TRADING SYSTEMS

THAT WORK
Building and Evaluating
Effective Trading Systems

THOMAS STRIDSMAN

McGraw-Hill
New York San Francisco Washington, D.C. Auckland Bogota
Caracas Lisbon London Madrid Mexico City Milan
Montreal New Delhi San Juan Singapore
Sydney Tokyo Toronto

Library of Congress Cataloging-in-Publication Data

Stridsman, Thomas.
Trading-systems that work : building and evaluating effective trading systems / by
Thomas Stridsman.
p. cm.
1SBN0-O7-13598O-X
1. Investment analysis.
HG4529.S77

2. Portfolio management.

I. Title.

2000

332.6<lc21
00-055008

McGraw-Hill
A Division

ofTheMcGravHiUCompanies

Copyright 2001 by The McGraw-Hill Companies, Inc. All rights reserved. Printed in the United States of
America. Except as permitted under the United States Copyright Act of 1976, no part of this publication may be
reproduced or distributed in any form or by any means, or stored in a data base or retrieval system, without the
prior written permission of the publisher.
2 3 4 5 6 7 8 9 0 AGM/AGM

0987654321

0-07-135980-X
This book was set in Times New Roman by Patricia Wallenburg.
Printed and bound by

Quebecor

World/Martinsburg.

This publication is designed to provide accurate and authoritative information in regard to the subject matter covered. It is sold with the understanding that neither the author nor the publisher is engaged in rendering legal,
accounting, futures/securities trading, or other professional service. If legal advice or other expert assistance is
required, the services of a competent professional person should be sought.
From
by a

Declaration

Committee

of Principles jointly

of the American
and a

adopted

Bar Association

Committee

of Publishers

McGraw-Hill books arc available at special quantity discounts to use as premiums and sales promotions, or for
use in corporate training programs. For more information, please write to the Director of Special Sales,
Professional Publishing, McGraw-Hill, Two Perm Plaza, New York, NY 10121-2298. Or contact your local bookstore.
This book is printed on recycled, acid-free paper containing a minimum of 50% recycled, de-inked fiber.

This book is dedicated to my mother, Britt-Marie

Stridsman. Thank you for your support and always
being there for all of us. There are not enough words
in the world to describe how much I love you.
All my forthcoming proceeds from this book will
be donated to Children's Wish List in Chicago, and
similar voluntary organizations, dedicated to help
kids with medical and other urgent needs, to help
them grow up as healthy and happy individuals.

DISCLAIMER
Hypothetical performance results have many inherent limitations, s o m e of which
are described below. No representation is b e i n g m a d e that any account will or
is likely to achieve profits or losses similar to those shown. In fact, there are
frequently sharp differences between hypothetical performance results and the
actual results achieved by any particular trading p r o g r a m .
O n e of the limitations of hypothetical performance results is that they are
generally prepared with the benefit of hindsight. In addition, hypothetical
trading does not involve financial risk, and no hypothetical trading record can
completely account for the impact of financial risk in actual trading. For
example, the ability to withstand losses or to adhere to a particular trading
p r o g r a m in spite of trading losses are material points which can also adversely
affect actual trading results. T h e r e are n u m e r o u s other factors related to the
markets in general or to the implementation of any specific trading p r o g r a m
which cannot be fully accounted for in the preparation of hypothetical
performance results and all of which can adversely affect actual trading results.
All trading systems and strategies in this book are intended for educational
purposes only, to provide a perspective of different market concepts. T h e y are
not meant to r e c o m m e n d or promote any trading system or approach. You are
advised to do your o w n research and testing to determine the validity of a
trading idea.

CONTENTS

INTRODUCTION xv
ACKNOWLEDGMENTS

xxiii

PART ONE
EVALUATING P E R F O R M A N C E

Chapter 1

Performance Measures

Total N e t Profit 4
M a x i m u m Intraday D r a w d o w n 5
A c c o u n t Size Required and Return on A c c o u n t 6
Average Trade 7
Largest W i n n i n g a n d L o s i n g Trades
7
Gross Profit and Gross Loss 8
The Profit Factor
8
Average W i n n i n g a n d Losing Trades
9
N u m b e r of (Winning/Losing) Trades a n d Average
N u m b e r of Bars per Trade
9
M a x Consecutive Winners a n d Losers a n d
Percent of Profitable Trades
10
C o d e 11

ix

Contents

Chapter 2
Better Measures

13

Slippage and C o m m i s s i o n
17
Profit per Trade
17
Largest Winning/Losing Trade
20
Cumulative Profit and Equity Tops
Drawdown
23
Flat Time and Run Ups
25
Code
26

21

Chapter 3
Futures Contract Data

31

Nonadjusted Data
31
Point-based Back-adjusted Data
Ratio-adjusted Data ( R A D )
37
Multimarket Portfolios
38
Perpetual Contracts 41

32

A Few Final Thoughts About Part 1

PART TWO
SYSTEM CONCEPTS

47

Chapter 4
Picking Tops and Bottoms
Chapter 5
Data Mining

55

Better Use of Your Data

62
Basic Exit Techniques
69
Chapter 6
Trade or Not to Trade

83

51

45

xi

Contents

Chapter 7
Following the Trend

85

M o v i n g Averages
86
D y n a m i c Breakout Systems
Standard Deviation B r e a k o u t

97
107

A Few Final Thoughts A b o u t Part 2

115

PART THREE
GETTING OUT

117

Chapter 8
Efficient Trades
Drawdowns

119

121

Chapter 9
Sweeney's MAE/MFE
Beyond M A E / M F E

143

156

Chapter 10
Adding Long-term Exits

167

T h e D y n a m i c Breakout System
168
T h e Standard Deviation Breakout System
Chapter 11
Working with Random Entries

187

T h e Gold D i g g e r System
191
T h e M e a n d e r System (Weekly Data)
T h e Black Jack System
204
A Few Final T h o u g h t s A b o u t Part 3

197

223

177

Contents

xii

PART FOUR
HIGH-PROBABILITY FILTERS

225

Chapter 12
Filtering

227

Trend Filters for Short-term Systems

T h e Gold Digger System
236
T h e M e a n d e r System 239
T h e Black Jack System 240

228

Chapter 13
Long-term Volatility Filters
T h e D B S System 2 4 6
T h e S D B System 2 4 7
T h e Directional Slope System

241

252

Chapter 14
What Makes a Trend

255

A F e w Final T h o u g h t s A b o u t Part 4

267

PART FIVE
MONEY MANAGEMENT AND
P O R T F O L I O C O M P O S I T I O N 269
Chapter 15
Money Management

271

Practical Applications for M o n e y M a n a g e m e n t

Short-term Systems 2 8 5
T h e Long-term Strategies 305
The SDB Strategy
305
The Directional Slope Strategy
313
The DBS Strategy
319

282

xiii

Contents

Chapter 16
Portfolio Composition

321

Total Equity Contribution I

321
Correlations and Covariances
324
Total Equity Contribution II
325
Chapter 17
Increasing Your Confidence

331

Out-of-sample Testing
331
Altering the Inputs
334
A Few Final T h o u g h t s About Part 5
Back to the Future
342

INDEX

345

339

INTRODUCTION

After I had been an editor, writer, and technical analysis expert for Futures magazine for two years it b e c a m e apparent to me that the main mistake most people
m a k e in the trading industry is to believe that there is nothing to it; that it is a piece
of cake to put together a trading strategy and then go out there and m a k e the big
bucks, k n o w i n g nothing about risk m a n a g e m e n t , m o n e y m a n a g e m e n t , why certain
systems simply do not work when traded on certain markets, why your worst drawdown is still to c o m e or w h y a system can work as it should but still break you. And
horrifyingly enough, this attitude s e e m s almost equally as c o m m o n a m o n g longtime professionals as a m o n g wanna-be amateurs.
A n o t h e r c o m m o n mistake, m a d e especially by smaller investors, is the thought
that lesser money needs less efficient and less thoroughly researched trading methods. That there will be plenty of time to learn about m o r e "sophisticated" strategies
and market intricacies as their accounts grow. This is not so. M o n e y is money; you
are equally as likely to lose it, whether you are a long-time, multimillion dollar
m o n e y manager or are new to the g a m e with a $ 10,000 savings account, if the methods you are using won't allow you to succeed in the first place. Think about it; why
would you be any m o r e successful trading $10,000 in a basic moving average
crossover system, with no m o n e y m a n a g e m e n t attached (just because this is all you
know about trading and all the m o n e y that you have), than if you were sitting on millions of dollars, knowing everything there is to know about trading?
If you think that you do not have enough money or knowledge to do it the
right way from the very beginning, you should not trade. Period. Because if you do,
you will find that there will in fact be plenty of time to learn about the more
"sophisticated" m e t h o d s , but not while trading. Ask m e I had the opportunity to
experience this first hand.
Trading Systems That Work is for all of you traders out there, no matter what
your level of experience is, w h o have c o m e to realize that it is not as easy at it seems,
XV

xvi

Introduction

but still haven't been able to pinpoint the missing pieces that will solve the whole
puzzle. I would guess that what you believe is missing, or what is keeping you from
succeeding as a trader, is the overall understanding of h o w it all ties together and
what really constitutes a m o r e sophisticated strategy with a higher likelihood for
success. This book is my most sincere effort to try to provide you with this overall
understanding, to help you solve the whole puzzle. Call me when you d o .
Today, I am primarily an analyst and writernot a trader, although I have
traded in the past and currently have a few systems out there, traded by others.
M a n y misinformed traders think that just because you are not a trader, you do not
know anything about h o w the markets work. In my opinion, nothing could be further from the truth. Just as good math skill is no guarantee of good language skill,
or good driving skill is no guarantee of good mechanical skill, good trading skill is
no guarantee of good analytical skilland vice versa for that matter. Furthermore,
being an analyst and writer also is a matter of choice: I simply cannot resist sitting
down in front of the computer and tinkering with something, then later trying to
write about it in one form or another.
As an analyst and specializing in systems a n d mechanical trading, I believe
it follows that I have the necessary trading skillsas long as I follow my o w n
strategies. If you do not believe m e , please continue reading and let me prove you
wrong. As you turn the pages, I will provide you with a set of useful tools based
on my personal and, I believe, highly innovative analysis techniques; all on a level
of sophistication a n d market knowledge widely surpassing most of what you can
find "out t h e r e " today.
Although I sometimes use scientific and academic jargonwords and phrases
such as standard deviation, kurtosis, and mathematical expectancyand s o m e times even indulge in the mathematics and calculus behind it all, I do not purport
to be a scientist, statistician, or psychiatrist, or to hold any other degree from which
I might have stolen s o m e of the techniques used throughout this book. In fact, I am
convinced that several of you will identify several situations where I expose my
ignorance in more topics than one. But that does not matter. The most important
thing that I will show you is that you do not need to be a rocket scientist to analyze
the markets successfully, but you must know a bit m o r e than what you have been
able to pick up from other books, and you must dare to think "outside of the b o x "
a little. A n d for that I m a k e no excuse.
W h e n trying to explain something new, a teacher often finds himself in a
dilemma. To explain subject C, he first must explain subjects A and B, but often
these subjects interweave with each other. So to learn A we need to first understand B and C. For the p u r p o s e of this book I have tried hard to keep it all in some
logical order and as simple as possible, but invariably there are instances where I
must talk about and explain two new subjects at the s a m e time. In other instances,
I have simply left a certain topic unexplained and gotten back to it later, in a m o r e
thorough analysis and explanation. On a few occasions, I also cheated a bit by

Introduction

xvii

oversimplifying things. I apologize for this, w i t h the explanation that I simply cannot give away the w h o l e store in one book. I h o p e you will bear with me and find
it worthwhile reading j u s t the s a m e .
Although I h o p e that you will perceive this as a practical book, it is not a
b o o k about a n e w set of ready-made technical analysis indicators, or a book about
foolproof trading systems, because there is no such thing. N o r is it a book that will
help you identify unique trading opportunities and show y o u how to exploit those
opportunities with a swift decision and then go do something else. And although
it might seem like it, it isn't even a book about n e w techniques or n e w ways to
develop trading strategies.
Instead, this is a book about how to go about reasoning even before you start
a development process that eventually and hopefully will allow you to put together
a trading strategy that is exactly thata strategy, a long-term work process, as
opposed to a series of single, isolated decisions. With the risk of sounding way too
pretentious right off the bat, this book is about philosophy: that is, the philosophy
behind the work of putting together a good, working trading strategy; the philosophy
behind what drives markets and what makes them behave as they do; and finally, the
philosophy that allows us to understand why a strategy works, or why it doesn't.
T h e key point I h o p e to get across is that a trading strategy is a "process
machine," where each decision m a d e automatically and immediately leads to the
next one, and next o n e , and next o n e . . . producing a long string of interacting decisions that form a never-ending p r o c e s s a l m o s t aperpetuum mobile. And, just like
a perpetuum mobile, the trading strategy is a very sensitive m a c h i n e , consisting of
a few parts that, by necessity, form a w h o l e greater than those parts and can allow
for no loss of energy. Ideally, for this to be possible, each part must be put together or constructed with all the other parts in m i n d but also be a part of all other
parts. (This is the philosophical aspect of this book.) It is almost.. . N o , in fact, it
is exactly as if each part is both one of the parts that m a k e a car work, and the car
itself, m a d e up of all the parts.
However, just as a perpetuum mobile is an impossibility, putting together a
good trading strategy is very m u c h like putting together, or buying, a good and
well-balanced car. While doing all this work, you also must think about your own
needs and current status in life, financially and otherwise. For instance, even if you
can afford it, y o u do not go and b u y yourself a N A S C A R racer for your Sunday
afternoon sightseeing drives with the family. N o r would y o u go and b u y a road
grader for the s a m e purpose. Instead you probably try to find a car that fulfills its
p u r p o s e for your everyday life, although you might think a station w a g o n or a mini
van a little boring when you know you really are the Lamborghini type. The s a m e
goes with trading strategies. First, you m u s t c o m e to grips with w h o you are and
what type of strategy is suitable for you, even if it is a little boring a n d leaves little r o o m for fast "overtaking m a n e u v e r s " or any other cool improvisations "behind
the wheel." In fact, my experience is that trading, at least systematic trading, is as

xviii

Introduction

boring as watching baseball without a beer in your hand a n d a few in your belly.
But the s a m e holds true for both the public roads and the trading rooms. They are
no places for fun a n d g a m e s .
W h e n putting together a trading strategy, this is h o w I look at it: the actual
system with its buy and sell rules is the engine that could be anything from a fast
intraday, market specific system (comparable to an 800 h.p., fast, N A S C A R - t y p e
engine) to a slow, long-term, universal system (comparable to a slow but steady
road grader engine). W h e n this is done, I take a look at the m o n e y m a n a g e m e n t ,
which is equal to the gearbox and the transmission. With the purpose of the system in mind, I now try to find a way to get its force down to the ground as efficiently and safely as possible. Continuing with the car analogy, you know it is a
recipe for disaster if y o u try to fit the engine of a N A S C A R racer with the transmission of a station wagon.
O n c e the engine (the system) and the transmission (the m o n e y m a n a g e m e n t )
seem to be working well in a balanced h a r m o n y with each other it's time to think
about the coach and the chassis. In a trading strategy, this is comparable to the
question of which markets to trade. In the case of a market-specific system, this is
already done (and a case of when subject C c o m e s before subject A ) . But whether
a strategy is market specific or not, it still is a good idea to make sure that it works
in as m a n y markets as possible. In the case of a multimarket strategy, it is equally
as important not to start curve-fitting the strategy by optimizing the markets in
regard to the system, as it is not to curve-fit the system to the markets.
There is a vast difference between a good system and a profitable system. It is
paramount to understand this difference. It also is very important to understand that
a good system can always be turned into a profitable system, while the opposite does
not necessarily hold true. This is because a well working system will usually work
well on a multitude of markets, catching the same type of moves, measured in percentage terms or s o m e other universal measurer. A profitable system, on the other
hand, is a well working system that generates a surplus of dollars when applied in
the context of a complete strategy to a specific market or portfolio of markets.
W h e n you have put together the car (the overall strategy) with its engine (the
system), the transmission (the money m a n a g e m e n t ) , a n d the chassis (the portfolio
of markets) there still are a few things missing: the fuel and the driver. T h e fuel is
your time and money. The driver is you. But before you fill the baby up and j u m p
into the driving seat, m a k e sure that this really is the vehicle for you (even though
y o u know in your heart that you really are the Lamborghini type).
Although Trading Systems That Work is not one of those 13-in-a-dozen books
on day trading that s e e m s to s w a m p the market these days, everything said here
could be translated to day trading techniques as well. Neither is it a book specifically aimed toward any specific type of market, although I have m a d e extensive
use of the c o m m o d i t y futures markets in my examples. Instead, words such as
stocks, markets, and contracts should be looked at as s y n o n y m s .

Introduction

xix

Part 1 takes a close look at how to measure the performance of a system

using a set of basic a n d universal m e a s u r e m e n t s and h o w to expand the analysis
further by incorporating a spreadsheet program, such as MS Excel or Lotus 1-2-3.
This part also takes a closer look at different types of data series a n d discusses
when and how to use them. This is especially important to understand if you are a
futures trader, but even if y o u have decided to stick to the stock market, this section should provide you with valuable insight on why so many of the systems y o u
have built so far have ceased to work as s o o n as you have taken them live.
In Part 2 , 1 put together a set of basic long-term and short-term trading syst e m s , using different types of data d e p e n d i n g on what I am trying to achieve. S o m e
of the systems are market specific, others are suitable for a basket of markets. A
lot of the analysis is done in a spreadsheet program, a n d for this we need the code
developed in Part 1. Other than the specific entry techniques that we will continue to use throughout the b o o k , the most important things to learn from Part 2 are
that a good working system does not necessarily have to be a profitable one and
that s o m e systems are d o o m e d to fail when traded on specific markets.
In Part 3, we take the systems that we put together in Part 2 a n d examine
them further for their statistical characteristics and h o w they can be improved
upon, using J o h n Sweeney's maximum adverse excursion ( M A E ) and maximum
favorable excursion ( M F E ) analysis techniques a n d different ways of splicing up
the drawdown. We also take a closer look at a few additional measurements, such
as kurtosis and skew. In m a n y ways, Part 3 is the most important part, thanks to
the exits that we later attach to a fixed fractional m o n e y m a n a g e m e n t regimen that
boosts the results ten-foldat least.
With the entries from Part 2 and the exits from Part 3, Part 4 takes a closer
look at different ways of filtering out favorable market situations and set-ups. A
lot of the work is done using r a n d o m entries to generate as many trades as possible. Using r a n d o m entries, we can take only a few y e a r s ' worth of data to produce
as m a n y years of u n i q u e trading sequences as we d e e m necessary. (In fact, after
writing this book, using the techniques described in this section, I tested a system
on the last ten years of data for all the D o w - 3 0 stocks, producing a total of m o r e
than 3,000,000 years of u n i q u e trading sequences. If that doesn't produce robust
results, nothing will.) At the end of this part a m o r e theoretically oriented chapter
provides the framework for understanding for what m a k e s a trend and why it is my
conviction that systematic trading is likely to work in the first place.
In Part 5, we tie it all together by combining all the systems with different
fixed fractional m o n e y m a n a g e m e n t strategies. We also take a closer look at how
to put together a portfolio consisting of several market/system combinations,
which help each other produce results that would have b e e n impossible to achieve
without all systems working in t a n d e m under a mutually shared m o n e y m a n a g e ment regimen. To coin a new word, what we will do is to "optisize" rather than
optimize the system. T h e main difference between "optisizing" and optimizing is

XX

Introduction

that w i t h optimizing we are fitting the system to the data; with "optisizing" we are
fitting the bet size to the system. T h e less optimized your system is, the m o r e
"optisized" you can m a k e the strategy. " O p t i s i z i n g " will almost always do m o r e
for your bottom line than optimizing. M o s t of the work is with the help of a
spreadsheet program, a n d there will be plenty of e x a m p l e formulas a n d code that
you can copy and use in your own work. Before we round off the entire b o o k , we
also take a brief look at how to test a system for robustness to further increase our
confidence in it before we start using it in real life trading.
A few people have asked me w h y I am giving away my ideas instead of m a k ing the most of them myself. Well, the truth is that I think I am m a k i n g the most
of t h e m , because even if this b o o k all of a sudden tops The New York Times bestseller list, these systems and strategies won't be widespread enough to arbitrage
away what I will show you over the next few pages. W h y ? B e c a u s e , first of all, not
everyone w h o reads this will use it, because they w o n ' t feel comfortable w i t h it or
b e c a u s e they simply don't like working w i t h other people's ideas, but like to c o m e
up with ideas of their own. Second, even if they do use t h e m , m a n y will still not
be able to m a k e any m o n e y out of them, because invariably they will c o m e up with
unthought-of ways of screwing things up anyway. A n d last but not least, because
the markets are m u c h bigger than any of us can fathom, these strategies will still
only m a k e up a m e r e fraction of all strategies out therestrategies that help m a k e
these work.
In fact, I h o p e that these strategies will be widespread e n o u g h to reinforce
and feed on themselves, m a k i n g them even m o r e profitable. If y o u look at it from
that perspective, y o u are not my e n e m y in this, but my best friend and accomplice.
My only true enemy in this g a m e is myself. As in almost any other situation in life
there are a thousand-and-one ways I will be able to screw things up without your
help. So, the bottom line is that even if I give this away to you, the danger that
p o s e s to my future wealth and well-being amounts to nothing, compared to the
danger I p o s e to myself.
L a s t but not least, please read the following quotation by the renowned
m o n e y m a n a g e r Ralph Vince, then p u t Trading Systems That Work aside for awhile
and contemplate it. Because if y o u do not understand, or do not a g r e e with it, there
is no need for you to continue:
" T h e key to ensure that you have a positive mathematical expectation in the future is to not restrict your system's degrees of freed o m . . . This is accomplished not only by eliminating, or at least
minimizing, the n u m b e r of optimizable parameters, but also by
eliminating, or at least minimizing, as many of the system rules as
possible. Every parameter you add, every rule you add, every little
adjustment and qualification you add to your system diminishes its
degrees of freedom. Ideally, you will have a system that is very

Introduction

xxi

primitive and simple, and that continually grinds out marginal

profits over time in almost all the different markets. Again, it is
important that you realize that it really doesn't matter how profitable the system is [by itself], so long as it is profitable. The
m o n e y you will m a k e trading will be m a d e by h o w effective the
m o n e y m a n a g e m e n t you employ is. The trading system is simply a
vehicle to give you a positive mathematical expectation on which
to use money management. Systems that work [show at least a
marginal profit] on only one or a few markets, or have different
rules or parameters for different markets, probably w o n ' t work
real-time for very l o n g . . . "
W h i l e you are reading through the rest of the book, please feel free to visit
me at my Web site, at w w w . T h o m a s S t r i d s m a n . c o m , for updates on how to order
additional material, to help me help the kids at the Children's M e m o r i a l Medical
Center, or to j u s t leave me a c o m m e n t or two.
Thomas

Stridsman

ACKNOWLEDGMENTS

Never in my whole life did I think that I would get a chance to thank everybody
w h o has m e a n t so m u c h to m e , both professionally and personally. Inevitably,
thanking everybody would result in a thank-you note longer than the rest of the
book, so I just have to settle for a few very special and recent people. T h a n k s to
N e l s o n Freeburg and his family, and to Dan and M a r y a n n e Gramza, for believing
in me and r e c o m m e n d i n g me for the j o b as a technical analysis editor for Futures
magazine.
Thanks to Ginger Szala and Jamey Holter and all the other editors at Futures
magazine for hiring me and believing in me as an editor at Futures, despite my limited
English. A special thank you to my friends Jim and Margie Kharouf for helping me get
settled in Chicago.
Thanks also to Jonas Vikstrom and Johan Ljung at Vikstrom, Ljung & Partners,
for providing me with a state-of-the-art computer, and to M a x von Liechtenstein for
contributing with the very important " W h a t makes a trend" chapter in Part 4 and for
helping me refresh my micro- and macroeconomics proficiencies.
A very w a r m "thank y o u " also to Melissa L a n g e for teaching me everything
there is to k n o w about w h a t an open and honest relationship is all about. To all the
rest of y o u that I have learned to k n o w and learned from throughout the years, know
that I am forever grateful.
A last minute thanks to Patty Wallenburg for being a great sport and of
tremendous help during the final editorial stages of putting this book together.

xxiii

PART

ONE

Evaluating Performance

W i t h the stock market in an unprecedented up-trend and with ever decreasing

prices for state-of-the-art computers, m o r e people than ever before are playing the
markets, trying to m a k e a living (and hopefully a fortune). M o r e and m o r e people
also try to do this in a systematic fashion, using one or several mechanical trading
strategies. In the backwater of this, several p r o g r a m vendors now offer p r o g r a m s
with w h i c h you can build and test your own trading strategies, or even b u y and plug
in strategies m a d e by others. A couple of the most popular p r o g r a m s for this are
O m e g a Research's TradeStation and E q u i s ' MetaStock. Other p r o g r a m s exist, but
thanks to their built-in p r o g r a m m i n g capabilities, these two programs are the m o s t
popular, a i m e d toward the retail customer.
T h a n k s to its PowerEditor and EasyLanguage, TradeStation probably provides
the m o s t professional way available today for coding and evaluating your own trading strategy. M o s t professional market analysts (myself included) probably have a
strange love-hate relationship to TradeStation. On the one hand, it provides you with
the most possibilities of any program, but on the other hand, its m a n y possibilities
also reveal a few embarrassing weaknesses, especially when it c o m e s to the evaluation p r o c e s s w h e r e it shares the s a m e weaknesses found in m a n y other programs.
This is very important, because before you can start to build a n d investigate
any type of trading strategy you must know what type of information to look for,
a n d if the information is not there, c o m e up with a way to p r o d u c e it yourself. In
this first part, you will learn which m e a s u r e m e n t s are m o s t important. S o m e of
t h e m you will be able to derive directly from TradeStation or MetaStock's performance s u m m a r i e s . Others m u s t be exported into a text file with the help of each
program's built-in p r o g r a m m i n g capabilities, for further analysis in Excel or any
i

PART 1

Evaluating Performance

other spreadsheet program. For c o m m o d i t y futures traders, it also is very important to use the right type of data a n d to r e m e m b e r that n o t all time series should be
treated the same.

CHAPTER

Performance Measures

W h i c h system-testing m e a s u r e s are likely to work and which are not? By necessity all system testing and design m u s t be m a d e on historical data. T h e trick, then, is
to m a k e as good use as possible out of this data, and to m a k e your evaluation m e a s ures as forward-looking as possible. This chapter presents a r u n d o w n of the m o s t
c o m m o n l y used measures in TradeStation's performance s u m m a r y a n d explains
which have some value, which don't, and which can be modified with the help of a
spreadsheet program. B u t before we go on, let us start with a little quiz.
If you can choose between b u y i n g two different stocks, one currently priced
at $12.50 and the other at $20, and you know for sure that the one priced at $12.50
will rise by 1.75 points over the next couple of days, while the one priced at $20
will increase 2.60 points (almost a full point more) over the same period of time,
which one would you choose? If you answer the $12.50 stock, you probably understand what I am hinting at and should have little trouble understanding this part of
the book.
If however, you answered the $20 stock, you probably are a little too anxiously chasing that elusive dollar. Just do the math. In this case, 12.50 divided by 20
equals 0.625, or 5/8. This m e a n s that for every 500 $20 stocks you buy you can use
the s a m e a m o u n t of m o n e y to buy, 800 $12.50 stocks. 500 times 2.60 points equals
1,300 points (or 13%) in profits if you b u y $10,000 worth of the $20 stock. 800
times 1.75 equals 1,400 points (or 14%) in profits if you b u y $10,000 worth of the
$12.50 stock.
If you think this difference isn't m u c h to worry about, what if you could
choose a m o n g 20 trades like this for the rest of the year, being able to use the profits from each trade in the next one? T h e n your initial $10,000 would grow to
3

PART 1

Evaluating Performance

$115,231 if you only b o u g h t the $20 stock, but to $137,435 if you only bought the
$12.50 stock. A n d w h a t if you could do this for three years straight? T h e n your initial $10,000 would grow to $15,300,534 if you only b o u g h t the $20 stock, but to
$25,959,187 if you only bought the $12.50 stock. This is a difference of m o r e than
$10,000,000 after only 60 trades.
A l t h o u g h these n u m b e r s are idealized, they illustrate t h e point that it pays
to take it easy and do the m a t h before y o u j u m p into a t r a d e . A n d it is exactly
this t y p e of m a t h that no m a r k e t analysis a n d trading software p a c k a g e s allow
y o u to d o .

TOTAL NET PROFIT

Probably the m o s t frequently used optimization m e a s u r e is total net profit. Often,
it is used together w i t h maximum intraday drawdown. Unfortunately, however, the
total net profit is of very little value when it c o m e s to evaluating a trading strategy's estimated future performance, no matter h o w rigorous the testing or how
robust the system. T h e reason for this is twofold and depends on whether you p r e fer to work with only one market at a time or with a basket of markets and/or systems to m a k e up a portfolio. W h a t applies to the one-market case also holds true
for a portfolio.
In the one-market case, the total net profit tells you nothing about w h e n your
profits occurred and h o w large they were in relation to each other. This is especially important if the market you are interested in is p r o n e to trending. For
instance, if the market has been in a prolonged up-trend it is likely that the dollar
value of each trade has increased with the increasing dollar value of the market.
This, in turn, m e a n s that the total net profit is unevenly distributed through t i m e
and mostly influenced by the very latest market action. In a down-trend the o p p o site holds true. Notice however, that the trend of the market says nothing about
whether the system has b e c o m e m o r e robust or not. In a market with several distinctive u p - a n d - d o w n trends, this matter is even m o r e complex.
In the multimarket case, the total net profit tells you nothing about h o w well
diversified your portfolio is. This is especially true if you stick to trading an equal
a m o u n t of shares for all equities, or o n e contract per c o m m o d i t y in the c o m m o d ity futures market. This is because w h a t are considered h u g e dollar moves in some
stocks or markets are only considered to be ripples on the surface in others. You
cannot, for instance, diversify a one-contract trade in the S & P 500 futures markets
with a one-contract trade in corn, no matter h o w well your system s e e m s to work
in each market. Simply stated, the larger the market value of the market, the larger the impact on the total net profit of the portfolio.
1

Your maximum equity loss, or loss of capital, calculated over both open and closed out profits. The term intraday indicates that this drop in equity can start at any time, at any day, and end at any time, at any day.

CHAPTER 1

Performance Measures

In the stock market, to value a system by its total net profit can take on
absurd consequences w h e n a c o m p a n y decides to do a stock split. For instance, say
that a stock currently is trading at $90, and a trading system that consistently buys
and sells 100 stocks per trade shows a historical, hypothetically back-tested profit of $150,000. Tomorrow, after the stock has been split 3:1 and the stock is trading at $30, the historical, hypothetically back-tested profit has decreased to
$50,000. D o e s this m e a n that the system all of a sudden is three t i m e s as bad as
the day before? Of course not; it's exactly the s a m e system, and from this e x a m ple it is easy to see that after the split you m u s t trade the stock in lots of 300 stocks
p e r trade to m a k e the new results comparable with old pre-split results. M a n y
times, however, w h e n we need to compare different markets, over different periods of time, and traded with different systems, it is not always this easy. T h e following chapters teach you how to work around this dilemma and build a welldiversified portfolio that is likely to hold up in the future and continue to provide
g o o d risk protection.

MAXIMUM INTRADAY DRAWDOWN

H o w many of us have not heard the old adage "your worst drawdown is still to
c o m e ? " While this is likely to c o m e true sooner or later, it does not have to happen
first t h i n g tomorrow, provided you have d o n e your h o m e w o r k correctly.
Unfortunately, the information most system testing software provides you will not
be enough, because the number given is in dollars, without any relation to where and
when this bad trade sequence struck. For instance, with a point value of $250 for a
S & P 500 futures contract, there is a vast difference between a $20,000 drawdown
when the market was trading around 500 and a $20,000 drawdown with the market
trading at 1,350. In the latter case, a $20,000 drawdown is similar to being caught in
a series of bad trades for about 5% of current market value, something that can happen to anybody. In the first case, however, an equally large drawdown in dollars
would amount to approximately 16% of the market value at that point in time. With
the market trading at 1,350 a 16% drawdown translates to $54,000. With only the
information from your regular performance summary, however, there is no way for
you to find that out. No wonder then, that so many traders blow out before they even
have a chance to get started. To calculate the true expected drawdown, you must first
find the largest drawdownin percentage termsin relation to the market value
when it occurred and then translate that percentage into today's market value.
A l s o , when evaluating your drawdown, you m u s t know exactly what you are
looking at. In TradeStation's case, the d r a w d o w n is calculated as the closed trade
drawdown ( C T D ) , plus the open trade drawdown ( O T D ) , m a k i n g up the total equity drawdown ( T E D ) at that point in time. B u t this is not necessarily the best way
to look at the d r a w d o w n w h e n it c o m e s to building a robust system. (We will
explore the reasons for this throughout this book). In Part 3 we also take a look at

PART 1

Evaluating Performance

how to divide the O T D into start trade drawdown ( S T D ) a n d end trade d r a w d o w n

( E T D ) a n d h o w to analyze t h e m for better entries and exits. N o t until this is d o n e
should you address the C T D for a better overall performance.

ACCOUNT SIZE REQUIRED AND RETURN ON ACCOUNT

T h e account size required a n d return on account are probably the m o s t deceiving
n u m b e r s of t h e m all. As you can see by looking at Figure 1.1, which shows the performance s u m m a r y for an early version of Black J a c k / M e a n d e r system in
TradeStation, the account size required is the s a m e as the value for the m a x intraday drawdown. But this is a purely theoretical figure produced with the help of
hindsight; there is no way for you to k n o w that n u m b e r before you start trading,
and there certainly is no guarantee for this value to hold up in the future.
To calculate the return on account, TradeStation simply divides the total net
profit by the account size required. T h e m a i n thing w r o n g with this n u m b e r is that

FIGURE

1.1

The performance summary for an early version of Black Jack/Meander system In TradeStation.

CHAPTER 1

Performance Measures

it is supposed to be calculated at one point in time (before you start trading) with
the hindsight from two totally different points in time (during trading for the worst
d r a w d o w n and when finished trading for total net profit). Perhaps even m o r e
important from a real-world point of view, no trader in his right m i n d would start
out with a trading account that is only expected to cover his worst historical drawdown, especially because this drawdown figure has no connection to the future
whatsoever and is very likely to be exceeded. T h u s , because there is no way to
k n o w your exact largest future drawdown or ending equity, you have no idea about
how large your trade account m u s t be and consequently no way of estimating the
return on your account either. Therefore, these two figures are completely u n n e c essary and provide you with no information whatsoever.

AVERAGE TRADE
O n e of the m o s t important things to consider before starting to trade any system
is the estimated average profit per trade in the future. Unfortunately, neither
TradeStation's nor MetaStock's performance summaries give you any such forward-looking information. Basically, what can be said about the total net profit
also can be said about the average profit per trade, at least as long as it is based
solely on historical dollar measures. W h e n it c o m e s to the average profit per trade,
however, the reasoning is that the trades that happened "way back when," when the
market was trading at a completely different level, will impact the value of the
average trade too m u c h . For instance, if the market has traded from 1,000 points
to 2,500 points, and y o u ' r e trading a system based on three profitable trades, o n e
for $ 1 0 0 when the markets was trading at 1,000; another for $200 w h e n the markets w a s trading at 2,000; and most recently, o n e for $150 when the market w a s
trading at 1,500, TradeStation will tell you that the average trade is $150. But this
is not what you can expect the average trade to be now, when the market has rallied and is trading considerably higher. F r o m this example it is not too difficult to
figure out that with the market trading at the 2,500 level, the next trade is expected to show a profit of $250. A m o r e disturbing example, however, can be seen in
Figure 1.1. H e r e the average profit per trade is $ 5 7 7 , while the true expected average profit per trade in today's market is $1,269 (as of October 1999).

LARGEST WINNING AND LOSING TRADES

Although the estimated largest drawdown holds information about how large your
account size must be and can give you an indication of whether you have the psychological profile to trade the system in question, the information about the largest
losing trade is m u c h more important than the drawdown for m o n e y m a n a g e m e n t
purposes. As is the case with so m a n y other of the most popular measures, however, the values for the largest winning trade and the largest losing trade essentially

PART 1

Evaluating Performance

hold no value as long as you can't put t h e m in relation to w h e n they happened and
where the market was trading at the time. Once this is known, the largest losing
trade can be used to set up a fixed fractional m o n e y m a n a g e m e n t strategy that m o s t
certainly will prove m u c h m o r e important for your b o t t o m line than the system
itself. (Fixed fractional m o n e y m a n a g e m e n t is discussed in Part 5.)

GROSS PROFIT AND GROSS LOSS

If the total net profit is of little good as a performance evaluator, it should follow that
so are the gross profit and the gross loss, right? Unfortunately, it is not that simple
and the answer is " y e s a n d n o . " L o o k e d at separately, it is likely that the gross
profit and the gross loss will be influenced in the very same way as the total net profit. That is, in a market prone to trending, the values of the winning and losing trades
are likely to vary with the value of the market. A n d in a portfolio, the larger the market value of the market, the larger the impact on the gross profit and the gross loss
of the portfolio. However, provided that the profits and the losses are evenly distributed through time and the relationship between t h e m stays approximately the same
over time, they can hold a wealth of information that is very useful in your initial
performance evaluation. This information is derived via the profit factor.

The Profit Factor

To calculate the profit factor, simply divide the gross profit by the gross loss; the
answer tells y o u h o w m a n y dollars you are likely to win for every dollar you lose.
For instance, say that y o u have $2 a n d place $1 in a bet, hoping to w i n $2 m o r e ,
ending up with $4. T h e first time you try, you lose and your gross loss is $ 1. W i t h
your last $ 1, you take another chance. This t i m e you w i n $2, ending up with a total
of $ 3 . Your gross profit is therefore $2. Two divided by one equals two, w h i c h is
your profit factor.
Now, do the same betting sequence all over again, but this time multiply all
values by 10. This time you first lose $10 and then w i n $20, ending up with $30.
20 divided by 10 also equals two. Hence, the profit factor is simply the relationship
between dollars lost and dollars gained and, because it is a ratio, it is a way of normalizing your results to m a k e t h e m comparable between time frames and markets.
For the profit factor to work, it does not matter if the market has b e e n trending or not as long as it is reasonable to a s s u m e that your profits a n d losses have
fluctuated at an equal p a c e a n d are evenly distributed through time. For the same
reason, it also is possible to use the profit factor as a c o m p a r i s o n between different systems a n d m a r k e t s . It is obvious that the higher the profit factor the better
the system. It is even m o r e important to determine how robust the profit factor is,
than h o w high it is. That is, h o w likely is it that the profit factor will hold up in the
future a n d in different m a r k e t situations?

CHAPTER 1

Performance Measures

M a n y system vendors and trading experts believe that you should not trade a
system with a hypothetically back-tested profit factor below three, because they
k n o w from experience that the profit factors for all their systems will decrease
considerably w h e n the system is traded live on u n s e e n data. Probably the only reason for this r e c o m m e n d a t i o n is that they do not understand how to build a robust
trading strategy in the first place. If nowhere else, this b e c o m e s evident w h e n they
continue to use the total net profit and drawdowns as their m o s t important performance evaluators. To build a robust trading system that is likely to continue to
perform in the future, m a k e sure that the underlying logic is sound and simple, that
the trading rules are as simple and as few as possible, and that the gross profit and
the gross loss are evenly distributed through time and in relation to each other. If
you m a n a g e to do all this, y o u will be surprised at how m u c h you can expect to
p r o d u c e from a system with a profit factor as low as 1.5, or even lower.

AVERAGE WINNING AND LOSING TRADES

As is the case w i t h gross profit and g r o s s loss, the average w i n n i n g and losing
trades can h o l d s o m e v e r y valuable information if u n d e r s t o o d a n d treated correctly. A g a i n , however, t h e trick is to m e a s u r e t h e m at today's m a r k e t value. For
instance, if y o u currently are in a d r a w d o w n a n d k n o w the values of y o u r average w i n n e r a n d loser and the frequency with w h i c h they are likely to occur, you
can calculate the m i n i m u m e s t i m a t e d n u m b e r of trades (and time) it s h o u l d take
to r e a c h a n e w equity high. For instance, if y o u k n o w that your average profit
per trade equals $ 4 0 0 , a n d y o u currently are in a $2,500 d r a w d o w n , the estimated n u m b e r of trades to get you out of the d r a w d o w n is seven ( I N T ( 2 , 5 0 0 / 4 0 0 )
+ 1).

NUMBER OF (WINNING/LOSING) TRADES AND

AVERAGE NUMBER OF BARS PER TRADE
M a n y traders and analysts pay very little or no attention to the n u m b e r of trades a
system is likely to generate. But this is very important information that gives you
the first clue to whether the system is suitable for you. T h e questions you m u s t ask
yourself are " d o e s this system trade often e n o u g h " and " d o e s it keep me in the
market enough to satisfy my need for a c t i o n ? " T h e s e are seemingly silly questions
at first glance, but the truth is that a specific system will not suit everybody, no
matter how profitable it is. If it does not fit your personality or style of trading,
y o u will not feel comfortable trading it.
W h a t is even m o r e important, however, is h o w m u c h time the system is
expected to stay in the market. This is because time spent in the market equals risk
assumed. Therefore, the less time you can spend in the market to reach a certain
profit, the better off you are. To calculate the relative time spent in the market,

PART 1

10

Evaluating Performance

multiply the n u m b e r of w i n n i n g trades by the average n u m b e r of b a r s for the w i n ners. A d d this to the n u m b e r of losing trades, multiplied by the average n u m b e r of
bars for the losers. Finally, divide by the total n u m b e r of b a r s examined. U s i n g the
performance s u m m a r y above, this equals approximately 0.4, w h i c h m e a n s that you
will be in a trade only four days out of ten. Obviously, you also will be better off
the fewer and shorter the losing trades are.

MAX CONSECUTIVE WINNERS AND LOSERS AND PERCENT OF PROFITABLE TRADES

You should try to keep the m a x i m u m n u m b e r s of consecutive winners and losers as
low as possible a n d the percent of profitable trades as high as possible. T h e n u m ber of consecutive losers is especially important, if you like to feel comfortable with
trading the system. For a correctly built system* however, with a relatively high
n u m b e r of profitable trades, these n u m b e r s hold very little value and should be
looked upon m o r e as freak occurrences than anything else. W h e n you examine your
system, it is very important to know whether the system has a tendency to produce
strings of trades with similar outcomes. If it does, the system and the market are
still holding valuable information that you haven't yet exploited. If you do not m a n age to get rid of this tendency, you need to keep this information in m i n d when you
are designing the m o n e y m a n a g e m e n t strategy to go with the system.
It also is a good idea, w h e n monitoring your real-time performance, to keep
track of h o w m a n y consecutive w i n n e r s it might take to bring you out of a drawdown. Continuing with the $2,500 drawdown example, it will take you four w i n ning trades in a row to reach a n e w equity h i g h if your average winning trade is
w o r t h $700 (INT(2,500 / 700) + 1). Once y o u k n o w this, ask yourself h o w likely
it is for this to happen. If the answer is it m o s t likely h a p p e n s once in a blue m o o n ,
that the m o s t winning trades y o u can expect in a row are two, that it is highly likely that the losers will c o m e in pairs, and that your average loser is for $300. T h e n
you k n o w that in the best of worlds, it can be expected to take at least ten trades
before you are in the black againprovided that you experience two winners in a
row, three times in a row, never experience m o r e than two losers in a row, a n d start
out with two winning trades. If you instead start out with two losing trades, taking
the d r a w d o w n down to $3,100, everything else being equal, the estimated n u m b e r
of trades w o u l d be 16.
Together with the profit factor, the percent profitable trades is the only
performance m e a s u r e that can be derived immediately from the performance
s u m m a r y a n d has any value w h e n it c o m e s to estimating the system's future
performance. Again, however, this assumes that the underlying logic is sound and

A bar is a marker in a chart that shows the trader how the price of the market fluctuated over a specific time
period. A bar usually shows the opening, high, low, and closing price. It can be specified to cover any
time period, such as 1 minute, 1 hour, 1 day, 1 week, etc.

CHAPTER 1

Performance Measures

11

that the system can be considered robust. Obviously you should try to keep this
n u m b e r as high as possible to feel comfortable with the strategy, but, as is the case
with the profit factor, a m o r e robust n u m b e r is better than a high one w h e n it
c o m e s to the system's tradability and when picking the best m o n e y m a n a g e m e n t
strategy. In fact, sometimes it even is a good thing to keep this n u m b e r down in
favor of a higher n u m b e r of profitable m o n t h s for the overall strategy or portfolio.
Nonetheless, with this n u m b e r in hand, we now can put together a m o r e generic
formula to calculate the estimated n u m b e r of trades, N, to get out of a drawdown:
N = I N T ( D D A / (X * AW - (1 - X) * A L ) ) + 1, w h e r e
DDA = Drawdown amount
X = Likelihood of a winner, between zero and one
AW = Average w i n n e r
AL = Average loser
For example: If the average winning trade is worth $700, the average losing
trade $ 3 0 0 , and the likelihood of a w i n n e r is 4 5 % , then the estimated n u m b e r of
trades it will take to bring you out of a $2,500 drawdown is 17. Furthermore, if the
average trade length is five days and the total a m o u n t of time spent in the market
is 3 3 % , then it will take you an estimated 250 trading days (17
5 + 0.33) to get
out of the red again, not counting w e e k e n d s and holidays. That is almost a full
year's worth of trading to get out of a situation that, at a first glance, only s e e m s
to require "just a few good trades."
This equation shows that trading is a very deceptive business, and that we
h a d better m a k e sure that we know w h a t we are doing. N o t e also that in this e x a m ple we are working with dollar amounts and a fixed n u m b e r of contracts traded on
one market only. In u p c o m i n g sections, where we start working with percentages
a n d an ever-changing n u m b e r of contracts traded on a w i d e variety of both markets and systems, this matter b e c o m e s even m o r e complex.

CODE
T h i s chapter discussed which of TradeStation and MetaStock's performance s u m m a r y measures are likely to work and w h i c h are not. T h e only two measures that
you can put to immediate use are the profit factor a n d the percent profitable trades.
With the c o d e below, you can create a file that exports this information to a spreadsheet program, together with the time spent in the market. T h i s can c o m e in h a n d y
if you, for instance, want to compare the same system on a w i d e variety of markets, or several variations of the s a m e system on o n e market during the initial steps
of the building process. By interchanging the SysVer input with all your other
inputs, you can keep track of which version of the system each row refers to.

12

PART 1

Evaluating Performance

In the next chapter we show h o w to use TradeStation's E a s y L a n g u a g e to

export m o r e data into Excel for further analysis, a n d h o w to c o m e up with a new
set of forward-looking evaluation measures that are better suited to holding up in
the future and helping you to build m o r e robust and reliable trading strategies.
Inputs: SysVer(O);
Vars: PFactor(O), WTrades(O), TotBars(O), TradeStrl("");
If CurrentBar = 1 Then Begin
TradeStrl = "Market" + "," + "Version" + "," + "P factor" + "," + "%
Winners" + "," + "% in trade" + NewLine;
FileAppend("C:\Temp\Chap 1 -1 .csv", TradeStrl);
End;
If LastBarOnChart Then Begin
PFactor = GrossProfit / -GrossLoss;
WTrades = NuraWinTrades * 100 / TotalTrades;
TotBars = (TotalBarsLosTrades + TotalBarsWinTrades) * 100 / BarNumber;
TradeStrl = LeftStr(GetSymbolName, 2) + "," + NumToStr(SysVer, 0) +
"," +

NumToStr(PFactor, 2) + "," + NumToStr(WTrades, 2) + "," +

NumToStr(TotBars, 0) + NewLine;
FileAppend("C :\Temp\Chap 1-1 .csv", TradeStrl);
End;

CHAPTER

Better Measures

So far, most testing packages (and back-adjusted time series for the commodity
futures markets) have only allowed you to back-test your strategies using dollar values. This is all fine and well if you are only interested in how m u c h money you could
have m a d e had you been able to trade your strategy in the past. The main disadvantage with this way of testing, however, is that it tells you very little about how well
your strategy is likely to hold up in the future. To achieve this, a whole new set of performance measures must be put together. In this chapter, we take a closer look at how
to export the necessary data from your system trading software into a text file and
how to calculate this new set of performance measures in a spreadsheet program.
In essence, what you m u s t do is to calculate all the necessary values in percentage terms rather than in dollars or points. In this way, you will be able to perform m o r e accurate comparisons of how a system is likely to work in different markets and time frames. For instance, say that a market (the S & P 500) currently is
trading at 1,350 and that each point m o v e is worth $250. If this market rises 1%,
that move is worth $3,375 (1,350 X 0.01 X 250). But if the s a m e percentage move
happened " w a y b a c k when," w h e n the market was trading at 2 5 0 , a 1% move was
only worth $625 (250 X 0.01 X 2 5 0 ) . If the market value increases, it is likely that
the dollar moves increase as well, while the percentage moves are likely to remain
approximately the same. (For proof of this, see Figures 2.1 through 2.3.) T h u s , by
using percentage-based calculations, you give each and every trade an opportunity
to influence your strategy to an equal degree and you are able to build better and
m o r e reliable strategies, which are m o r e likely to hold up in the future. T h e same
reasoning can also be applied to a comparison of different markets with different
market values. O n c e you are done with your percentage-based calculations, you can
transform t h e m into dollar values based on w h e r e the market is trading today.
13

14

PART 1

FIGURE

2.1

The higher the market value, the higher the point-based volatility.

Evaluating Performance

CHAPTER 2

FIGURE

Better Measures

2.2

Least-square regression line of the point-based volatility.

15

16

PART 1

FIGURE

2.3

Least-square regression line of the percentage-based volatility.

Evaluating Performance

CHAPTER 2

Better Measures

17

SLIPPAGE AND COMMISSION

Although there is no way around these costs once the system is up and running, you
should not concern yourself with slippage and commission when building and
researching a trading strategy. This sounds strange, but there are several reasons for
this. First of all, when you're back-testing on historical data, you should not try to
squeeze out as many dollars or points as you can, but rather try to capture as many and
as large favorable moves as possible, while spending as little time in the market as necessary, because the m o r e time you spend in the market, the more risk you're assuming.
For instance, if you are building a system to trade the S & P 500 stock index
futures contract, it can prove tricky to c o m e up with a system that beats a simple
buy-and-hold strategy. B u t with a buy-and-hold strategy, you are in the market
1 0 0 % of the time, with the same a m o u n t of contracts for the entire period. W h a t
if you, instead, could c o m e up with a strategy that only kept you in the market 5 0 %
of the time, while the profit per contract traded only decreased 4 0 % ? To assume
the s a m e effective risk as in the buy-and-hold strategy, you now can trade twice as
m a n y contracts per time units spent in the market, but with only a 4 0 % drop in
return per contract traded. T h e final net outcome, measured in dollars, is still 2 0 %
higher than it would have been for the buy-and-hold strategy.
T h u s , taking into consideration slippage and c o m m i s s i o n at this point, which
favor systems with few and long-lasting trades, only results in the risk for a suboptimal solution that can prove to be less profitable and robust w h e n you r u n your
system real-time. In a trending market, the slippage and c o m m i s s i o n settings have
a larger impact on your b o t t o m line, the lower the value of the market. In Chapter
1, about different performance measures, we looked at how the dollar value of the
moves involved is likely to increase with the value of the market. Subtracting the
same cost for slippage and c o m m i s s i o n across all trades in such a market only lowers the impact of the low-value trades even further. T h e s a m e goes for comparing
different markets with each other. A n d finally, as already mentioned, in most cases
you shouldn't even use dollar-based calculations when y o u ' r e putting your system
together. This m a k e s the dollar-based slippage and c o m m i s s i o n assumptions obsolete as well.
First calculate the expected percentage move you are likely to catch, then
transform that m o v e into dollar t e r m s in today's market by multiplying by today's
market level and point value. T h e n deduct the proper amount for slippage and
c o m m i s s i o n . If this dollar value still looks good enough, you should take the trade.

PROFIT PER TRADE

Figure 2.4 shows you the dollar profit for every trade m a d e w i t h a simple stock
index and b o n d system traded on historical data for the S & P 500 stock index
future contract, using dollar-based calculations. Notice h o w the profits and loss-

18

PART 1

FIGURE

Evaluating Performance

2.4

The dollar value of a series of Individual trades.

es per trade have increased dramatically for the last h u n d r e d trades or so. In this
case, the average profit per trade c o m e s out to $ 7 5 5 , which also can be seen in
Figure 1.1. Now, contrast this to Figure 2.5, which shows the s a m e sequence of
trades but this time with percentage-based calculations. Notice h o w the profits
a n d losses per trade now are m u c h m o r e evenly distributed in time and c o m e out
to $ 1 , 2 6 9 , based on a 0 . 3 8 % average profit per trade and a m a r k e t value, (in
today's marketplace), of 337,500 (1,350 X 250). If you w e r e to start using this
s y s t e m today (and provided that it is robust a n d your reasoning is sound), this is
the average profit p e r trade that you could expect to m a k e in the i m m e d i a t e
future. N o t $ 7 5 5 , as implied by the performance summary. T h a t is, by m e a s u r i n g
the profit per trade in dollars, for a robust system, the value for both the profits
a n d the losses increases (decreases) at the s a m e rate at w h i c h the m a r k e t trades
higher (lower). T h i s in turn gives m o r e weight to the m o r e distant results and consequently lowers (raises) the value of the average trade. M e a s u r i n g the profit per
trade in p e r c e n t a g e s , however, all trades are given an equal weighting. For a
robust system, the percentage returns for both the profits and the losses stays the
s a m e no matter what the market is doing. O n c e the system building process is
over, it is easy to transform the average percentage return into a dollar value
applicable to today's m a r k e t situation.

CHAPTER 2

19

Better Measures

FIGURE

2.5

The same sequence of trades as in Figure 2.4 but with percentage-based calculations.

To calculate the percentage profit per trade in TradeStation you can use the
following Easy L a n g u a g e code:
Vars: TotTr(O), Prof(0);
TotTr = TotalTrades;
If TotTr > TotTr[l] Then Begin
Prof = 1 + PositionProfit(l)/(EntryPrice(l) X BigPointValue);
To export your data to a text file, the E a s y L a n g u a g e code is as follows:
Vars: Prof(0), TradeStr2("");
TradeStr2 = NumToStr((Prof - 1) XI00, 2) + NewLine;
File Append("C:\Temp\Chap 1 -2.csv", TradeStr2);
O n c e in Excel (or the spreadsheet p r o g r a m of your choice), type in the following formula at the b o t t o m of the column w h e r e the data are stored:
=AVERAGE(A1 :AX)
w h e r e A denotes the column for data stored and X denotes the n u m b e r of
rows/trades.

20

PART 1

Evaluating Performance

To calculate your average w i n n e r ( " > 0 " ) and loser ( " < = 0 " ) use the following formula:
=SUMIF(A1 :AX,">0")/COUNTIF(A1 :AX,">0")
If you also would like to calculate these values for the short and the long side
respectively, you m u s t first m u s t modify your E a s y L a n g u a g e export function to:
TradeStr2 = NumToStr(MarketPosition(l), 0) + "," + NumToStr((Prof - 1)
X100, 2) + NewLine;
Finally, to calculate the dollar value of all trades into today's market value,
multiply the results by the current trading level and dollar value per trade, either
in the s a m e or in an adjacent cell. If typed into the s a m e cell, the formula for the
average dollar value p e r trade would look like:
=AVERAGE(A 1: AX)/100 X 250 X1,350,
w h e r e 250 is the dollar value per point a n d 1,350 is the current trading level for
the S & P 500.

LARGEST WINNING/LOSING TRADE

F r o m a m o n e y m a n a g e m e n t standpoint, it is m u c h m o r e important to know your
largest losing trade than to k n o w your drawdown. To calculate your largest w i n ning trade, type in the following formula in the spreadsheet:
=MAX(A1:AX)
w h e r e A denotes the column for w h e r e the data is stored, and X denotes the n u m b e r of rows/trades.
A n d for your largest losing trade, the formula is:
=MIN(A1:AX)
w h e r e A denotes the c o l u m n for where the data are stored, a n d X denotes the n u m ber of rows/trades.
It also is very important, from a m o n e y m a n a g e m e n t standpoint, to keep all
trades as uniform as possible. Therefore, it is of importance to know your standard
deviation intervals. To calculate the standard deviation interval for all trades, type
in the following formula in Excel:
=STDEV(A1:AX)
To get the two standard deviation interval, simply multiply by two. In our
case, with the system traded on the S & P 500 stock index futures contract, this
interval c o m e s out to 0 . 2 3 % 3.36%, or $781 11,326 in dollar terms in today's

CHAPTER 2

21

Better Measures

market. This m e a n s , with 9 5 % certainty, the true average profit for this system
should lie somewhere within the interval $10,545 to $12,107. T h e smaller the
interval, the m o r e uniform the trades and the less risky the system. Notice, however, that in this case the interval stretches well into negative territory. This
explains why a seemingly good system suddenly can start to p r o d u c e large losers.
M a n y t i m e s the system has not b r o k e n down, but with a large n u m b e r of trades
m a d e on unseen data, it simply has m o v e d closer to its true average per trade,
which happens to be negative. To keep the standard deviation interval completely
in positive territory is virtually impossible, a n d I have yet to see such a non-curve
fitted system. In fact, I believe that the system that I ' m using here for demonstration p u r p o s e s is v e r y robust, with very uniform trades, and that it will continue to
stay in positive territory for m a n y years to c o m e , although it undoubtedly has a few
very scary features as well.
With this information in hand, you also can calculate the likelihood of failure
as the distance between the zero-profit line down to the lower standard deviation
boundary, divided by the total distance between the two standard deviation boundaries. In this case, it comes out to 0.46, or 4 6 % (= 10,545 / (10,545 + 12,107)).
Obviously, this too is a risk measure that you should try to keep as low as possible.
To get a better feel for w h a t you can expect from your system (less any freak
occurrences) and depending on w h a t type of system you're building/trading, it also
may be a good idea to w e e d out all outlayer trades before you continue to investigate it further. R e m e m b e r , however, that s o m e s y s t e m s ' sole p u r p o s e is to catch
as m a n y "outlayers" as possible, to m a k e up for long strings of expected small
losses. It also is questionable to w e e d out too m a n y (if any at all) losers and, hence,
fool yourself into believing that your system is better than it really is, especially
considering that you cannot rule out that the true, future value of the average trade
easily could be negative.
1

CUMULATIVE PROFIT AND EQUITY TOPS

T h e cumulative profit isn't really of any interest at this point, but is only needed
to calculate other important measures such as drawdowns and flat times (the time
spent b e t w e e n two equity highs). To calculate and export the cumulative profit and
equity tops in TradeStation's E a s y L a n g u a g e you can use the following formulas:
Vars: CumProf(l), ETop(0), TradeStr2("");
CumProf = CumProf x Prof;
ETop = MaxList(ETop, CumProf);
TradeStr2 = NumToStr((CumProf - 1) X 100, 2) + "," + NumToStr((ETop - 1)

An outlayer trade is a trade that is judged to be so far away from the average trade that it can be considered a "freak occurrence". Usually trades that are more than three standard deviations away from the average trade are considered outlayers.

22

PART 1

Evaluating Performance

X 100,2) - NewLine;
FileAppend("C:\Temp\Chapl-2.csv",TradeStr2);
T h e plot in Figure 2.6 shows the cumulative, percentage-based equity curve for
a version of the Black Jack system. The trick is to m a k e all lines coincide with each
other as m u c h as possible and keep the " b o x e s " as small and as few as possible. T h e
squiggly line is the cumulative profit, the upper straight line depicts the latest equity peak, and the lower straight line is the latest equity drawdown.
T h e reason the cumulative profit is of lesser interest at this point is that as
long as the system is considered to be robust, with a stable profit factor a n d a constant relative a m o u n t of winners, a n d has a positive expectancy p e r trade b a s e d on
today's market value a n d less expected costs for slippage and commission, the
final equity g r o w t h is only a result of how aggressively you choose to trade it
and/or what type of m o n e y m a n a g e m e n t strategy you use.
Also, in the calculation above we are assuming that the entire equity, including
previously m a d e profits, is being reinvested each time. This is not all that correct,
especially not in the futures markets, but could be so in the stock market, provided
that you can b u y fractions of a share. W h a t it does allow you to do, however, is to
compare different systems and markets with each other on an equal basis, or test how

FIGURE

2.6

The cumulative, percentage-based equity curve for a version of the Black Jack system.

CHAPTER 2

Better Measures

23

your strategy would have held up compared to a buy-and-hold strategy over the same
time period.

DRAWDOWN
M a n y traders and analysts prefer to calculate the drawdown as the m a x i m u m intratrade drawdown, or total equity drawdown ( T E D ) , including both closed trade
drawdown ( C T D ) and open trade d r a w d o w n ( O T D ) in the same figure. To build a
robust trading system, however (as already mentioned in Chapter 1), this is not the
way to go about finding the optimal solution. T h e problem really consists of several parts. First, you must m a k e each and every trade as efficient as possible, by
dividing the O T D into the start trade drawdown ( S T D ) and the end trade drawdown ( E T D ) . This is achieved by a trade-by-trade efficiency analysis primarily
using John Sweeney's m a x i m u m adverse excursion ( M A E ) and m a x i m u m favorable excursion ( M F E ) techniques {Campaign Trading: Tactics and Strategies to
Exploit the Markets [Wiley Finance Editions, 1996] and Maximum Adverse
Excursion: Analyzing Price Fluctuations for Trading Management [Wiley Trader's
Advantage Series, 1997]).
O n c e this is done, you should have a trade m a n a g e m e n t technique that will
let you trade (at least marginally) profitably over a wide variety of markets, time
frames, and market conditions, no matter what entry technique you use. T h e n , a n d
only then, should you finalize your entry technique and start to look for a high
probability filter technique to c o m b i n e with your trade-by-trade management. The
drawdown n u m b e r s you then c o m e up with should be for closed out trades only.
As mentioned, the drawdown really isn't that interesting when it comes to deciding
whether a system is robust and profitable enough to trade. But that doesn't free you
from the responsibility to investigate it as thoroughly as you can to make sure that
you're properly capitalized when you start trading. Take another look atTradeStation's
performance summary in Figure 1.1. This system has a drawdown of $18,178.
"Great," you think: "I can stomach a drawdown of less than $20,000, and if I start out
with $60,000, the worst expected drawdown will only make up a third of my initial
capital." But nowhere does it say when in time this drawdown happened or how large
it was in relation to where the market was trading at the time (in this case it happened
during mid-1998, when the market was trading around the 1,000 level).
A good way to get a feel for the drawdown for any trading strategy is to chart
it as a so-called underwater equity curve, as in Figure 2.7, which shows the depth
of the historical drawdown. From this chart you can see that the worst equity dip
for this system is about 8%, which corresponds to approximately $27,000 in today's
market value (1,350 X 250 X 0.08). This, in turn would have taken away close to
5 0 % of your initial equity of $60,000, as mentioned above. If you do not want to
experience a drawdown of, for instance, one-third of your equity right off the bat,
you would need an initial trading capital of at least $81,000 were you to start trad-

24

PART 1

FIGURE

Evaluating Performance

2.7

The underwater equity curve measures the depth of the historial drawdowns.

ing this system today. To c o m e to the necessary conclusions, the drawdowns must
be measured in percentages and then transformed into today's market value.
O n l y the most risk-seeking people, w h o thrive on stomach aches and sleepless nights, would trade a system with an expected drawdown closing in on 5 0 %
of their equity. That is in addition to all those w h o do not know that the expected
d r a w d o w n of their system usually is m u c h higher than w h a t their system trading
software's performance s u m m a r y tells them. R e m e m b e r also that your worst drawdown is always still to c o m e . In fact, not only is your worst drawdown still to
c o m e t h e truth is that you will eventually be completely w i p e d out. It is only a
question of h o w long you stick around. If we all stick around long enough, the
probabilities have it that we all will be w i p e d out sooner or later. It is j u s t a matter of time. T h e trick then, is to do your homework as well as possible so that you
can keep disaster in front of you and quit while you are still ahead. (Or, as in the
case above, avoid getting started in the first place.)
This little piece of TradeStation code shows you h o w you can export the n e c essary drawdown information for your closed out trades to a text file for further
analysis in Excel:
Vars: CumProf(l), EBot(O), EDraw(O), TradeStr2("");
EBot = MinList(EBot, CumProf);

CHAPTER 2

Better Measures

25

EDraw = CumProf / ETop;

TradeStr2 = NumToStr((EBot - 1) X 100, 2) + '*," + NumToStr((EDraw - 1) X
100, 2) + NewLine;
FileAppend("C:\Temp\Chapl-2.csv",TradeStr2);
From Figure 2.7 it also is evident that this system will keep you in the red an
awfully long time and only on occasion will you be able to get b a c k to the surface
and set a n e w equity high. Even the best of systems will spend most of their t i m e
in drawdowns.

FLAT TIME AND RUN UPS

Take a look at Figure 2.6 again. This looks like a pretty decent equity curve with
a steady upward slope and tolerable drawdowns. But h o w long do you have to
spend between two equity highs? This is revealed in Figure 2.8, which shows the
so-called flat time, the t i m e spent between two equity highs. (There is a risk of
confusion of terms here. For m a n y analyst a n d traders, flat time m e a n s the time
spent in between trades, or in a directionally neutral position. In this book, however, this alternative interpretation only m a k e s up a part of the total flat time, as

FIGURE

2.8

The flat time measured as the time spent between two equity highs.

PART 1

26

Evaluating Performance

we define it.) A good system should have a m a x i m u m flat time of no longer than
18 months. In this case, with a flat time of little m o r e than 4 0 0 days, we c o m e pretty close to w h a t is tolerable.
If you started to trade this system just prior to this drawdown, y o u would
have started out losing, stuck it out for approximately 1.5 years just to get your
m o n e y back, a n d eventually broke even. But d o n ' t despair; drawdowns are simply
something you have to learn to live with and the truth of the matter is that even the
m o s t robust and profitable trading strategies will keep you feeling on your way
d o w n m o r e often than on your way u p .
To chart the time spent between two equity highs and to calculate the average, you m u s t add the following E a s y L a n g u a g e code (which also allows you to
calculate the time spent between equity lows) to the system:
Vars: Etop(0), TopBar(O), EBot(O), BotBar(O), TopInt(O), Botlnt(O);
IfETop > ETop[l]Then
TopBar = CurrentBar;
IfEBot < EBot[l]Then
BotBar = CurrentBar;
Toplnt = CurrentBar TopBar;
Botlnt = CurrentBar - BotBar;
TradeStr2 = NumToStr(TopInt, 0) + "," + NumToStr(BotInt, 0) + NewLine;
FileAppend("C:\Temp\Chap 1 -2.csv", TradeStr2);
T h e average time for this system spent in a flat is 86 bars, and the average
time between equity lows is 66 bars. In "psychological t e r m s " this m e a n s that only
about four times per year, on average, will you be allowed to feel like a really successful trader, setting a n e w equity high; you will spend five times every year feeling really rotten, setting a new equity low before the markets finally start to move
your way again.
M o s t institutional investors will not tolerate a drawdown over 30 to 3 5 % of
total equity, or a m a x i m u m flat time of m o r e than 18 months. So, if you are aspiring to selling services to the "big money," these are the measures within which you
m u s t keep yourself.

CODE
In this chapter we looked at how to build a new set of system testing measures that
give you m o r e realistic results that are m o r e likely to hold up in the future. T h e
most important m e a s u r e s w h e n you are building your system are average expected profit p e r trade and largest expected losing trade. Others, like the expected
d r a w d o w n and expected flat time, are only of interest from a psychological point
of view, as long as you can keep t h e m within tolerable levels. However, all vari-

CHAPTER 2

27

Better Measures

ables m u s t be e x a m i n e d equally thoroughly. If you do not feel comfortable with all

of them, you probably are b o u n d to lose, no matter w h a t the market is doing or
h o w well the s a m e system s e e m s to be w o r k i n g for others.
B e l o w is all the code used in this chapter, put together into one piece:
Vars: FName(""), TotTr(O), Prof(0), CumProf(l), ETop(l), TopBar(O), Toplnt(O),
,m

BotBar(O), Botlnt(O), EBot(l), EDraw(l), TradeStr2( );

If CurrentBar = 1 Then Begin
FName = "C:\Temp\" + LeftStr(GetSymbolName, 2) + ".csv";
FileDelete(FName);
TradeStr2 - "E Date" + "," + "Position" + "," + "E Price" + "," + "X Date" +
"," + "X Price" + "," + "Profit" + "," + "Cum. prof." + "," + "E-Top" + "," +
"E-Bottom" + "," + "Flat time" + "," + "Run up" + "," + "Drawdown" +
NewLine;
FileAppend(FName, TradeStr2);
End;
TotTr = TotalTrades;
If TotTr > TotTr[l] Then Begin
Prof = 1 + PositionProfit(l)/(EntryPrice(l) X BigPointValue);
CumProf = CumProf X Prof;
ETop = MaxList(ETop, CumProf);
Tf ETop > ETop[l] Then Begin
TopBar = CurrentBar;
EBot = ETop;
End;
EBot = MinList(EBot, CumProf);
IfEBot<EBot[l]Then
BotBar = CurrentBar;
Toplnt = CurrentBar - TopBar;
Botlnt = CurrentBar - BotBar;
EDraw = CumProf / ETop;
TradeStr2 = NumToStr(EntryDate(l), 0) + "," +
NumToStr(MarketPosition(l), 0) + "," + NumToStr(EntryPrice(l), 2) +
"," + NumToStr(ExitDate(l), 0) + "," + NumToStr(ExitPrice(l), 2) + ","
+ NumToStr((Prof - 1) X 100, 2) + "," + NumToStr((CumProf - 1) X
100, 2) + "," + NumToStr((ETop - 1) X 100, 2) + "," +
NumToStr((EBot - 1) X 100, 2) + "," + NumToStr(TopInt, 0) +
"," + NumToStr(BotInt, 0) + "," + NumToStr((EDraw - 1) X 100, 2) +
NewLine;

28

PART 1

Evaluating Performance

FileAppend(FName, TradeStr2);
End;
O n c e you have exported all the necessary data into Excel or any other
spreadsheet program, you can use it to put together your own performance s u m mary, which will be m u c h m o r e telling about your system than any other performance s u m m a r y you are likely to find in any other program. F r o m n o w on, we
will work extensively with a type of performance s u m m a r y that resembles that in
Table 2 . 1 .
To put together a performance s u m m a r y like this, type the following formulas into your spreadsheet p r o g r a m :
For the average percentage profit, in cell C ( X + 4 ) :
=AVERAGE(F$2: FX)/100
where F denotes the c o l u m n w h e r e you have stored the profit from each individual trade, and X denotes the n u m b e r of rows/trades.
For the standard deviation in percent, in cell C ( X + 5 ) :
= STDEV(F$2:FX)/100
For the n u m b e r of winners, in cell F ( X + 2 ) :
,,

=COUNTIF(F$2:FX, >0"),
For the largest (%) winner, in cell F ( X + 3 ) :
=MAX(F$2:FX)/100
For the average (%) winner, in cell F ( X + 4 ) :
=SUMIF(F$2:FX,">0")/COUNTIF(F$2:FX,">0")/100
For the cumulative (%) profit, in cell F ( X + 5 ) :
=GX/100
w h e r e G denotes the c o l u m n w h e r e you have stored the cumulative profit.
For the n u m b e r of losers, in cell I ( X + 2 ) :

TABLE

2.1

Example of performance summary.

Total trades

150

Profit factor

1.80

Winners
Lrg winner

Avg profit

0.38%

1,269

Avg winner

St Dev

1.61%

5,421

Cum profit

92

61.33%

58

38.67%

4.88

16,470

Losers
Lrg loser

-5.74%

-19,373

1.38

4,667

Avg loser

-1.22%

-4,121

74.12

250,155

Drawdown

-7.76%

-26,190

CHAPTER 2

Better Measures

29

=COUNTIF(F$2:FX,"<=0")
For the largest (%) loser, in cell I ( X + 3 ) :
MIN(F$2:FX)/100
For the average (%) loser, in cell I ( X + 4 ) :
=SUMIF(F$2 :FX," <=0")/COUNTIF(F$2:FX," < = 0 " ) / 1 0 0
For the largest (%) drawdown, in cell I ( X + 5 ) :
=MIN(L$2:LX)/100
w h e r e L denotes the c o l u m n for w h e r e you have stored the drawdown.
For today's dollar value of the average trade, in cell D ( X + 4 ) :
= C ( X X 4 ) X 1350X250
w h e r e 1350 denotes today's index value, and 250 denotes dollar value per point.
For today's dollar value of the standard deviation, in cell D ( X + 5 ) :
= C ( X + 5 ) X 1350X250
For the n u m b e r of trades, in cell D ( X + 2 ) :
=F(X+2)+I(X+2)
For the percentage of winners, in cell G ( X + 2 ) :
=F(X+2)/D(X+2)
For today's dollar value of largest winner, in cell G ( X + 3 ) :
= F ( X + 3 ) X 1350X250
For today's dollar value of average winner, in cell G ( X + 4 ) :
= F ( X + 4 ) X 1350X250
For today's dollar value of the cumulative profit, in cell G ( X + 5 ) :
= F ( X + 5 ) X 1350X250
For the percentage of losers, in cell J ( X + 2 ) :
=I(X+2)/D(X+2)
For today's dollar value of largest loser, in cell J ( X + 3 ) :
= I ( X + 3 ) X 1350X250
For today's dollar value of average loser, in cell J ( X + 4 ) :
= I ( X + 4 ) X 1350X250

30

PART 1

Evaluating Performance

For today's dollar value of the largest drawdown, in cell J ( X + 5 ) :

= I ( X + 5 ) X 1350X250
For the profit factor, in cell D ( X + 3 ) :
= A B S((F(X+2) X G(G+4))/(I(X+2) X J(X+4)))

CHAPTER

Futures Contract Data

In Chapters 1 and 2 we looked at those performance measures that could be derived

directly from TradeStation's performance summary, and those that we had to export
into a spreadsheet p r o g r a m for further analysis to m a k e t h e m m o r e forward looking. In Chapter 2 we also concluded that to achieve the latter, we had to base our
calculations on percentages rather than dollars or points.
If you are a c o m m o d i t y futures trader, one of the m a i n hurdles to overcome
w h e n testing trading strategies on historical futures data is the limited life span of
a futures contract. To overcome this, various m e t h o d s have been invented to splice
several contracts together to form a longer time series. In this chapter, we take a
closer look at these different m e t h o d s , look at their pros and cons, and explore suggestions on w h e n and how to use them. This is especially important when it c o m e s
to m a k i n g your system reports as forward-looking as possible, using the percentage-based calculations described in the previous chapter.
There are three different m e t h o d s to go about splicing contracts together. T h e
nonadjustment method, the back-adjustment method, and the perpetual-adjustment
method. T h e back-adjustment m e t h o d can further be subdivided into point-based
adjusted a n d ratio-adjusted.

NONADJUSTED DATA
With the nonadjustment method, you simply stop charting o n e contract when it
expires or w h e n you otherwise d e e m it justifiable to do so and continue to chart the
next contract in line, w h i c h n o w is the new front contract. Usually, this coincides
with w h e n the market as a whole moves from one contract to the next, resulting in
31

32

PART 1

Evaluating Performance

an increase of the open interest for the n e w front contract surpassing the open
interest for the old contract. M a n y traders, however, prefer to roll over on a certain
day of the month or when there are a certain n u m b e r of days left on the life of the
old contract. If there happens to be a difference in price between the new and the
old contract, it is left unadjusted.
T h e m a i n advantage of the nonadjusted time series is that, at each point in
time, it shows you exactly how the front contract at that particular time w a s traded, with all trading levels and price relationships intact and exactly as they
appeared in real life. Figure 3.1 illustrates w h a t this looked like for the D e c e m b e r
contract on the S & P 500 index during the crash of 1987. F r o m the high of 333 on
October 2, the market fell a total of 152 points, or $38,000 (152 X 2 5 0 ) to the low
of 181 on October 20. In percentage terms, this equals a drop of 4 5 . 6 % (152 / 333)
of the total market value ( r e m e m b e r these n u m b e r s ) .
T h e m a i n disadvantage of the nonadjustment m e t h o d is the difference in
price that frequently appears on the day of the roll, which will distort your backtesting results. Therefore, nonadjusted data is best for day traders or other c o m modity futures traders with very short trading horizons, w h o m o r e often than not
close out all their trades at the end of the day. With the TradeStation code provided in Chapter 1 and 2 it is easy to export all trading results from several different
and separately tested contracts into the same spreadsheet for a c o m b i n e d analysis.

POINT-BASED BACK-ADJUSTED DATA

M o s t markets trade at cither a p r e m i u m or a discount the longer in t i m e you look.
If we continue to use the S & P 500 as an example, it usually trades at a p r e m i u m ,
m e a n i n g that the new front contract usually trades slightly higher than the old one
on the day of the roll. H a d you been in an open position at this time, during your
historical profit testing, this difference would have been added or subtracted to the
result of that particular trade. Over time, this error adds u p . To overcome the distortions on the historical systems testing results induced by the nonadjustment
method, the point-based back-adjusted contract was invented.
If the new front contract is trading at a p r e m i u m c o m p a r e d to the old contract, the entire historical time series leading up to the roll is adjusted upward by
that distance. For instance, if, on the day of the roll, the new contract closes at
1,309.5 and the old contract at 1,296.9, a n d the day before at 1,318.6, the entire
time series up to that point of the roll is adjusted upward by 12.60 points (1,309.5
1,296.9), resulting in a new artificial value of 1,331.2 for the last close used
from the old contract (12.6 + 1,318.6). A similar but opposite adjustment takes
place each time the n e w contract is traded at a discount.
Figure 3.2 shows w h a t the October 1987 market action looks like using a
point-based back-adjusted contract, with the latest roll m a d e in September 1999.
In this chart, the high on October 2 is at 567.35 and the low on October 20 is at

CHAPTER 3

FIGURE

Futures Contract Data

3.1

The crash of 1967 viewed through the nonadjusted contract

33

34

PART 1

FIGURE

3.2

The crash of 1987 viewed through the point-based back-adjusted contract.

Evaluating Performance

CHAPTER 3

Futures Contract Data

35

4 1 5 . 3 5 , for a total difference of 152 points. B e c a u s e of the transition upward, however, the original percentage difference of 4 5 . 6 % n o w has decreased to 2 6 . 8 % (152
/ 567.35). What's m o r e , for each roll all these contract values will continue to
change slightly, presumably upward, resulting in an ever-decreasing percentage
difference.
This continuing decrease (increase) of the relative importance of historical
moves in an up-trending (down-trending) market a n d markets that are traded at a
p r e m i u m (discount) is the m a i n disadvantage of the point-based back-adjusted
contract. For instance, if the 2 6 . 8 % d r o p a s implied by the back-adjusted contractwere to be used to calculate the dollar value of the drop from the actual top
of 3 3 3 , we would c o m e up with $22,311 (0.268 X 333 X 250) instead of $38,000
(0.456 X 333 X 250). Therefore, to keep the relative importance of all historical
moves constant, they m u s t be viewed in percentage terms. By the s a m e token,
those w h o claim that the volatility of the stock market has increased over the years
and, therefore, has b e c o m e m o r e difficult to trade, do not k n o w what they are talking about. Yes, the dollar- or point-based volatility has increased considerably, but
this is only a natural consequence of the market's trading higher.
Viewed in percentage terms, the volatility, if anything, actually has decreased
somewhat. This is illustrated by Figures 2.1 through 2.3, which show a nonadjusted monthly chart of the nearest future on the S&P 500 index, covering the period
late 1985 to late 1999, together with the absolute monthly changes in points and
percentages, relatively. If we look at Figure 2 . 1 , it seems that the price bars have
b e c o m e wider and wider in recent time, indicating that the market has b e c o m e m o r e
volatile and difficult to trade. This is only a consequence of the increasing market
value, however, a n d should prove no problem if dealt with properly.
T h e straight lines that run through the two latter charts are best-fit regression
lines. As you can see, the regression line for the absolute point changes has
increased over the years from close to 0 to over 30 points p e r month. Contrast this
to the regression line for the absolute percentage change, which has stayed approximately constant (the slope is slightly down) around 3 . 5 % p e r month. That is,
measured in dollars or points, the volatility has indeed increased considerably over
the years, as the regression line in Figure 2.2 indicates. But, measured in percentages, the volatility has stayed approximately the s a m e since late 1985. In fact, if
anything volatility has decreased somewhat.
W h a t immediate consequences does this have for you as a trader? Well, aside
from those topics discussed in the previous chapters, one important thing to
r e m e m b e r is never to use dollar- or point-based stops. Consider, for instance the
following example:
U s i n g TradeStation, a $5,000 m o n e y m a n a g e m e n t stop (equal to approximately 1.5% of current market value, with the market trading around 1,350) a n d a
$10,000 trailing stop (approximately 3% of current market value) were added to a
standard breakout system. In back-testing this system on nonadjusted S & P 500

PART 1

36

FIGURE

3.3

The crash of 1987 viewed through the RAD contract

Evaluating Performance

CHAPTER 3

Futures Contract Data

37

data, covering the period January 1983 to April 1999, it took 55 trades (out of a
total of 197 trades) before any of these stops got hit in October 1987. After this, it
t o o k another 35 trades before they were hit again in October 1990, only showing
up again in M a r c h 1996. F r o m t h e n on, however, the hits b e c a m e m o r e frequent;
for the period D e c e m b e r 1996 to present, 25 out of 33 trades were stopped out, and
from July 1998 to present, 13 out of 13 trades w e r e stopped out this way.

RATIO-ADJUSTED DATA (RAD)

To overcome the d i l e m m a induced by the point-based back-adjusted contract, I
suggested in an article for Futures m a g a z i n e ("Data Pros and C o n s , " Futures m a g azine, June 1998) and also later in a second article ("Truth Be Told," Futures m a g azine, January 1999) that a better wa y to adjust a spliced-together time series
would be to use percentages rather t h a n points or dollars. If we do so, this n e w
time series will also vary in level as c o m p a r e d to w h e n the true contract was traded, but instead of keeping the point- or dollar-based relationship between two
points in time intact, the percentage relationship stays the s a m e . This c o m p a r i s o n
is d o n e w i t h the help of the following formula:
Cmew = c

i o l d

X(l+(C-c)/C)

Where:
c

inew
ioid

price i bars ago.

T h old price i bars ago.

e

C T h e close of the n e w contract on the day of the roll,

c = T h e close of the old contract on the day of the roll.
Figure 3.3 shows what the October 1987 crash looks like through the perspective of the ratio-adjusted data ( R A D ) contract, with the latest roll m a d e in September
1999. This time the high of October 2 is at 486 and the low of October 20 at 264.15,
for a point difference of 221.85 points, but with a percentage difference once again
equal to 4 5 . 6 % ((486 - 264.15) X 100 / 486). This increase in magnitude measured
in points is due to the upward transition because of the many rolls. T h e important
thing, however, is that the percentage-based move stays intact at 4 5 . 6 % .
After the crash of 1987 it t o o k close to 11 years before the S & P 500 experie n c e d a larger d r o p m e a s u r e d in points. This h a p p e n e d in the a u t u m n of 1998. This
was a n n o u n c e d as big n e w s by the media, with s o m e analysts and market writers
even c o m p a r i n g it to the crash of 1987, implicitly suggesting that this w a s equally as bad or at least w a s soon to be. To get a better feel for the benefits of the R A D
contract, let us take a closer look at this recent event and c o m p a r e it to the crash
of 1987 (see Figure 2.1). D u r i n g October 1987, the market fell by 152 points, or
$ 3 8 , 0 0 0 , or 4 5 . 6 % ; this we know. D u r i n g the a u t u m n of 1998, the market fell from
a July high of 1199.4 to an October low of 929, for a total drop of 270.4 points (as

38

PART 1

Evaluating Performance

measured on the nonadjusted contract). In dollar t e r m s , this equaled a drop of

$67,600 (270.4 X 250). T h a t is almost twice as m u c h as the October 1987 drop.
In percentage terms, however, the drop of 1998 only equaled 2 2 . 5 % of total market value. That is, the drop in 1998 was not even half as b a d as the crash of 1987.
In fact, to equal the crash of 1987 in relative t e r m s , the 1998 drop would have had
to continue all the way down to the 652.5 level, for a total drop of 546.9 points, or
$136,725.
T h u s , the crash of 1987 is still the largest relative drop in equity in modern
times, a n d if you, in your system building and analysis work, would like to treat it
as such, to m a k e your systems m o r e robust and give t h e m a better chance to hold
up in the future, the only way to do that is to use the R A D contract in combination
with the percentage-based performance m e a s u r e s described in the previous chapters. A n d while at it, do not forget to get rid of those dollar-based stops and profit targets as well.
Let's stop for a little quiz: B a s e d on the information so far, what is the
expected worst drop in equity for the S & P 500 if it were to equal the drop of 1987
and with the market trading at the 1,350-level? Is it
A.

$38,000 (152 X 250)

B.

$23,477 (0.282 X 333 X 250)

C.

$54,625 (218 X 250)

D.

N o n e of the above

If you answer " n o n e of the above," what is the correct answer?

T h e answer is: N o n e of the above. B a s e d on what we have learned so far the
expected worst drop in equity, with the market trading at the 1,350 level is
$153,900 (1,350 X 250 X 0.456). Furthermore, just because the drop in October
1987 stopped at 4 5 . 6 % , there is no saying that there will be no larger drops c o m ing. No wonder then that your software supplier, system vendor, or market guru of
choice says that your worst drawdown is still to c o m e a n d they do not even know
how to calculate it.

MULTIMARKET PORTFOLIOS
T h e benefits of the R A D contract also b e c o m e evident when you want to put
together a multimarket portfolio. We discuss multimarket portfolios and trading
strategies in m o r e detail in Part 5. For now we only state that the percentage-based
calculations do not take into consideration h o w m a n y contracts y o u ' r e trading and,
therefore, give each market an equal weighting in the portfolio. Let us look at the
following example:
U s i n g a regular 20-day breakout system that is always in the market, I h y p o thetically traded the Japanese yen and the S & P 500 over a period of 10 years,
resulting in 71 trades for the Japanese yen and 79 trades for the S & P 500. Figure

CHAPTER 3

Futures Contract Data

39

3.4 shows what the individual and c o m b i n e d equity curves would have looked like
had we b e e n able to place all these trades for real, trading o n e contract. In short,
the Japanese yen would have e n d e d up with a profit of close to $60,000, and the
S & P 500 with a profit of approximately $20,000, resulting in an overall profit for
the portfolio of about $80,000. However, to m a k e results comparable between the
two markets and at the same time relevant to today's markets, we m u s t look at how
m u c h a percentage m o v e in each market is worth today. It turns out that a 1% move
in the S & P 500, in today's market, is worth approximately $3,275, but only $1,200
in Japanese yen. H e n c e , had you been able to place all trades in today's markets,
you would have had to trade approximately five Japanese yen contracts for every
two S & P 500 contracts (3,275 / 1,200), which in turn would have resulted in an
overall profit of approximately $240,000.
Now, wait a minute, you say, those results are purely hypothetical. H o w can
I place all the trades in the same market at presumably the s a m e point in time?
Well, you can't, so that is a good and valid question; but let me ask you, can you
place any of these trades for real, no matter how you do it? N o , of course not. They
all represent foregone opportunities. Isn't it better then to at least place t h e m h y p o thetically in today's marketplace to get a feel for what might happen today, rather
than in a ten-year-old market situation to get a feel for how the situation was back

FIGURE

3.4

The results from a multi-market portfolio traded on a one-contract basis.

PART 1

40

Evaluating Performance

then? I think we all can agree that it is better to k n o w w h a t might happen today,
rather than what h a p p e n e d ten years ago.
Nonetheless, this is looking good. By m a k i n g the profits from the two markets m o r e comparable to each other we have raised the total profit from $80,000
to $240,000. W h a t we forget, however, is that this relationship changes with time,
something we have not taken into consideration in the above example.
Furthermore, from the previous chapters we k n o w that only looking at hypothetical dollars m a d e is not the m o s t efficient way to j u d g e a trading strategy. A n d
because both the dollar moves and the contract relationships between different
markets are constantly changing, we m u s t put together a chart based on the (cumulative) percentage moves we would have been able to catch. As you can see from
Figure 3.5, this results in an approximately 4 0 % c o m b i n e d move in the right direction (the market moved in favor of our position) for the Japanese yen, but in an
almost 2 5 % c o m b i n e d move in the w r o n g direction (the market moved against our
position) for the S & P 500, resulting in an overall c o m b i n e d move that barely m a n aged to reach a profit of 5 % . (To understand how a 4 0 % positive m o v e and a 2 5 %
negative m o v e can end up being close to zero, multiply 1 by 1.4 (1 + 0.4) and then
by 0.75 (1 - 0.25), that is: 1 + 4 0 % - 2 5 % = 1 X 1.4 X 0.75 = 1.05 = 1 X (1
- 0.05) = 1 + 5%.)

FIGURE

3.5

The results from a multi-market portfolio traded on a percentage basis with compounded returns.

CHAPTER 3

Futures Contract Data

41

T h u s , what at a first glance seemed to be a profitable portfolio w h e n tested

on a one-contract basis in dollar terms, turns out to be a close-to-losing endeavor
w h e n giving both markets an equal percentage weighting and c o m p o u n d i n g the
results.

PERPETUAL CONTRACTS
No matter h o w good the back-adjusted contracts are there is a certain situation
w h e n n o n e of them should be used. That is when you m u s t c o m p a r e the developm e n t of two different markets, p e r h a p s to find the one that seems to have the
strongest relative strength. T h i s is because none of these contracts stays stationary
in time, but each is recalculated each t i m e there is a roll from one contract to the
next, which in turn can lead to perceptual changes in the historical relationship
between the markets.
Figure 3.6 shows a ratio spread between the S&P 500 and the T-bill, as it is
perceived to have looked in D e c e m b e r 1984, using the point-based back-adjusted
contract. W h e n the squiggly line is below its fast and slow moving averages, the
trend for the spread is down a n d the T-bill can be expected to outperform the stock
market and, in this case, trigger a short position in the S & P 500 futures. Overall,
this looks like a smooth ride with few erroneous signals a n d whipsaw trades,
w h i c h leaves plenty of r o o m for a quick buck, trading the S & P 500 from the short
side. Now, look at Figure 3.7, which shows you the s a m e market action, as it
looked in D e c e m b e r 1984, through the actual contracts traded at the time. This is
what you actually would have seen had you monitored this spread at the time. This
time the picture is no longer as clear, with m o r e j a g g e d curves and several w h i p saw trades.
W h a t looked like a fairly smooth ride w h e n looked at through the pointb a s e d back-adjusted contracts, turns out to be a little bit m o r e j a g g e d a n d dangerous w h e n looked at through the original contracts as they actually were trading at
the time. T h e difference in the two pictures is caused by the back-adjustment
process in combination with the ratio calculation, which also adds s o m e smoothing to the historical data. This can be very deceiving if you're not aware of the
problem and might m a k e you build trading systems that will not work as well as
they seemed to have done in the past. Worse yet, they might not work at all because
the underlying assumption for your strategy was not there in the first place.
To work a r o u n d this problem, you m u s t put together a so-called perpetual
contract that stays stationary in time, despite the rollovers. That is, once a historical value is calculated, it stays the same. T h e m e t h o d for calculating a perpetual
series m a d e up of two contracts (the nearest contract and the second nearest contract) is as follows:
First set a forward quote day a certain n u m b e r of days in the future and keep
this n u m b e r of days constant as time passes. Then calculate the n u m b e r of days

PART 1

42

FIGURE

3.6

A spread between two markets put together using point-based back-adjusted contracts.

Evaluating Performance

CHAPTER 3

FIGURE

Futures Contract Data

3.7

A spread between two markets put together using the actual contracts at the time.

43

44

PART 1

Evaluating Performance

between the expiration days for the two contracts a n d the forward quote day. W h e n
the distance to the forward quote day is kept constant (i.e. roll it forward one position for every new bar) the relative distance for the two contracts in relation to this
day changes for each day. Finally, multiply the relative distance for both contracts
by their respective prices and add t h e m together, for today's perpetual price. T h e
closer the nearest contract c o m e s to expiration, the less weight it gets. W h e n it
expires it b e c o m e s 0; the second nearest contract is now the nearest contract, and
the third nearest contract is the n e w second nearest contract.

PART

ONE

A Few Final Thoughts About Part 1

This part started out by taking a closer look at which system testing measures are
m o r e useful than others, and why it could be a good idea to expand the analysis
work a bit with the help of a spreadsheet program, like MS Excel or Lotus 1-2-3.
To properly evaluate a trading system, it is of paramount importance to use a set of
universal measures that give an equal weighting to all the trades, no matter where
and when they are derived. To accomplish this, it also is important to use the right
type of data. As we have seen, not all data can be used all the time; knowing when
to use what is vital in building a robust and profitable trading system.
Later chapters take a closer look at how to evaluate a system or a strategy that
is already up and running, to m a k e sure that it stays on track and performs as it
should. Before we go any further in the evaluation process, we need a couple of systems to trade. Building a system suited for a specific type of market action or trading perspective is the major topic in the next section.

45

PART

TWO

System Concepts

In the first part, we discussed a few important concepts to understand and think
about before the system building process can begin. Unfortunately, however, it does
not stop there. There is m o r e than one thing to think about w h e n trying to put
together a mechanical trading strategy. Aside from the psychological a s p e c t s
whether the system fits your style of trading, personality, trading budgetseveral
other m o r e technical questions m u s t be answered as well. Before we can start
designing and investigating any trading system, we must understand what it should
try to accomplish.
We must know if the trading methodology should be long-term, short-term, or
perhaps even intraday. Should the system be market specific or tradable over a wide
variety of markets? If the system is a multimarket system, h o w should it be built
and tested so that all the markets have an equal impact on the final result? Should
we try to capture a certain type of trending move, or should we try to capture the
shorter-term moves within a previously defined trend? Or are we trying to a c c o m plish something completely different, like arbitraging o n e market or contract month
against another, or buying or selling volatility with the help of options? T h e n there
is the question of what type of m o n e y m a n a g e m e n t strategy to use, and w h e n this
should be taken into consideration.
But even before these fundamental questions can be answered, there is a set
of basic questions of a m o r e philosophical character, which must be addressed
a n d i f not answeredat least contemplated. W h a t is long t e r m and short term? Or
what is high and low? There really are no generic, r e a d y - m a d e answers to these
questions, because long t e r m for me might be short t e r m for you, and something
completely different for s o m e o n e else. Whatever your answer is, it is correct for you
and not for anyone else to question.
47

48

PART 2

System Concepts

Before you answer, however, here is some food for thought: w h a t if, instead
of measuring long-term or short-term market activity in regular time units, such as
minutes, hours, or days, you were to m e a s u r e it in n u m b e r of bars (market time
units) encapsulating as m u c h time (regular time units) as you d e e m necessary? Or
perhaps, long t e r m and short term simply should be defined by the system, where
a breakout-type system is always considered long term while a top-and-bottom
picking system is always considered short term, no matter for how long the trade
actually lasts, whether measured in regular time units or market t i m e units?
A n o t h e r important thing to consider before researching a trading system (and
a c o m m o n l y debated question a m o n g system builders) is whether it is a good idea
to save some of the data for out-of-sample testing. T h e answer is that it depends
on the system and its underlying assumptions; m a n y times it shouldn't m a k e any
difference. In a correctly built system, y o u ' r e interested in exploiting something
that should, on average, work well in several different markets and over several different t i m e periods, not with any specific type of anomaly that might or might not
still exist in a particular market. By saving some of your data for out-of-sample
testing, you can strengthen your trust in the m o d e l , but in doing so, you also use
up valuable data that could have added even further to the robustness of the system. Only you can m a k e that choice. T h a t said, in m a n y of these examples, we
work with an out-of-sample period for comparative reasons. Whatever you choose,
it is important to r e m e m b e r that by necessity all system testing and design m u s t be
m a d e on historical data. T h e trick, then, is to m a k e as good use as possible out of
these data and m a k e your evaluation measures as forward looking as possible.
Building systems is no easy matter, and you h a d better k n o w the o d d s against
you before you get started. For instance, if you only traded the S & P 500 index
approximately every fifth day (I picked out all days randomly 20 times, calculated the c o m p o u n d e d returns, and averaged the results), between M a y 1992, and
October 1999, and if you could have been absolutely sure that all these days would
have been d o w n days, you would have m a d e a w h o p p i n g 71 times the index. If you
instead h a d b e e n able to pick the same a m o u n t of up moves, you would have m a d e
an incredibly 161 t i m e s the index. This tells us that we are m u c h better at going
with the underlying trend (as it undoubtedly has been up during the last few years),
and if we can only t e m p e r ourselves and trade only on a few selected days, the
rewards can be v e r y good. No wonder day trading is such an alluring business. But
rewards like this do not c o m e without a price. T h e above examples assume that we
can be 1 0 0 % sure of the move, but what if we can only be 5 0 % sure? Well, then
the expected return sinks to only 0.25 t i m e s index, if going long only, or to a mere
0.11 times index, if going short only.
Similarly, if y o u ' r e a long-term trader and prefer to c o m p a r e yourself to a
buy-and-hold strategy, you could have increased you return to 4.9 times the index,
had you been able to stay out of the market every 20th d a y i f you also were
absolutely sure that day also was a down day. But if you only were 5 0 % sure about

CHAPTER

Picking Tops and Bottoms

W h e n it c o m e s to trying to trade shorter term, or picking tops and bottoms, m a n y

system traders usually start out trying to use any or all of the m a n y oscillator-type
indicators that have swamped the field of technical analysis in recent years.
Examples of such indicators are RSI, Stochastic, Momentum, Rate-of-change,
MACD, and Plus and Minus DMI. M a n y others exist, and the list can be m a d e m u c h
longer, but those mentioned above are the m o s t c o m m o n and popular ones. T h e s e
indicators are believed to be g o o d top and bottom pickers because they're believed
to be able to anticipate or even predict the market by moving in and out of overbought and oversold regions, or simply by not " c o n f i r m i n g " the moves of the market. I believe, however, that one main reason you have picked up this book in the
first place is that you have c o m e to realize that m a n y of these indicators simply do
not work nearly as well in real-time as they seem to do when looked at on a historical price chart.
If all these indicators do not work, how then should one go about finding profitable short-term trading opportunities? Well, probably the most important task for
you as a trader, especially if you prefer to trade fairly short term, is to mine the data
from which you hope to m a k e a living. If, for nothing else, it m u s t be done to get
an in-depth feel for the markets you are trading, which in turn should help you to
better interpret the current market situation and the indicators you are using to
g a u g e it. But before you can go about doing this, you must know exactly what it is
you should look for and how to measure it.
To build a system that works equally well in all markets, you must decide what
type of data to use. With the ordinary point-based back-adjusted contract, you can
estimate how m u c h profit your system would have made historically in each market,
51

52

PART 2

System Concepts

but because of the way it is constructed you cannot channel any information from
one market to another. To give all moves an equal importance, and to derive the necessary input for your calculations, it is of paramount importance that you only use
percentage-based calculationsand if you are a futures trader, also use the newly
invented R A D contract (see "Data Pros and Cons," Futures magazine, June 1998,
and "Truth Be Told," Futures magazine, January 1999). In a trending market, the
dollar value of the magnitude of the moves changes with the market value, so that if
the market is in an up trend, with a steadily increasing market value, the dollar- or
point-based move is likely to increase as well. T h e relationship between the moves
and the market value stays the same, resulting in a average percentage move that is
m o r e likely to stay the same, no matter at what level the market is trading. This is
also the contract you need as soon as you start to work with any type of percentagebased decisions, such as rate-of-change analysis or percentage-based stops.
Notice that the key words here are "equally w e l l " instead of "equally profitable." To build a trading system that continues to work equally well in the future
as it did during the building process, w h e n it w a s hypothetically traded on historical data, it also is important to understand the difference between a good working
system and a profitable system. Furthermore, once this is fully comprehended, it
is even m o r e important to understand w h y a good working system is not necessarily a profitable system, but why a profitable system always m u s t be a good system. T h e s e key concepts that will be stressed continuously throughout this book.
W h e t h e r a good working system will be profitable or not has nothing to do
with the system, but rather with the level at which the market is currently trading
a n d the dollar value of that level. For instance, suppose the dollar value of the S & P
500 futures contract were to be lowered from today's $250 per point to $2.50 per
point. H o w many S & P 500 systems would still be profitable in the future, or even
test profitable in hypothetical back-testing? Probably not many. T h e point is that
the actual value of the market and its move is a technicality decided by the
exchange, which has nothing to do with whether the system is good at capturing
the moves of the market.
D u r i n g the building process, the actual dollar values are of no concern.
Instead, we focus on the universal measures: the profit factor, the percentage
move, and the n u m b e r profitable trades. To m a k e sure the results do not differ too
m u c h between markets, we also look at the standard deviations of all these measures. T h u s , a well-working s y s t e m a good systemis a system that captures as
m a n y and as large moves as possible in any and all market environments, as m e a s ured with a universal m e t h o d like those mentioned. For a system to be profitable,
however, it must be traded in a market where the moves it is designed to catch also
are worth taking. This has nothing to do with the system, but rather with the level
at which the market currently is trading and the dollar value of that level.
Once you k n o w what to look for in terms of the data and the different measurements, you also must figure out exactly what it is you would like to do. That is,

CHAPTER 4

Picking Tops and Bottoms

53

should you only take trades that go in the s a m e direction as the longer-term underlying trend, or should you also look for counter-trend moves? Whatever your
answer, you then m u s t ask yourself whether you should try to anticipate the turning
point with an early limit-order entry; or play it a little safer and look for proof that
the preceding m o v e has run its course, and then enter with a stop? Furthermore,
once you're in the trade, should you then try to ride it for as long as possible, or
should you only go with the short-term thrust and perhaps exit m a n y of the winners
with a limit order at a specific price or after a move of a certain magnitude?
In this chapter, we took a closer look at a few different system ideas that all
have a high likelihood of producing a large n u m b e r of profitable trades for a
decent overall profit. T h e first two are a couple of market-specific data m i n i n g
systems, w h e r e we only are interested in the specific characteristics of that particular market, or group of related markets, although it is fully possible and realistic
to apply both the research technique and the systems to any market. T h e next is
based on a proprietary indicator perhaps best described as a mix b e t w e e n percentage-based analysis, Bollinger b a n d s , and pivot-points. T h e last system, which really is only a demonstration of how to c o m e up with a good set of exit rules, I have
given the n a m e Black Jack, because of its ability to m a k e u s e of small statistical
advantages, w h i c h are the s a m e for all markets, in an attempt to p r o d u c e long-term
profits out of several smaller a n d sometimes just barely profitable trades. It is
especially designed to w o r k well together with m o r e advanced m o n e y m a n a g e m e n t strategies, s u c h as Ralph Vince's Optimal f, or fixed fractional investing. (In
Part 3, we will continue to work with different ways of deriving exits, which we
then apply to all our systems.)

CHAPTER

Data Mining

W h a t will the market do today? E a c h day we ask our fellow trader friends and ourselves the same old question, and each day the market answers us at the end of the
day. One would think that after having asked the question and observed the answer,
say, a couple of thousand times or so, we either would learn the answer or stop wasting time pondering the question. Yet, we keep on asking and each day the market baffles us, by always behaving m o r e or less out of sync with what we had expected.
O n e day the stock market is behaving exactly as anticipated, except for that
little dip after lunch that had us stopped out with a loss. W h i c h , by the way, reminded almost everybody about how the T-bond used to behave a couple of years ago,
w h e n the typical characteristics of the European currency crisis were casting their
shadow over the financial world. Then, the next day, the very same market behaves
in a way never seen before. Except for your b u d d y Joel that is, w h o traded the grain
and meat markets back in the 1970s and therefore recognizes the similarities to how
the corn market used to behave back then.
T h e point I am trying to m a k e here is that it's very difficult to distinguish the
behavior of one market from that of another, and the very s a m e second that we
think w e ' v e nailed it, it all changes and the experience gained loses almost all its
value. Let's face it: can you really say that the stock market today behaved exactly
as the stock market usually does, or w a s it m o r e like the recent coffee market or the
lumber market in the 1980s? And, if it behaved like the coffee market, is the recent
behavior for the coffee market really typical for coffee at this time, and on and on
and on? Is there really a specific, consistent behavior for each and every market,
and, if so, can it be traded profitably? To find out you m u s t e x a m i n e each and every
market very carefully and ask yourself an a b u n d a n c e of questions:
55

56

PART 2

System Concepts

For h o w long does a typical trend last? W h a t are the typical characteristics of
any corrections? W h e n can a m o v e be considered to have g o n e too far? W h a t is
the likelihood for a certain amount of days in a row in the s a m e direction? A n d
m o s t important: h o w can you benefit from this information and i m p l e m e n t it in
your arsenal of existing trading tools? T h o s e are only s o m e of the questions to ask
when you start m i n i n g your data to c o m e up with high-probability trading opportunities; or at least, in trying to avoid the m o s t obvious bad ones.
For example, if the market has been in a down trend for four m o n t h s straight
but you k n o w that a typical d o w n trend should last for only two months, it probably isn't too good an idea to a d d to your position w h e n your breakout system signals that you once again should go short. Granted, the market might have changed
from w h a t is typical for o n e market to what is typical for another, but at least you
still have the long-term statistics on your side. Alternatively, if you already are
short, perhaps it is a good idea to start scaling back, no matter what your system
is telling you. Or, if you k n o w that only 2 2 % of all down moves, measured on daily
data, last for m o r e than two days, you could speculate on the high statistical likelihood for a short up trend to follow, set up a contrarian trading system, and p r o b e
the market with a small position whenever the market has fallen for two days or
m o r e . Similarly, if you k n o w that only 9% of all d o w n moves result in a decline of
8% or m o r e , measured on monthly data, you could take a long position, betting on
an extended up m o v e to follow, as soon as the market has declined that amount.
To get a feel for relationships like these, it is a good idea to put together a set
of tables like Table 5.1 through Table 5.3, which serve as good starting points for
experimentation. All tables have been put together using the R A D contract for the
S & P 500 futures market, with data expanding from January 1985, to D e c e m b e r
1994, and January 1995 to October 1999. (The m o s t up-to-date data, from January
1995 to October 1999, w e r e used for out-of-sample testing, and are placed within
parentheses.) F r o m Table 5.1 you can see that during the period January 1, 1985,
to D e c e m b e r 3 1 , 1994, the average up m o v e for the S & P 500, measured on weekly data, went on for 2.29 weeks, for an average total gain of 3 . 3 % , while the average down move, measured on monthly data, went on for 1.35 m o n t h s with an average total decline of 4 . 8 2 % . F r o m Table 5.2 you can see that measured with daily
data, 2 5 % of all up moves lasted for longer than two weeks, while only 6% of all
down moves could be expected to last for three weeks or m o r e . In Table 5.3 the
monthly amplitudes are within the parentheses in the headers. M e a s u r e d with
weekly data, 4 2 % of all up moves resulted in an increase of the market value of
m o r e than 3 % . M e a s u r e d with monthly data, 4 6 % of all up moves resulted in an
increase of the market value of m o r e than 6%.
To put together Tables 5.1 through 5.3 you m u s t be able to export your data
into a text file, which can be opened with the help of MS Excel or any other
spreadsheet program. O n c e in Excel, you first must calculate the percentage
change of the closing price with the following formula:

CHAPTER 5

Data Mining

TABLE

57

5.1

Data mining summary.

Periods in move

Move per period

Amplitude

Daily data
All moves

0.68%

1.95

1.34%

Up moves

0.69% (0.77%)

2.04 (2.17)

1.41% (1.68%)

Down moves

0.68% (0.71%)

1.86 (1.93)

1.26% (1.36%)

Weekly data
All moves

1.52%

1.97

2.99%

Up moves

1.44% (1.78%)

2.29 (2.21)

3.30% (3.98%)

Down moves

1.64% (1.67%)

1.65 (1.52)

2.70% (2.52%)

Monthly data
All moves

3.32%

1.72

Up moves

3.22% (3.47%)

2.09 (3.31)

6.72% (11.93%)

Down moves

3.56% (3.50%)

1.35 (1.25)

4.82% (4.36%)

TABLE

5.73%

5.2

Periods of moves of certain length.

1

>1

>2

>3

Daily data
All moves

50%

50%

23%

11%

Up moves

4 6 % (43%)

54% (57%)

2 5 % (33%)

13% (15%)

Down moves

5 3 % (53%)

4 7 % (47%)

2 2 % (25%)

10% (12%)

Weekly data
All moves

50%

50%

25%

11%

Up moves

4 1 % (43%)

5 9 % (57%)

3 3 % (28%)

17% (15%)

Down moves

6 0 % (67%)

4 0 % (33%)

17% (13%)

6% (4%)

Monthly data
All moves

57%

43%

16%

Up moves

3 7 % (38%)

6 3 % (61%)

2 3 % (46%)

Down moves

7 6 % (83%)

24% (17%)

9 % (8%)

6%
9 % (46%)
3 % (0%)

PART 2

58

TABLE

System Concepts

5.3

Percentage of moves of certain amplitude.

1(2)%

>1(2)%

>2(4)%

>4(8)%

>3(6)%

>5(10)%

Daily data
All moves

55%

45%

20%

9%

4%

2%

Up moves

5 1 % (42%)

4 9 % (58%)

2 2 % (28%)

1 1 % (18%)

4 % (11%)

2 % (4%)

Down moves

5 9 % (57%)

4 1 % (43%)

18% (25%)

8 % (11%)

4 % (6%)

2 % (3%)

Weekly data
All moves

26%

74%

53%

36%

25%

16%

Up moves

2 3 % (10%)

7 7 % (90%)

6 1 % (75%)

4 2 % (59%)

2 8 % (41%)

18% (24%)

Down moves

2 8 % (28%)

7 2 % (72%)

4 5 % (43%)

3 1 % (27%)

2 2 % (21%)

14% (18%)

Monthly data
All moves

25%

Up moves

2 6 % (0%)

Down moves

75%

2 4 % (25%)

7 6 % (75%)

35%

19%

12%

4 6 % (62%)

2 9 % (46%)

1 7 % (46%)

2 4 % (17%)

9% (8%)

6 % (8%)

45%

7 4 % (100%) 5 7 % (92%)
3 2 % (33%)

=E3/E2
w h e r e E denotes the column where the data is stored.
To calculate h o w m a n y periods in a row a certain move lasts, first use the following formula in the adjacent column:
=IF(OR(AND(F3 > 1 ;F2 < 1); AND(F3 < 1 ;F2 > 1)); 1 ;G2 + 1 )
T h e n , continue the calculation in the next column:
=IF(G3>=G4;SIGN(F3-1)XG3;"")
Finally, in the last column, use the following formula to calculate the percentage move:
M

=IF(H3<> ";PRODUCT(INDEX(F:F;ROW()-ABS(H3)+l;l):
INDEX(F:F;ROW();l))-1;"")
After you have filled in all the calculations, type in the following sets of formulae at the b o t t o m of the spreadsheet to derive the necessary n u m b e r s for the
tables. (For the down moves, simply change " > " t o " < = " . )
To calculate the total n u m b e r of periods u p :
=ABS(SUMIF(H$3:H4429;">0"))
To calculate the total n u m b e r of moves up:
=ABS(COUNTIF(H$3:H4429;">0"))

CHAPTER 5

Data Mining

59

To calculate the average n u m b e r of periods in an up move:

=H4431/H4432
To calculate the average percentage amplitude for an up move:
,,

,,

=SUMIF(I$3:I4429; >0")/COTJNTIF(I$3:I4429;">0 )
To calculate the average percentage amplitude for each period within the up
move:
A

=((H4434+1) (1/H4433)-1)
To calculate the likelihood for an up m o v e to last for two periods or m o r e :
=COUNTIF(H$3 :H4429;" > 1 ")/H4432
To calculate the likelihood for the amplitude of an up m o v e to be greater than
1%:
=COUNTIF(I$3:I4429;">0.01")/H4432
To test a simple trading system that m a k e s use of information like this, you
can, for instance, set up a system that only goes long as soon as you have a down
day that follows a down week that follows a down month. Because of the natural
u p w a r d drift in the stock market and because of w h a t the data mining has shown
about up moves going on for a longer period of time, the requirements for a short
position could be two up days, two up w e e k s , and two up m o n t h s . T h e
TradeStation code for this simple system, G o l d D i g g e r I, looks something like
this:
Condition 1 = CloseM(l) > C and CloseW(l) > C and C[l] > C;
Condition2 = CloseM(2) < CloseM(l) and CloseM(l) < C and
CloseW(2) < CloseW(l) and CloseW(l) < C and C[2] < C[l] and C[l] < C;
If Condition 1 = True and MarketPosition = 0 Then
Buy ("Go long") at open;
If C[2] < C[l] and C[l] < C Then
ExitLong ("Exit long") at close;
If Condition2 = True and MarketPosition = 0 Then
Sell ("Go short") at open;
I f C [ l ] > CThen
ExitShort ("Exit short") at close;
U s i n g the trade-by-trade export function from Part 1, together with the R A D
contract for exporting the results into a spreadsheet program, and then using the
Excel formulae we also derived in Part 1, you can put together a performance s u m m a r y table like those in Tables 5.4 and 5.5.

PART 2

60

TABLE

System Concepts

5.4

Performance summary for Gold Digger I, January 1985-December 1994.

Total trades

251

Winners

Profit factor

1.34

Lrg winner

0.18%

597

Avg winner

2.01%

6,799

C u m profit

Avg profit
St Dev

Losers

15

63.35%

7.76

26,190

Lrg loser

1.09

3,684

Avg loser

47.74

161,123

Drawdown

36.65%

-19.12

-64,530

-1.40

-4,739

-26.71

-90,146

Over the period January 1, 1985, to D e c e m b e r 3 1 , 1994, Gold D i g g e r I p r o duced 251 trades, a m o n g w h i c h close to 6 4 % w e r e profitable, for an average gain
p e r trade of 0 . 1 8 % (or $597 in today's m a r k e t value, with the S & P 500 trading
around 1,350). A fairly low d r a w d o w n and standard deviation also m a k e us want
to continue the research.
W h e n tested on out-of-sample data, the average profit h a d increased to
$ 1 , 4 6 7 , while the largest loss h a d d e c r e a s e d to 4 . 2 0 % (or $14.175 in today's m a r ket value). An ever-low standard deviation a n d a h i g h p e r c e n t a g e profitable trades
k e e p us interested. No m o n e y w a s deducted for slippage a n d c o m m i s s i o n , but
w i t h an expected $75 in slippage and c o m m i s s i o n , you can expect this system to
generate approximately $1,392 p e r contract traded ($1,467 $ 7 5 ) in the i m m e diate future. Table 5.5 also shows that t h e d r a w d o w n decreased substantially from
close to 2 7 % , for the in-sample period, to 7% for the out-of-sample period. For
d r a w d o w n a n d cumulative profit values to be correct, however, we are a s s u m i n g
that the entire equity, including previously m a d e profits, w a s reinvested at each
new trade. As m e n t i o n e d in Part 1, this is s e l d o m possible, especially not in the
futures m a r k e t s , but it could be so in the stock market, provided that you can b u y
fractions of a share. W h a t this does allow you to do, however, is to c o m p a r e different systems a n d m a r k e t s with each other on an equal basis, or c o m p a r e h o w
your system would have held up against a b u y - a n d - h o l d strategy over the s a m e
t i m e period.
Despite the not too b a d results, it is important to r e m e m b e r that Gold Digger
I is put together merely to illustrate a simple system for taking advantage of a market's statistical characteristics. This particular system is n o t a good system to trade
if you compare the statistical characteristics in Tables 5.1 to 5.3 for the m a r k e t dur-

TA B L E

5.5

Performance summary for Gold Digger I, January 1995-October 1999.

Winners

Total trades

105

Profit factor

2.27

Lrg winner

Avg profit

0.43%

1,467

Avg winner

St Dev

1.40%

4,719

Cum profit

66.67%

33.33%

3.45

11,644

Losers
Lrg loser

-4.20

-14,175

1.16

3,928

Avg loser

-1.02

-3,456

56.10

189,338

Drawdown

-7.20

-24,300

APTER 5

61

Data Mining

ing the first 10-year period, with the characteristics for the latest four years.
C o m p a r i n g the in- a n d out-of-sample periods with each other shows that ever-sosmall differences between the daily and weekly data soon add up to quite large differences in the monthly time perspective. For instance, when looking at the m o n t h ly data, you notice the average m a g n i t u d e of an up move, measured on monthly
data, has grown from 2.09 m o n t h s and 6 . 7 2 % to 3.31 m o n t h s and 1 1 . 9 3 % . Also,
in the table describing the likelihood for periods of moves of certain length, for the
period up until 1994 there was only a 9% chance for an up move to last for m o r e
than three m o n t h s , but for the period spanning January 1995 to October 1999,
there was a 4 6 % chance for the s a m e type of move.
In the s a m e table, the m o s t recent study suggests that 5 7 % of up moves m e a s ured on weekly data go on for more than one week. This is two percentage points
less than what could be expected looking at the period ending in 1994. Hence,
while the n u m b e r of persistent up trends measured on weekly data has decreased,
the n u m b e r of persistent up trends measured on monthly data has skyrocketed. T h e
s a m e p h e n o m e n o n also can be seen in the table for percentage of moves of certain
amplitude, w h e r e the percentage of up moves with an amplitude of m o r e than 5 % ,
measured on weekly data, has only increased by 3 3 % (from 1 8 % to 2 4 % ) , while
the n u m b e r of up moves with an amplitude of m o r e than 10%, measured on
monthly data, has skyrocketed by 1 7 0 % (from 1 7 % to 4 6 % ) .
This suggests that when the market changes its characteristics or m o d e , these
changes can be very hard to detect w h e n looking at a shorter time frame within the
bigger picture. Or, put differently: no matter what long-term m o d e the market is
in, the short-term statistical characteristics are likely to still look the s a m e and be
close to stationary in nature. This is a very important conclusion, because if this is
true, the only way to build a mechanical trading system that holds up and behaves
the s a m e way in the future as it does during testingno matter what the longert e r m underlying trend looks likeis to focus on the shorter time perspective, with
trades lasting for no longer than approximately a week and using as little historical data for the signals as possible.
To test if this could be true, a modified version of the original system considers only daily and weekly data, as suggested by the following TradeStation
code, which also does not differentiate between up trends and down trends:
Conditionl = CloseW(2) > CloseW(l) and CloseW(l) > C and C[2] > C[l] and
C[l] > C;
C o n d i t i o n = CloseW(2) < CloseW(l) and CloseW(l) < C and C[2] < C [ l ] and
C[l] < C;
If Conditionl =True and MarketPosition = 0 Then
Buy ("Go long") at open;
IfC[2] < C[l] and C[l] < C T h e n
ExitLong ("Exit long") at close;

62

PART 2

System Concepts

If Condition2 = True and MarketPosition = 0 Then

Sell ("Go short") at open;
IfC[l] > CThen
ExitShort ("Exit short") at close;
T h e results for this version of the system can be seen in Table 5.6. Tested on
the out-of-sample period this system h a d 107 trades, with 6 3 % profitable trades
and an average profit of 0 . 3 1 % (or $1,045 in today's market value). No m o n e y was
deducted for slippage and commission, but this is easily accounted for, for trades
in the immediate future, by deducting an appropriate a m o u n t from the expected
value of the average trade. B e c a u s e Gold D i g g e r II is a simpler system (less curvefitted) than G o l d D i g g e r I, its performance is not quite as good, as j u d g e d by a
slightly lower profit factor and percentage profitable trades and a m u c h lower
cumulative profit. It does, however, s e e m to be a little m o r e robust, as j u d g e d by
a lower standard deviation. A l t h o u g h the average profit and profit factor are slightly lower for Gold D i g g e r II w h e n c o m p a r e d to Gold D i g g e r I, the lower standard
deviation, in combination with Gold D i g g e r IPs b e i n g m o r e symmetrical in nature
(less curve-fitted), m a k e s it clear that Gold Digger II is the preferable system.
F r o m the TradeStation code, however, you can see that the only exit criteria
we have on the long side is two up closes in a row, while we will exit a short trade
as soon as we have a down day. B e c a u s e the market does not always behave as we
want it to, the losses can be quite severe before we are allowed to exit the trade.
This can, for instance, happen if the m a r k e t immediately takes off in the w r o n g
direction, or if in a long position, every other day is an up day and every other day
is a d o w n day, but with d o w n days of greater m a g n i t u d e than up days. Therefore,
the lack of properly developed stops still m a k e s this a dangerous system, as can be
seen from the values for the largest a n d average losing trades. (We will talk m o r e
about stops a n d exits in Part 3.)

BETTER USE OF YOUR DATA

If you are better off using as few bars as possible in your system, it should follow
that you should m a k e as g o o d use of those bars as possible. For instance, a m o v ing average usually only looks at one particular price within a t i m e series, not takTABLE

5.6

Performance summary for Gold Digger II, January 1995-October 1999.

Total trades

107

Profit factor

1.98

Winners
Lrg winner

Avg profit

0.31%

1,045

Avg winner

St Dev

1.38%

4,666

Cum profit

62.62%

7.31

24,671

1.00

3,364

37.84

127,710

Losers
Lrg loser

37.38%

-4.06

-13,703

Avg loser

-0.84

-2,840

Drawdown

-7.26

-24,503

CHAPTER 5

Data Mining

63

ing into account that each b a r actually holds three other important price levels.
That is, w h e n calculating a moving average of closing prices, all open, high, and
low prices usually are ignored. Using regular m o v i n g averages a n d Bollingerb a n d - t y p e indicators to identify trading opportunities also present a problem in
that the readings are not always comparable over time. For e x a m p l e , a value of
1,350 for the S & P 500 index m a y very well be below its moving average and standard deviation bands today, but m a y not be so a couple of years from now, if and
w h e n the market trades at the same level again. It all depends on where the market is c o m i n g from at that point in time.
Wouldn't it be great to c o m e up with an indicator that can m a k e use of all
four of these price extremes, shorten the lookback period by 7 5 % , a n d at the s a m e
time m a k e it possible to m a k e all of the indicator observations comparable with
each other, regardless of w h e n they happened or at w h a t level the market w a s trading at the time?
Contrast the above with a statistical examination, where you are measuring the
height of a random sample of m e n that you meet throughout the day. At the end of the
day you write down the height of the first man, the tallest man, the shortest man, and
the last man you have measured that day. It does not matter in what order you write
t h e m down. After five days you have 20 observations, which is the m i n i m u m amount
required to make any statistical conclusions. (This number varies; some say 20, others say 30, while still others say 100 or more. In the medical field, it is close to 1,000.
Aside from the fact that you will not always have that m u c h data to experiment with,
there is no reason why you shouldn't treat your m o n e y any less seriously.)
Let us say that the average for all 20 m e n was 6 feet and the standard deviation was 3 inches. T h e one standard deviation boundaries and the two standard
deviation boundaries hold approximately 6 8 % and 9 5 % , respectively, of the m e n
m e a s u r e d and the expected height of all m e n to be measured in the future. With
this information at hand y o u now can classify all m e n that you met as, for instance,
average height (between 5 feet 9 inches to 6 feet 3 inches), short (between 5 feet
6 inches to 5 feet 9 inches), very short (below 5 feet 6 inches), tall (between 6 feet
3 inches to 6 feet 6 inches), and very tall (above 6 feet 6 inches). For each and
every investigation d o n e this way, and depending on w h e r e they are done, these
levels will vary slightly. B u t for each investigation, at least s o m e of the previous
information will be useful in estimating the future outcome or what can be expected for the next day. For instance, because the chances of meeting a tall or very tall
m a n are approximately 1 6 % ((1 0.68) X 100 / 2), you also know that the c h a n c e
of meeting two such m e n in a row is a m e r e 2 . 5 % (0.16 X 0.16 X 100).
To develop and use the same type of research in the markets, it is important
to note not at what level the markets are trading, but the amplitude of the moves. To
m a k e each move comparable to all others, both in time and between different markets, we m u s t use the R A D contract, which keeps the percentage changes constant
instead of using the actual point or dollar values. To calculate the move rather than

PART 2

64

System Concepts

the level, it also is necessary to relate each move to a base or ground level.
Continuing the comparison with measuring m e n on the street, it m a k e s little or no
sense to measure each man's height from sea level instead of from the ground where
he is standing. That is, for a comparison to be meaningful, it is necessary to normalize the data; the obvious level from which to normalize a person's height is the
ground on which he stands.
In m a r k e t research, there are a few levels to which e a c h m o v e can be normalized, like the previous bar's close or the average price of the previous bar. H e r e
each move has been normalized to the previous bar's closing price. A n o t h e r difference between m e a s u r i n g m e n and market action is the use of absolute and percentage values. If you are 6 feet tall, you will be 6 feet tall whether you are standing on a pair of skis in Vail or lying down on a beach towel in Hawaii. It would
m a k e little or no sense to c o m p a r e your height in percentage terms to h o w high
you are from the ocean. B u t in the market there is a huge difference between a
move worth $20,000 if the market is trading at the 1,350 level, compared to the
s a m e dollar move if the market is trading at 2 5 0 . In the first case, a $20,000 move
equals only a 6% move, while in the latter the s a m e dollar move a m o u n t s to over
3 0 % . T h u s , it is of p a r a m o u n t importance to m e a s u r e market moves in percentages
rather than point or dollar values so that you can c o m p a r e different types of moves,
no matter where and w h e n they happened.
In our m e a s u r i n g - m e n example, four observations were kept each day. This
can be done for the markets as well with what I call the M e a n d e r indicator. The
following TradeStation c o d e sets up an array containing all the opening moves,
moves to the highs, moves to the lows, and moves to the close, for a total of 20
data points for the last five bars:
Input: VSStd(l);
Vars: SumVS(O), AvgVS(O), DiffVS(O), StdVS(O), SetArr(O), SumArr(O),
DiffArr(O), VSLow(O), VSMid(O), VSHigh(O), FNameC*"), TradeStrl("");
Array: VS[20](0);
For SetArr = 0 To 4 Begin
VS[SetArr * 4 + 0] = (0[SetArr] - C[SetArr + 1]) / C[SetArr + 1];
VS[SetArr * 4 + 1] = (H[SetArr] - C[SetArr + 1]) / C[SetArr + 1];
VS[SetArr * 4 + 2] = (LfSetArr] - C[SetArr + 1]) / C[SetArr + 1];
VS[SetArr * 4 + 3] = (C[SetArr] - C[SetArr + 1]) / C[SetArr + I];
End;
For SumArr = 0 To 19 Begin
If SumArr = 0 Then
SumVS = 0;
SumVS = SumVS + VS[SumArr];
If SumArr = 19 Then

CHAPTER 5

65

Data Mininj

AvgVS = SumVS / 20;

For DiffArr = 0 To 19 Begin
If DiffArr = OThen
DiffVS = 0;
DiffVS = DiffVS + Square(VS[DiffArr] - AvgVS);
If DiffArr = 19 Then
StdVS = SquareRoot(DiffVS / 20);
End;
End;
VSLow = C * (1 + (AvgVS - StdVS * VSStd));
VSMid = C * (1 + AvgVS);
VSHigh = C * (1 + (AvgVS + StdVS * VSStd));
Plotl(VSLow, "VS Low");
Plot2(VSMid, "VS Mid");
Plot3(VSHigh, "VS High");
If CurrentBar = 1 Then Begin
FName = "C:\Temp\" + LeftStr(GetSymbolName, 2) + ".csv";
FileDelete(FName);
TradeStrl = "Date" + "," + "Open" + "," + "High" + "," + "Low" + ","
+ "Close" + "," + "VS Low" + "," + "VS Mid" + "," + "VS High" +
NewLine;
FileAppend(FName, TradeStrl);
End;
If CurrentBar > 5 then Begin
TradeStrl = NumToStr(Date, 0) + "," + NumToStr(Open, 2) + "," +
NumToStr(High, 2) + "," + NumToStr(Low, 2) + "," +
NumToStr(Close, 2) + "," + NumToStr(VSLow[l], 2) + "," +
NumToStr(VSMid[l], 2) + "," + NumToStr(VSHigh[l], 2) + NewLine;
FileAppend(FName, TradeStrl);
End;
All moves are measured with the previous bar's close as a base. O n c e the data
are collected, the c o d e calculates the average move and the standard deviation
before the data are denormalized to fit with the latest price action. W h e n the calculation is completed, you have an indicator that m a k e s the m o s t out of each bar's
price action and allows you to trade intraday using only end-of-day data or
intraweek, using a m i x of weekly a n d daily data. T h e last part of the code holds the
n e c e s s a r y instructions for exporting all the data into a text file for further analysis

PART 2

66

System Concepts

in a spreadsheet p r o g r a m , such as Excel. Figure 5.1 shows w h a t this pivotp o i n t - t y p e indicator looks like w h e n charted together with the latest price action
for the S & P 500 futures contract. As you can see, the Meander indicator consists
of three lines, a m o n g which the upper (VS High) and lower ( V S L o w ) lines can be
chartered one or two standard deviations away from the middle (VS M i d ) line. T h e
M e a n d e r indicator works exactly the s a m e for all markets.
With the necessary data imported into Excel, you can type in the following
formula in adjacent c o l u m n s to calculate the long entry level, the long risk level
(stop loss), the long exit level, and the result of the trade. (Reverse the calculations
in the next four c o l u m n s for the short side.)
= IF(B2<F2;B2;IF(D2<F2;F2;"")),
C o l u m n B denotes the opening price, c o l u m n F denotes the VS L o w level,
and c o l u m n D the Low.
=IF(I2<>"";I2*(1-H$1212/100);"")
C o l u m n I denotes the long entry level a n d cell HI212 refers to the cell where
you have typed in your percentage risk.
=IF(I2<>"**;IF(D2<J2;J2;IF(AND(C2>H2;E2<H2);H2;E2));"")
C o l u m n J denotes the long risk level, column C denotes the High, column H
the VS High level, and column E the closing price.
n

= IF(I2<>"'*;(K2 - I2)/I2; - ")

C o l u m n K denotes the long exit level.
At the bottom of the spreadsheet type in the following formulas to calculate
the total n u m b e r of trades, percent winning trades, average percentage profit per
trade and average dollar profit per trade. (Again, reverse the calculations for the
short side.)
=COUNT1F(L2:L1208;"<>-")
C o l u m n L denotes the percentage result for each trade.
=COUNTIF(L2:L1208;">0")/L1210
=SUMIF(L2:L1207;"<>-")/L1210
=L1212*1350*250
T h e percentage risk that you are willing to take is, in this case, typed into cell
H1212.
This simple system puts on a long (short) trade immediately at the open if
the opening price is below (above) the VS Low ( V S High) level, or at the VS Low

CHAPTER 5

FIGURE

Data Mining

5.1

The Meander indicator makes the most out of your data.

67

PART 2 System Concepts

68

(VS High) level if the market trades below (above) that level. It exits with a loss if
the market moves against the position with a certain percentage, or with a profit if
it trades above (below) the VS H i g h ( V S L o w ) level but then moves b a c k down
again, as suggested by the closing price. All trades are also closed out at the end
of the day. B e c a u s e this is an intraday system based on daily data, and sometimes
we do not k n o w what came first, the high or the low, we m u s t play it safe and look
for the losing trades first. Table 5.7 shows the results for this system broken down
into long and short trades, tested on the S & P 500 futures contract, for the period
January 1995 to October 1999. No m o n e y has been deducted for slippage and
c o m m i s s i o n . The percentage risked at each trade w a s 0 . 7 5 % . In this version, the
M e a n d e r had over 5 0 % profitable trades on the long side, for a total average profit in today's m a r k e t value of $287. With an estimated $75 deducted for slippage
and commission, the average profit for the long side will be $ 2 1 2 . N o t too b a d for
an intraday trading system on a one-contract basis. R e m e m b e r also that because
we played it safe by always considering a trade to be a loser if it reached the stop
loss level, although we very well might have reached the profit target first, there
is a high likelihood for the percentage profitable trades and the average profit to
be even higher than shown.
In this case, the standard deviation was set to 1, w h i c h m e a n s that the system
triggers a trade fairly frequently. With the standard deviation set to 2, the system
generates fewer trades, but with a higher probability for success. A l s o , the highest
probability trades are those that enter immediately at the open. This is because if
the open is below (above) the trigger level, the low (high) will be, too. That is,
there will be two low probability occurrences in a row, analogous to the chance of
meeting two tall m e n in a row, which increases the probability for a move in the
opposite direction. Or rather, if we continue on the tall m e n analogy, it increases
the likelihood that we have run out of tall m e n .
Now, let us take a look at the very s a m e system, with the very same settings,
but this time traded on corn. As you can see in Table 5.8, the system continues to
work fairly well if we only look at the percentage average profit and the n u m b e r
of profitable trades.

TABLE

5.7

Basic results for Meander I trading system on the S&P 500.

Short trades

Long trades
Number of trades
Percent winners

522
52.30%

Number of trades
Percent winners

459
43.79%

Average profit (%)

0.09%

Average profit (%)

0.03%

Average profit ($)

287.16

Average profit ($)

112.81

CHAPTER 5

69

Data Mining

TABLE

5.8

Basic results for Meander I trading system on com.

Long trades
Number of trades
Percent winners
Average profit (%)
Average profit ($)

Short trades
439
48.97%
0.07%
6.88

Number of trades
Percent winners
Average profit (%)
Average profit ($)

476
44.75%
0.07%
7.02

But when it comes to the dollar profits, the results look less promising. This
is because there is a huge difference between a good working system and a profitable system. This is an interesting observation, which adds to the never-ending
debate on whether or not to trade the s a m e system on several different markets. T h e
conclusion from this simple test is to test the same percentage-based system on several different markets to m a k e sure that it works equally well on all markets, but
only to trade those markets in which the percentage-based moves also equal a high
enough dollar profit. Or, in continuing our analogy to measuring men, if you were
to sell pants to as m a n y of the m e n you have measured as possible, choosing your
sizes and markets would be beneficial in selling as many pairs as possible.
To put it in trading terms: just because a system is equally good at capturing
all 2% moves in corn as it is in S & P 500, this doesn't m e a n that it will be equally
as profitable in both markets. For a system to be profitable, it needs to be traded
in a market w h e r e the moves it is designed to catch also are worth taking. Because
a system is working well on all markets, it does not m e a n it will be tradable on all
markets, because not every system will produce big enough dollar profits to overc o m e the costs. In the case of corn, the M e a n d e r does not even c o m e close. But
r e m e m b e r that this does not m e a n that M e a n d e r is a b a d system. W h e t h e r a good
working system will be profitable or not has nothing to do with the system, but
rather at what level the market currently is trading and the dollar value of that
level. And, as previously noted, this is a technicality decided by the exchange,
w h i c h has nothing to do with whether the system is good at capturing the moves
of the market.

BASIC EXIT TECHNIQUES

T h e Black Jack Trading System derives its n a m e from the popular Las Vegas card
g a m e of the s a m e n a m e . Black Jack is probably the only betting g a m e where you
can m a k e sure that the long-term odds work in your favor and therefore allow you
to beat the house in a consistent manner. W h e n you are playing Black Jack, it is
essential to know w h e n to hold, when to ask to be hit with yet another card, and
w h e n to bet m o r e or less aggressively (if at all). To accomplish this you must know

70

PART 2

System Concepts

the current composition of the deck and how to analyze your odds correctly as
compared to the dealer's. T h e basic rules are to bet m o r e aggressively, but hold at
a lower n u m b e r if there are m o r e high cards left in the deck, especially if the dealer's card is high as well. If you know how to do this, in the long run it will not matter what cards you are being dealt, you will end up being a winner anyway.
In the trading system Black Jack, all this is comparable to being able to trade
profitably, no matter what the entry rule looks like, as long as you only take trades
in the direction of the underlying trend, and know exactly when to exit and why. You
are trying to keep your average trade (hand) in positive territory with just enough
money to make it worthwhile. If you can accomplish this, the rest is just a matter of
trading as frequently (playing as many hands) as possible in as short a time as possible. Therefore, it is not only essential that your entry rule allows you to take as
many trades as possible, but even more important that your trades don't last too long.
W h e n exiting a trade, you can do so under four different assumptions. In the
best of worlds, exit technique n u m b e r one is the profit target, which lets you exit
the market with a limit order at the m a x i m u m open profit. Exit technique n u m b e r
two is the trailing stop, which c o m e s into play when the market is starting to move
against you after you have accumulated an open profit. N u m b e r three is the stop
loss for those occurrences when the trade goes against you right off the bat.
Finally, you also can exit with a time-based stop, possibly in combination with any
of exit techniques one to three.
W h e n it c o m e s to the entries and trade triggers, there are probably as m a n y
ways to go about designing them as there are traders (and then some), so instead
of delving any deeper into this matter, and because of the reasoning above, this
version of Black Jack was put together with a very simple top-and-bottom-picking
entry technique paired with a short-term breakout system in case the retracement
w a s n ' t deep e n o u g h to trigger an entry. In Part 3, where we focus m o r e explicitly
on efficient exits and overall trade efficiency, we will substitute these entries with
random entries. That said, for a set of good short-term entry ideas, look into the
work of Linda Bradford Raschke or T o m D e M a r k .
Finally, there is the trend filter. In the academic c o m m u n i t y and a m o n g fundamental analysts w h o pay h o m a g e to the efficient market hypothesis, for the
longest time there was no such thing as a trend. However, in recent years even
these people have had to cave in and admit that there is something that they call
"a r a n d o m walk with a drift." For decades, this drift has been measured by technical analysts with the help of moving averages of various lengths. Probably the
m o s t simple trend filter, w h i c h also has withstood the test of time, is the 200-days
moving average. To continue to keep it as simple as possible, this is also what we
use for Black Jack. If you would like to look into m o r e sophisticated long-term
techniques, d o n ' t overlook the work of N e l s o n Freeburg.
Before we can put together the system, we also need to understand what it is
we are trying to achieve, and consequently, how to go about doing it. Picture your-

CHAPTER 5

Data Mining

71

self as a quality controller standing at the end of an assembly line. Your pay (final
estimated profit) for this j o b is dependent on how m a n y units (number of trades) you
and your machine (system) can produce, but also on how little you can m a k e the
look of all these units deviate from each other or from the average look (a low standard deviation). Obviously, you would like the assembly line to run as fast as possible, but only after you have examined one unit can the next unit be placed on the line,
so the longer the time spent examining a single unit, the lower your payoff will be.
A l o n g c o m e s an obviously defective unit that is bound to lower your results.
Without wasting any time, you toss it b e h i n d your back right away, freeing up the
line for another unit to be placed upon it. T h e next unit looks pretty average, but
to be sure, you e x a m i n e it for a while longer than you did the previous one. In fact,
without k n o w i n g it, with this average unit you spent an average a m o u n t of time.
But the third unit that c o m e s along is m u c h trickier. This o n e shows all the
right features for being a real m o n e y maker, and while y o u ' r e watching it, these
features grow even bigger, p r o m i s i n g to m a k e this unit at least three times as valuable as the average unit. W h a t to do? Should you let it grow even bigger or should
you free up the line for m o r e units? But then, all of a sudden its value-added features start to diminish. You freeze in panic, and not until its value has sunk b a c k to
twice the value of the average unit do you take it off the line. T h e problem is that
in the time you spent processing this unit you could have processed three average
units. W h a t w a s it that h a p p e n e d here? By trying to increase your profits from an
individual unit you put yourself in unfamiliar territory, w h i c h m a d e you panic,
increased the standard deviation of the o u t c o m e , and lowered the n u m b e r of units
produced, which in the end resulted in less overall profits.
But wait a minute, you say. W h a t about the big winners? D o e s n ' t the old
adage say that you should let your profits run? Yes, it does, and it is probably good
advice to listen to every now and then, but in trading it also has its price. T h e price
is that you lose track of what is average. T h e really big winners are usually few and
far between; the rest are just a b u n c h of look-alikes there to confuse you. By staying away from all of them, you are freeing both your m i n d and capital to milk that
big w i n n e r several times over in a sequence of average trades, while at the s a m e
time allowing yourself to step in and out of other m a r k e t s as well.
Translated into trading t e r m s , for a trading system to be both as robust and
as profitable as possible you should strive to m a k e each trade as similar to all other
trades as possible, even if it m e a n s that you have to cut your profits short every
now and then. By doing this, you are freeing up time and m o n e y that can be spent
on other average tradestrades with which you are familiar and k n o w how to handle without panicking, which in the end should prove itself m o r e profitable.
N o w w h a t exactly does this m e a n ? It m e a n s that to c o m e up with a reliable
system with a positive average trade and a standard deviation that is as low as p o s sible, and with as m a n y profitable trades as possible, it is of p a r a m o u n t importance
that the system work, on average, equally well on several different markets over

PART 2

72

System Concepts

several different time periods. This holds true even if you only are going to trade
it on o n e market, because you never k n o w w h e n the characteristics of this market
will change to resemble m o r e closely those of another market. Therefore, if and
w h e n this happens, you must have taken this into consideration beforehand by
m a k i n g sure that the system (at the least) will not go berserk in this new market
environment.
Hence, if the underlying assumption isn't market specific, always m a k e sure
that your system runs at least marginally profitably, if ever so slightly, in percentage terms, on as m a n y markets as possible. If you wish, this could be comparable
to m a k i n g sure that your Black Jack card playing skills stay independent from any
external factors, such as how m a n y other people are playing, h o w the deck is shuffled, w h o the dealer is, or the n a m e of the hotel.
For this system we use data covering the period January 1985 to October
1999, from 14 different markets. T h e s e are: coffee, copper, corn, cotton, C R B index, D-mark, Eurodollar, Japanese yen, crude oil, live cattle, natural gas, orange
j u i c e , T-bond, a n d the T-bill.
We know that we should look for as m a n y profitable trades as possible to
generate an as high as possible average profitable trade in percentage terms, but
with am as low as possible standard deviation. We can modify the code from Part
1 before we attach it to the b o t t o m of the system, so that we can export the right
data into a text file for further analysis in Excel. The TradeStation code for the
Black Jack I trading system follows:
Inputs: BarNo(O), SL(0), PT(0), MP(0);
Vars: Trigger(O), BSLevel(O), TotTr(O), ExpVar(O), Prof(0), TotProf(0),
TradeStrl ("");
Filters *****}
Trigger = Average(C, 200);
I ***** Retracement Entry *****}
If C > Trigger and MarketPosition = 0 and (L < XAverage(L, 6) or
L Crosses above XAverage(L, 6)) and C Crosses above XAverage(C, 6) and
H < XAverage(H, 6) Then Begin
Buy ("Buy Support") at C;
BSLevel = C;
End;
If C < Trigger and MarketPosition = 0 and (H > XAverage(H, 6) or
H Crosses below XAverage(H, 6)) and C Crosses below XAverage(C, 6) and
L > XAverage(L, 6) Then Begin
Sell ("Sell Resist.") at C;
BSLevel = C;
End;

CHAPTER 5

73

Data Mining

{***** Breakout Entry *****}

If C > Trigger and MarketPosition = 0 and H Crosses above XAverage(H, 6) and
C Crosses above XAverage(C, 6) Then Begin
Buy ("Buy Break") at C;
BSLevel = C;
End;
If C < Trigger and MarketPosition = 0 and L Crosses below XAverage(L, 6) and
C Crosses below XAverage(C, 6) Then Begin
Sell ("Sell Break") at C;
BSLevel = C;
End;
{***** Exit techniques *****}
If BarsSinceEntry >= BarNo and BarNo <> 0 Then Begin
ExitLong ("Long time") at C;
ExitShort ("Short time") at C;
End;
If SL <> OThen Begin
ExitLong ("Long loss") tomorrow at BSLevel * (1 + SL*0.01) Stop;
ExitShort ("Short loss") tomorrow at BSLevel * (1 - SL*0.01) Stop;
End;
IfPT < > OThen Begin
ExitLong ("Long profit") tomorrow at BSLevel * (1 + PT*0.01) Limit;
ExitShort ("Short profit") tomorrow at BSLevel * (1 - PT*0.01) Limit;
End;
IfTS < > OThen Begin
If C > BSLevel * (1 + MP*0.01)Then
ExitLong ("Long stop") tomorrow at BSLevel * (1 + MP*0.01) Stop;
If C < BSLevel * (1 - MP*0.01)Then
ExitShort ("Short stop") tomorrow at BSLevel * (1 - MP*0.01) Stop;
End;
As you can see from the code, so far all input values are denoted by zeros.
This is because we would like to activate t h e m one at a time. T h e first thing we
must do is find the optimal length of our trades. Therefore, we start by adding the
following piece of code to the bottom of the system:
TotTr = TotalTrades;
If TotTr > TotTr[l] Then Begin
Prof = PositionProfit(l) / (EntryPrice(l) * BigPointValue);

74

PART 2

System Concepts

TotProf = TotProf + Prof;

End;
If LastBarOnChart Then Begin
ExpVar = BarNo;
TotProf = TotProf / TotalTrades;
TradeStrl = LeftStr(GetSymbolName, 2) + "," + NumToStr(ExpVar, 2) +
"," + NumToStr(TotProf * 100, 2) + "," + NumToStr(Percentprofit, 2) +
NewLine;
FileAppend("D:\Temp\B J.cvs", TradeStrl);
End;
O n c e this is done, we can r u n an optimization process in TradeStation. For
e a c h t i m e TradeStation runs a test, it also exports the data we need into a text file
for further analysis in Excel. N e v e r m i n d the optimization report in TradeStation.
In Excel, start by sorting and separating the data by n u m b e r s of days in trade, then
type in the following formulae at the b o t t o m of each group.
To calculate the average percentage profit for a specific trade length for all
markets:
=AVERAGE(C1:C14)
w h e r e column C denotes the average profit for a specific market a n d trade length.
To calculate the standard deviation of the percentage profit for a specific
trade length for all markets:
=STDEV(C1:C14)
To calculate the ratio between the average profit and standard deviation of
outcomes:
=E14/E15
w h e r e cell El 4 denotes the average profit and cell El5 denotes the standard deviation.
To calculate the percentage profitable trades for a specific trade length for all
markets:
=AVERAGE(D1:D14)
w h e r e c o l u m n D denotes the percentage profitable trades for a specific market and
trade length.
To calculate the standard deviation of the percentage profitable trades for a
specific trade length for all markets:
= STDEV(D1:D14)

CHAPTER 5

75

Data Mining

To calculate the ratio between the percentage profitable trades and its standard deviation:
=F147F15
w h e r e cell F14 denotes the percentage profitable trades a n d cell F15 denotes the
standard deviation.
S u m m e d up in a table for easier interpretation, the results resemble those in
Table 5.9, w h i c h shows the data for the time-based stop, which is triggered at a certain n u m b e r of days, no matter what. In this case, we are looking for a high average
profit and a high percentage of profitable trades. T h e higher the ratio between the
critical variable and its standard deviation, the m o r e robust the system.
F r o m Table 5.9 we can see that the ratio between the average profit and its
standard deviation is at its highest with a trade length of 14 days. A trade length
of ten to 11 days, however, also s e e m s to have a decent risk/reward relationship,
especially considering that the ratio between the percentage profitable trades and
its standard deviation also is slightly higher for these trade lengths than for the 14day alternative. If we were to set a m a x i m u m trade length of either 10 or 11 days,
however, the actual average trade length is slightly shorter w h e n the other stops are
added. Therefore, we will change the time-based stop slightly and turn it into a
very tight trailing stop that will be triggered if the trade lasts for 11 days or m o r e .

TABLE

5.9

Summary tradelength statistics for Black Jack I trading system.

Days in trade
2

Average profit

0.05

0.04

0.01

0.04

0.00

0.03

0.05

Standard deviation

0.10

0.13

0.16

0.18

0.21

0.26

0.25

Ratio

0.51

0.32

0.08

0.22

-0.01

0.13

0.21

52.99

51.64

52.30

53.30

53.12

52.91

52.82

Percent profitable
Standard deviation

4.32

4.08

3.91

3.70

4.28

4.37

4.23

Ratio

12.27

12.64

13.38

14.39

12.40

12.12

12.49

10

11

12

13

14

15

Average profit

0.08

0.20

0.17

0.10

0.11

0.14

0.15

Standard deviation

0.38

0.36

0.35

0.35

0.30

0.24

0.32

Ratio

0.21

0.56

0.50

0.29

0.36

0.58

0.45

Percent profitable

54.04

53.96

54.12

53.61

53.32

53.21

53.54

Standard deviation

3.89

4.79

4.78

5.01

5.96

5.15

5.05

Ratio

13.91

11.28

11.31

10.70

8.94

10.34

10.60

PART 2

76

System Concepts

If the trade runs on for too long, our profit target eventually stops it out instead.
T h e TradeStation code for the time-based trailing stop is :
If BarsSinceEntry > = BarNo Then Begin
ExitLong ("Long time") tomorrow at C Stop;
ExitShort ("Short time") tomorrow at C Stop;
End;
With a time-based trailing stop that kicks in after 11 days in place, it is time
to set a profit target. In the TradeStation code for the export function simply
change the line "ExpVar = B a r N o " to "ExpVar = P T " . R e m e m b e r that the profit
target is only there to catch the m o s t extreme trades, but because the entire idea
behind Black Jack is to keep all trades as uniform as possible, we d o n ' t want our
profits to run for too long either. In this case, we aim for the 5 to 1 0 % region. Table
5.10 shows what our results look like.
From Table 5.10 we can see that the ratio b e t w e e n the average profit and its
standard deviation is at its highest with a profit target of 9 . 5 % . This value also
seems to be surrounded by other fairly high values, which indicates that this is a
robust level, likely to produce good results in the future, even though the optimal
level might m o v e around slightly. Therefore, we decide that we will add a profit
target of 9 . 5 % to the system.

TABLE

5.10

Summary profit target statistics for Black Jack I trading system

Percentage profit target
5

5.5

6.5

7.5

Average profit

0.10

0.09

0.10

0.08

0.10

0.13

Standard deviation

0.48

0.45

0.45

0.49

0.47

0.46

0.23

0.17

0.21

0.27

55.10

54.41

54.21

54.14

Ratio

0.20

0.21

Percent profitable

55.91

55.41

Standard deviation

4.06

3.97

4.07

4.32

4.55

4.63

Ratio

13.77

13.97

13.54

12.59

11.92

11.69

8.5

9.5

10

10.5

Average profit

0.13

0.13

0.13

0.15

0.13

0.13

Standard deviation

0.43

0.43

0.45

0.43

0.41

0.42

Ratio

0.30

0.31

0.29

0.36

0.33

0.32

54.05

53.90

53.72

53.64

53.55

53.47

Percent profitable
Standard deviation

4.69

4.71

4.90

4.99

4.91

4.99

Ratio

11.52

11.44

10.96

10.75

10.91

10.72

CHAPTER 5

Data Mining

77

With a time-based trailing stop and a profit target in place, the s a m e m a n e u vers are repeated for the m i n i m u m profit stop a n d the stop loss. T h e m i n i m u m
profit stop we try to place somewhere between zero profits and halfway up to the
profit target, which m e a n s that we must look in the 0 to 5% region. Table 5.11
shows the result for these levels.
From Table 5.11 we can see that the higher the m i n i m u m accepted profit, the
lower the average trade, but also that the uncertainty, as m e a s u r e d by the ratio
between the average profit a n d its standard deviation, is increasing. To be sure, it
s e e m s as if we should go with a fairly low accepted m i n i m u m profit. In doing so
we also can expect a high percentage of profitable trades even though the insecurity about this n u m b e r decreases with a lower m i n i m u m accepted profit. For this
system, we decide to aim for a m i n i m u m accepted profit of 1%.
T h e stop loss, finally, we would like to keep as tight as possible, but at the
s a m e time m a k e sure that we give the trade s o m e initial slack. Let us see what we
can find in the 0.5 to 2.5 percent region. Table 5.12 shows our findings.
I n l o o k i n g a t Table 5.12, w e i m m e d i a t e l y can m a k e o n e v e r y i m p o r t a n t
o b s e r v a t i o n . T h e ratio b e t w e e n t h e a v e r a g e profit a n d its s t a n d a r d d e v i a t i o n
n o w is a b o v e 1 in m o s t i n s t a n c e s . ( R e m e m b e r , the h i g h e r t h e ratio b e t w e e n t h e
critical variable a n d its s t a n d a r d deviation, t h e m o r e r o b u s t the s y s t e m . )

TABLE

5.1 1

Summary minimum profit level statistics for Black Jack I trading system.
Percentage minimum target
0.5

1.5

2.5

Average profit

0.16

0.17

0.14

0.16

0.11

Standard deviation

0.23

0.26

0.25

0.25

0.33

Ratio

0.69

0.65

0.54

0.61

0.33

69.39

65.47

62.35

60.58

58.86

Percent profitable
Standard deviation

5.73

5.84

5.81

5.02

4.21

Ratio

12.10

11.20

10.73

12.07

13.98

3.5

4.5

Average profit

0.12

0.11

0.14

0.13

0.12

Standard deviation

0.35

0.40

0.41

0.43

0.40

Ratio

0.33

0.28

0.33

0.30

0.30
54.69

Percent profitable

57.21

56.43

55.52

55.17

Standard deviation

3.86

3.83

3.98

4.04

4.18

Ratio

14.83

14.72

13.93

13.66

13.08

PART 2

78

TABLE

System Concepts

5.12

Summary stop loss statistics for Black Jack I trading system.

Percentage stop loss
0.50

0.75

1.00

1.25

Average profit

0.17

0.18

0.18

0.17

0.15

Standard deviation

0.14

0.14

0.16

0.16

0.15

Ratio

1.50

1.27

1.24

1.18

1.08

1.00

Percent profitable

33.40

38.93

43.57

46.88

49.25

Standard deviation

11.04

9.35

7.92

6.93

6.73

3.02

4.16

5.50

6.76

7.32

1.75

2.00

2.25

2.50

2.75

Ratio

Average profit

0.13

0.13

0.14

0.16

0.16

Standard deviation

0.12

0.15

0.15

0.17

0.16

Ratio

1.03

0.84

0.95

0.94

1.01

Percent profitable

51.37

53.35

55.24

56.79

57.97

Standard deviation

6.50

5.99

5.53

5.54

4.93

Ratio

7.90

8.90

9.99

10.25

11.77

With 6 8 % certainty, we can say that the average profit will be positive. For
instance, with the stop loss at 0 . 5 % , the average profit is 0 . 1 7 % and the standard
deviation 0.14%. This m e a n s that with 6 8 % certainty, we can say that the true
value for the average profit will lie somewhere in the interval 0.03 to 0 . 3 1 % . Table
5.12 also reveals that, after adding the stops, the uncertainty (as measured by the
standard deviation) about the o u t c o m e has decreased at the s a m e time as the average expected profit has increased somewhat for the tightest stops. T h e price for
this is a lower percentage of profitable trades, however. By the s a m e token, we can
see that the percentage of profitable trades s e e m s to increase somewhat the further
away we place the stop, but that this c o m e s at the price of a deteriorating profit
factor. A good c o m p r o m i s e between these two p h e n o m e n a , therefore, s e e m s to be
to place the stop in the m i d d l e of the interval, at the 1.5% level. That this is a compromise is emphasized by the fact that n o n e of the markets tested actually had the
1.5% level as its best alternative.
With all exit levels in place, it is n o w t i m e to look at what performance can
be expected from each of the individual markets. Table 5.13 shows the average
percentage profit for each market, together with its current dollar value.
As you can see, for most of our 15 markets, the average percentage profitable
trade is not big enough to generate a big enough dollar profit to be worthwhile
trading. A l t h o u g h Black Jack I w o r k s well on almost all markets, this does not
m e a n it will be tradable on all markets, because not every system produces big

CHAPTER 5

79

Data Mining

TABLE

5.13

Market summary for Black Jack I trading system.

Market

Average profit

Dollar value

Percent profitable

Orange Juice

0.21

28.69

46.20

Copper

0.18

36.24

44.26

Live cattle

0.30

80.78

55.64

Japanese yen

0.01

13.97

48.50

Corn

0.13

13.48

48.17

T-bond

0.22

243.93

58.48

Eurodollar

0.04

82.52

60.13

Natural gas

0.37

119.40

37.13

-0.13

-88.68

49.38

0.28

99.89

40.97

-0.01

-7.73

52.02

Cotton

0.11

29.74

47.96

T-bill

0.02

51.70

56.71

Crude oil

0.33

76.55

43.93

All markets

0.15

49.25

S&P 500

0.18

579.14

52.86

D-mark
Coffee
CRB-index

e n o u g h dollar profits to overcome the costs. This does not m e a n that Black Jack I
is a b a d system. In fact, of all the markets tested, the only market that might be
worthwhile trading is T-bonds and even this market is a border case. Other m a r k e t s
that, while not ruining you, but at the s a m e t i m e only feed your b r o k e r a r e natural gas, coffee, live cattle, Eurodollar, and crude oil. This does not mean, however, that the m o d e l is not working for these markets, but simply that these markets,
at this particular point in time, are not trading at the appropriate price levels and
with a high enough point value to generate a profit after slippage and commission.
H e n c e , a good working m o d e l is not the s a m e as a profitable model.
The S & P 500 was not a m o n g the markets chosen for the testing process,
because I wanted to save this, perhaps the most popular of all markets, for some outof-sample testing. The bottom row in Table 5.13 shows the initial findings for how
our model would have fared on the S & P 500 futures contract. A n d as you can see, it
would have done fairly well, with an average percentage profit of 0.18%, equal to an
average dollar profit of $579, not counting slippage and commission, in today's market value. This is very interesting information, because the way TradeStation and
several other testing packages calculate this number, it would have come out to
approximately $387. Again, this is because all of these packages look at how many
dollars you could have made, not how m a n y dollars you can expect to make; in an
up-trending market, the historical value will almost always (providing a steady working model) be lower than the actual value for today's market.

PART 2

80

System Concepts

On a separate note, it is once again worthwhile m e n t i o n i n g that the reason

you don't want to adjust your results for slippage and c o m m i s s i o n while you're
building the system is that you would like to find a solution that can trade as efficiently as possible (squeeze out the m o s t of the move) with as few constraints
(rules) as possible. Applying slippage and c o m m i s s i o n at this point will result in a
suboptimal solution and m a k e the system m o r e unreliable when traded on realtime data. Ideally, you should not bother with slippage and c o m m i s s i o n until it is
time to trade the system with real-time data and then decide if the expected profit is worth the expected cost.
Let us use our detailed export function from Part 1 to take a closer look at
h o w Black Jack would have fared trading the S & P 500, as presented by Table 5.14.
Tested on the S & P 500 with data from January 1985 to October 1999, Black Jack
I produced 280 trades, with an average profit of $581 in today's market value.
However, with a drawdown of m o r e than $102,000, Black Jack can hardly be considered a tradable m o d e l a t this point.
Note, however, that in TradeStation this drawdown figure would only have
c o m e out to $43,695 on a one-contract basis. W h y is this? It's because when this
drawdown happened, the market was trading at a m u c h lower level than today, and
consequently the dollar moves were smaller as well. But because the percentage
moves stay approximately constant, you can take this sequence of trades, measured
in percentages, and transfer t h e m to today's market level for an estimate of what the
dollar value would be today. No wonder they say, "your largest drawdown is still to
come." Remember, however, that this value assumed that all trades would have been
taken from today's market value and provided that all profits and losses could have
been reinvested. Therefore, you cannot use this value when comparing your system
with other ready-made systems or the results from a CTA, w h o m you might contemplate giving your m o n e y to instead. Instead, it should only be used for your own
research during the building process. As such, is it m u c h better than any value any
technical analysis or system-trading package will give you.
Now, you probably are thinking, "Yeah, but even so, how likely is it that all
trades would be repeated from t o d a y ? " Well, it's just as likely (or shall we say
unlikely) that the market will repeat itself level for level, move for move, price for
price, which is what you're assuming w h e n you're doing it the old-fashioned way.
If neither of these two assumptions is likely to happen again, ask yourself, " a m I
TABLE

5.14

Performance summary for Black Jack traded on the S&P 500, January 1985-October 1999.
Total trades

280

Winners

148

52.86%

132

47.14%

Profit factor

1.25

Lrg winner

9.50%

30,991

Lrg loser

-2.23%

-7,275

Avg winner

1.69%

5,521

Avg loser

-1.52%

-4,958

C u m profit

55.38%

180,662

Avg profit

0.18%

581

St Dev

2.02%

6,594

Losers

Drawdown - 3 1 . 4 7 % - 1 0 2 , 6 6 2

CHAPTER 5

Data Mining

81

m o r e interested in k n o w i n g what profits or losses I could have m a d e historically,

or am I m o r e interested in getting to know h o w I can m a k e my results as forwardlooking as possible, to get a feel for h o w m u c h I might be m a k i n g t o d a y ? "
Nonetheless, in its current version Black Jack would be very difficult to
trade. But do not despair. As we will see in later chapters, there are m u c h better
ways of building a system like Black Jack than this example.

CHAPTER

Trade or Not to Trade

If a specific system is not worth trading on a specific market, although it works as

well on that market as on any other markets, then it b e c o m e s interesting to know
h o w m u c h the market in which you are interested really is worth, and if the risks
are worth taking considering the volatility of the market and other viable trading
alternatives. This is especially interesting if it is a short-term system, with trades
lasting no longer than a week or two. As it turns out, the weekly moves of most agricultural markets have too small dollar values, as c o m p a r e d to m a n y other markets,
such as currencies, indexes, and interest rates.
Table 6.1 shows how m u c h half the average percentage weekly true range is
worth in today's market value in several different markets. For instance, for coffee
half the average weekly true range for the last ten years and for the last year are
worth $1,343 and $ 1 , 4 5 3 , respectively. At the s a m e time, a 1% move is worth $ 3 5 9 .
For coffee a $1,459 move equals a 4 . 0 5 % move. Hence, the higher the n u m b e r in
the rightmost comparative volatility column, the more volatile the market, on a percentage basis, as c o m p a r e d to the other markets.
Let us take a closer look at some markets in this table. It is interesting to see
that basically all weather-sensitive markets (like the A m e r i c a n agricultural markets), such as corn, wheat and soybeans, have fairly high comparative volatility.
Even m o r e weather-sensitive markets (and perhaps traditional gambling hot spots),
such as coffee, natural gas, and orange j u i c e , have even higher volatility.
On the other hand, the macroeconomic-oriented markets, such as the currencies
and interest rates, all have very low comparative volatility, many times below 1, which
m e a n s that these markets seldom produce a half-weekly average true range of above
1%. With true ranges this low it is doubtful that one can trade a market like this profitably with a short-term system if the current market level and price value aren't high
enough. Fortunately, most macro-markets are still viable trading alternatives.
83

PART 2

84

TABLE

System Concepts

6.1

Dollar value of half average weekly true range (AWTRO, 1% move, and their comparative volatility (AWTR/% move)
Market
Bean oil
Cacao
Canada dollar
Coffee
Copper
Corn
Cotton
CRB index
Crude oil
Dollar index
Eurodollar
Feeder cattle
Gold
Heating oil
Japanese yen
Lean hogs
Live cattle
Lumber
Municipal bonds
Natural gas
Nikkei
Oats
Orange juice
Pork bellies
S&P 500
Silver
Soy beans
Soy meal
Sugar
T-bills
T-bonds
T-notes
Unleaded gas
Wheat

10 years

5 years

1 year

1%

Comp. Vol.

203
231
345
1,343
452
187
502
825
703
866
204
473
339
776
1,425
413
345
714
840

206
200
369
1,512
490
208
499
828
708
781
173
573
310
763
1,617
478
391
756
884
1,628
1,723
149
428
1,045
5,387
641
477
347
209
171
1,046
676
836
316

272
278
447
1,453
492
218
570
1,063
859
825
139
526
393
941
1,825
780
385
650
745
1,542
1,645
135
476
1,311
6,390
616
574
406
325
162
1,030
659
979
321

98
87
680
359
199
100
267
1,020
229
992
2,347
398
294
257
1,206
186
272
251
1,087
322
874
55
134
248
3,264
261
241
147
77
2,374
1,115
1,084
268
128

2.78
3.20
0.66
4.05
2.47
2.18
2.13
1.04
3.75
0.83
0.06
1.32
1.34
3.66
1.51
4.19
1.42
2.59
0.69
4.79
1.88
2.45
3.55
5.29
1.96
2.36
2.38
2.76
4.22
0.07
0.92
0.61
3.65
2.51

148
409
982
4,767
614
444
299
213
192
1,037
680
818
280

But even many highly volatile markets can be difficult to trade if the dollar values of their moves are not high enough. For instance, say that you have a short-term
system with a profit factor as high as 2.2 that is expected to produce 5 5 % profitable
trades. In the best of worlds, you can expect the average winning trade to be half the
average true range of that market. So far, you have only traded it on T-bonds, which
is a very liquid and modestly volatile market for an average profit per trade of $294
(with $75 deducted for slippage and commission). N o w you would like to trade this
system on a m o r e volatile market, like orange juice. With a profit factor of 2.2 and
an average winning trade of $476 (see Table 6.1), the average losing trade is $216
($476 / 2.2). With 5 5 % profitable trades, a sequence of 20 trades will, on average,
produce a profit of $3,292 (11 X $476 - 9 x $216), which in turn breaks down to
an average profit of $165 per trade ($3,292 / 20). Deduct from that another $75 to
cover slippage and commission, and you are down to $90 per contract traded, or less
than one third of what you are making in the T-bonds. Considering that you only get
one or two chances like this per week, making only a little over a $100 a week simply is not worth sticking your neck out and running the risk of losing it all.

CHAPTER

Following the Trend

From the previous chapters we have seen that short-term trading can be incredibly
rewarding, but also very risky, especially if the market you are trading simply does
not p r o d u c e the moves you seek. B e c a u s e not every market can produce a shortterm average profit big enough in dollar t e r m s , the only way to trade these markets
is to go with less frequent trading a n d longer trades that raise the value of the average trade to a level worth trading.
In the previous chapters we discussed how to classify trades and what might
constitute long t e r m and short term. It might be a good idea to always try to be as
short t e r m as you can w h e n it c o m e s to market time units (bars). W h e n y o u ' r e
designing a long-term system, measured in regular time units (days), you should
use weekly (or perhaps even monthly) data during the building and research process
so that you can work with as few market time units (bars) as possible. After an
optimal n u m b e r of time units for the system in question, the further away from the
signal you move, the less profitable and reliable your results will be, as can be seen
from Table 7 . 1 .
Table 7.1 shows the average profit per trade from a 20-day (four-week) breakout system, traded on the long side only on the S & P 500 stock index futures contract
over the period June 1983 to August 1999, and with all trades lasting for a certain
amount of days (weeks) no matter what. F r o m the top half of this table you can see
that, in this case, the longer the holding period, the higher the average profit. But
with a higher average profit comes a higher standard deviation of the individual outcomes, indicating that the uncertainty about the results also is increasing. This can
be measured by the ratio between the average profit and the standard deviation. As
long as the ratio is increasing as well, however, the average profit is growing faster
85

PART 2

86

TABLE

System Concepts

7.1

Trade/time reliability comparison.

Days in trade
10

20

30

40

0.152

0.462

0.854

0.886

1.293

1.304

St. Dev.

2.033

2.822

3.368

3.842

4.913

4.982

Ratio

0.075

0.164

0.254

0.231

0.263

0.262

Profit/day

0.015

0.023

0.028

0.022

0.026

0.022

10

12

Avg. perc. profit

50

60

Weeks in trade
2

Avg. perc. profit

0.366

0.818

1.086

1.205

1.468

1.780

St. Dev.

2.136

2.651

3.433

3.559

4.571

4.746

Ratio

0.171

0.309

0.316

0.339

0.321

0.375

Profit/week

0.183

0.205

0.181

0.151

0.147

0.148

Profit/day

0.037

0.041

0.036

0.030

0.029

0.030

than the standard deviation, indicating a decreasing relative risk. In this case, the
ratio seems to level off after about 50 days, indicating that there is little to be gained
by staying any longer in the trade. This is confirmed by the average profit m a d e per
day, which starts to decrease after 30 days. Similar observations also can be made
from the bottom, weekly part of the table. What is even more important is that by
comparing the weekly data with the corresponding daily data, we can see that the
weekly data produce both m o r e profitable and less risky results, as measured by the
average profit and the standard deviation. That said, in the following long-term systems we work both with weekly and daily data.

MOVING AVERAGES
M o r e than one book or article has been written on the topic of moving averages,
m o s t of t h e m focusing on the basic crossover signal. T h e crossover signal is produced either by the price crossing o n e or several averages, or that of a faster moving average crossing a slower one. (A fast moving average is an average that is calculated with fewer data points, or a shorter lookback period, than a slow moving
average, and vice versa.) W i t h a basic system like this, most of us probably have
produced fantastic results on historical back-testing, only to discover that reality
looks less glamorous when the model is traded live or tested on previously unseen
data. To cure this, m a n y try to filter out the actual buy/sell signals, either by adding
additional requirements like only taking the trade after a retracement and/or on the

CHAPTER 7

Following the Trend

87

second crossover, or by adding additional m o v i n g averages or other indicators of

various lengths.
A n o t h e r strategy that looks good on paper is to avoid the m o v i n g average
model completely w h e n the market is in a trading range, and consequently only
use it when the market is trending. T h e problem is that the only way to classify the
market as either trending or nontrending, using this method, is by hindsight, which
is a luxury that few of us are granted in actual real-time trading. Therefore, for a
long-term system to be stable over time it m u s t be constructed in such a way that
it leaves you with enough slack, so you are not stopped out time and time again
during less advantageous times.
W h e n it c o m e s to m o v i n g averages, the directional slope m e t h o d is a better
choice than the basic crossover method. As an example, we can apply a 100-day
moving average to the C R B index, which is a notoriously choppy market. From its
inception in M a y 1986 to October 1999, there have been 214 occurrences where
the closing price has crossed over the average, but only 160 occurrences where the
average has changed its direction from one day to another. On a trendier market,
like the Japanese yen, there have been 184 crossovers and 127 directional changes
from M a y 1972 to October 1999. With a 200-days average, the s a m e n u m b e r s
c o m e out to 122 and 82 for the C R B index, and 170 and 92 for the Japanese yen.
W h a t does this m e a n ? If it is reasonable to a s s u m e that in a prolonged trend, the
trend of the moving average will follow, you should be able to reduce the n u m b e r
of w r o n g signals considerably by using the slope of the average as a trigger for
your trade, instead of the actual crossover.
B u t that is not all. Notice that for the Japanese yen, with the 200-day m o v ing average the n u m b e r of crossovers is still larger than the n u m b e r of directional
changes for the 100-day average. With the directional slope method, you also can
build a system that uses less and m o r e up-to-date data, and still achieve the same
effect. For instance, for the C R B index the percentage profitable trades went from
1 6 % for the crossover m e t h o d to 3 3 % for the directional slope, with the 200-day
moving average, and from 10 to 3 2 % with the 100-day moving average. For the
Japanese yen the s a m e n u m b e r s went from 17 to 3 6 % with the 200-day moving
average, and from 31 to 4 0 % with the 100-day moving average.
A n o t h e r c o m m o n trading technique, w h e n it c o m e s to moving averages, is to
u s e two moving averages and to require the fast moving average to cross over the
slow one before entering a trade, but only to require the price to cross the fast average to exit. W h e n it c o m e s to the directional slope technique, the same effect can
be achieved by using the slope of a slower moving average for the entry (Entry
M A ) , but the slope of a faster moving average for the exit (Exit M A ) . W h e n the
two contradict e a c h other, you stay out. Yet another way of doing this is to use one
moving average (or one set of averages) for the long side and another average (or
set of averages) for the short side. This could be a good idea for markets with a
natural upward drift, such as the stock market. W h e n you test such a combination,

88

PART 2

System Concepts

r e m e m b e r to treat each side as two different markets, traded with two different systems; w h e n the systems contradict each other, you can either choose to stay out or
trade in smaller positions.
Let's p u t together a m o d e l using the directional slope method, consisting of
two m o v i n g averages that will be the s a m e for both the long a n d the short side and
that will work equally as well on as m a n y markets as possible. A l s o , in accordance
with our finding that we will be better off using as few bars as possible when
examining the system, we will put it together using weekly data substituting a 200day moving average with a 40-bar moving average. At the end of the code we
attach our export function for the profit factor, percentage profitable trades, and
time-in-market, and tailor it so that it also exports the current moving average
combination. T h e TradeStation code for this system follows:
Input: EntryMA(lO), ExitMA(5);
Vars: EntryVal(O), ExitVal(O), PFactor(O), WTrades(O), TotBars(O), TradeStrl("");
EntryVal = Average(Close, EntryMA);
ExitVal = Average(Close, ExitMA);
Conditionl = EntryVal > EntryVal[1];
Condition2 = ExitVal > ExitVal[l];
If Conditionl = True and Condition2 = True Then
Buy at Close;
If Conditionl = False and Condition2 = False Then
Sell at Close;
If Condition2 = False Then
ExitLong at Close;
If Condition2 = True Then
ExitShort at Close;
If LastBarOnChart Then Begin
PFactor = GrossProfit / GrossLoss;
WTrades = NumWinTrades * 100 / TotalTrades;
TotBars = (TotalBarsLosTrades + TotalBarsWinTrades) * 100 / BarNumber;
TradeStrl = LeftStr(GetSymbolName, 2) + "," +
NumToStr(EntryMA, 0)+ "," + NumToStr(ExitMA, 0) + ",'* +
NumToStr(PFactor, 2) + "," + NumToStr(WTrades, 2) + "," +
NumToStr(TotBars, 0) + NewLine;
FileAppend('"D:\Temp\MaDirect.txt", TradeStrl);
End;
T h e m o d e l w a s tested on all moving average combinations on 16 different
markets over the period from January 1979 to October 1994. T h e rest of the data

CHAPTER 7

89

Following the Trend

were saved for some out-of-sample testing. T h e 16 markets tested were: T-bonds,
live cattle, Japanese yen, corn, C a n a d a dollar, crude oil, dollar index, lumber,
orange j u i c e , S & P 500, copper, Eurodollar, C R B index, cotton, gold, and coffee.
No m o n e y w a s deducted for slippage and commission. T h e data w a s put together
using the ordinary point-based back-adjustment method.
With all the data exported into Excel, we build a matrix as a base for a surface chart. This lets us graphically examine our data and look for robust and stable
moving average combinations, which are likely to continue to hold up in the future.
Start by using the data-sort function to sort first by the Entry MA and then
by the Exit M A . Then type in the following formulae in two adjacent c o l u m n s , and
fill d o w n to the bottom of the spreadsheet.
To calculate the average profit factor for each moving average combination:
=IF(C17<>C18;AVERAGE(D2:D17);"")
where column C denotes the Exit MA and c o l u m n D the profit factor for each
moving average combination and market.
To calculate the ratio between the profit factor a n d its standard deviation:
=IF(G2<>"";G2/STDEV(D2:D17);"")
where column G denotes the previously calculated average profit factor.
=INDEX(B:B;2+((ROW()-2)*16))
where column B denotes the Entry M A . T h e value 16 stems from the fact that we
are testing on 16 markets.
=INDEX(C:C;2+((ROW()-2)*16)),
= INDEX(G:G;2+((ROW()-2)* 16))
=INDEX(H:H;2+((ROW()-2)* 16))
where column H denotes the ratio between the profit factor and its standard deviation.
W h e n this is done, in an adjacent column, leave the top cell blank and fill in
the values 0 to 11 in the following rows. Starting at the top row in the next colu m n , fill in the values 0 to 11 in the following c o l u m n s . Figure 7.1 shows what
this can look like.
Figure 7.1 shows a sample matrix created in a spreadsheet p r o g r a m after we
exported the necessary data from TradeStation and applied the suggested formulas. In cell Q2 fill in the following formula and c a r r y it out all the way to cell
AA2:
=INDEX($K:$K;2+(P1*1))
w h e r e column K denotes the Exit MA and cell PI denotes the value zero.

90

PART 2

FIGURE

System Concepts

7.1

The average profit factor for different MA combinations.

In cell P3 fill in following formula, down to cell P I 3 :

=INDEX(J:J;2+(02*11))
w h e r e column J denotes the Entry MA and cell 02 denotes the value zero.
Finally, in cell Q3 fill in the following formula, out and down, all the way to
cell A A 1 3 to complete the matrix in Figure 7 . 1 :
=OFFSET($L$ 1 ;($02*$AA$ 1)+Q$ 1 ;0)
w h e r e column L denotes the average profit factor.
With the matrix ready, it is now easy to create a surface chart similar to that
in Figure 7.2, which shows how the average profit factor for all markets varies with
the length of the moving averages that m a k e up the system. T h e higher the profit
factor, the better the system. Similar charts also can be m a d e for the n u m b e r of
profitable trades or the time spent in the market, or even of the standard deviation
of any of the above, as yet another measure over the robustness of the system.
From Figure 7.2, we can see that an Entry MA in the 16- to 20-bar region
s e e m s to produce very high profit factors with almost any Exit MA from 10 bars
and u p , but also with an Exit MA in the 7- to 8-bar region. T h e most profitable Exit
M A s seem to be the 8-bars and 12-bar Exit M A s . However, from Figure 7.3, we
also can see that the ratio between the profit factor and its standard deviation seems
to be the highest for the 8-bar Exit M A , either together with an Entry MA around
12 bars or with an Entry MA around 18 bars. T h e ratio between the profit factor
and its standard deviation measures the robustness of the system. T h e higher the
ratio, the m o r e robust the system. T h e m o s t robust combinations seem to be with
an 8-bar Exit M A , together with a 12- or 18-bar Entry M A . Therefore, the two
charts in conjunction tell us to go with an 18-bar Entry MA and an 8-bar Exit M A .

CHAPTER 7

91

Following the Trend

FIGURE

7.2

The average profit factor for different MA combinations.

A n o t h e r m e t h o d is to calculate how often a specific moving average c o m b i nation happens to be a m o n g , for instance, the top 10 combinations for each market or a m o n g , for instance, the top 200 for all markets. Using our Entry MA as an
e x a m p l e , in Figure 7.4 we can see that all Entry M A s between 14 and 20 bars were
a m o n g the best 200 combinations at least 15 times each, with the 16-bar MA topping the crowd with a total of 27 occurrences. This strengthens our conclusion that
an Entry MA of 18 bars will be a robust and stable choice that is likely to hold up
in the future.
Figure 7.5 shows yet another e x a m p l e of this technique, but this time we
have sorted on the percent profitable trades c o l u m n and for how m a n y times we
could find any Exit M A s a m o n g the best 200 c o m b i n a t i o n s . As you can see, if
y o u ' r e looking to b o o s t the n u m b e r of profitable trades, you should go for an Exit
MA in the 6- to 10-bar region. T h e jewel of the crown is the 8-bar M A , with 27
occurrences.

92

PART 2

System Concepts

Exit MA
FIGURE
7.3
The ratio between the average profit factor and the standard deviation shows the robustness of the system.

Tables 7.2 and 7.3 show how the directional slope system, with an 18-bar
Entry MA and an 8-bar Exit M A , would have performed on the individual markets
on both the in-sample data and the out-of-sample data. For the in-sample period,
the average profit factor w a s 1.91. With the ratio between the profit factor and its
standard deviation equal to 3.13, the standard deviation c o m e s out to 0.61 (1.91 /
3.13), which m e a n s that we can be 6 8 % sure that the true average profit factor will
be no lower than 1.30 (1.91 - 0.61). D u r i n g the out-of-sample period, the average
profit factor was 2.09. With the ratio between the profit factor and its standard
deviation equal to 0 . 7 3 , the standard deviation c o m e s out to 2.86 (2.09 / 0.73),
w h i c h m e a n s that we no longer can be 6 8 % sure that the true average profit factor
will be above 1. Hence, the profit factor w a s slightly higher for the out-of-sample
data than for the in-sample data, but the price for this came in the form of a smaller ratio between the profit factor and its standard deviation, which m e a n s that the
system has b e c o m e less robust.

CHAPTER 7

93

Following the Trend

FIGURE

7.4

Number of top 200 profit factor occurrences for different entry MAs.

FIGURE

7.5

Number of top 200 percentage winners for different exit MAs.

PART 2

94

TABLE

System Concepts

7.2

Performance summary I for the directional slope system.

Data from January 1979 to October 1994, with 18/8 setting
Market

Entry MA

Exit MA

P factor

% Winners

% In trade

Corn

18

2.06

36.99

Canada dollar

18

46.25

77.00
77.00

Crude oil
CRB index

18

54.35

77.00

18

8
8

1.29
2.73
1.02

33.33

72.00

Cotton

18

2.28

81.00

Dollar index

18

2.07

Eurodollar

18

8
8

45.71
42.86

2.94

48.00

76.00

Gold

18

1.85

42.11

76.00

Copper

18

8
8

1.70

77.00

Japanese yen

18

2.28

35.29
54.84

Coffee

18

2.59

41.03

77.00

Lumber

18

1.99

42.86

76.00

Live cattle
Orange juice

18

8
8

1.48

35.71

79.00

18

1.96

44.16

78.00

S&P 500

18

0.63

T-bonds

18

1.73

33.33
45.83

77.00

Average

1.91

42.67

77.19

Ratio

3.13

6.43

36.68

TABLE

76.00

78.00

81.00

7.3

Performance summary II for the directional slope system.

Data from October 1994 to October 1999, with 18/8 setting
Market

Entry MA

Exit MA

P factor

% Winners

% In trade

Corn

18

12.16

69.23

75

Canada dollar

18

1.19

40

76

Crude oil

18

2.61

52.17

80

CRB index

18
18

40

2.5
0.79

71
80

Dollar index
Eurodollar

18

0.87

18

0.58

Gold

18

1.72

37.5
47.62

Copper

18

0.85

26.92

85

Japanese yen

18

4.37

66.67

76

Coffee
Lumber

18

0.92

25

83

18

1.36

52.38

76

Live cattle

18

1.13

74

Orange juice

18

0.38

45.45
24.14

S&P 500
T-bonds

18

8
8

0.72

40.91
41.67

82

Cotton

Average
Ratio

18

1.23
2.09
0.73

25.93
28.57

73
74
80

83

41.51

78
77.88

2.98

18.93

CHAPTER 7

Following the Trend

95

Just because the system has b e c o m e less robust, however, does not have to
m e a n that it is no longer working as well. It could be that it is still just as good in
catching the right types of moves, but there simply have been no such moves to
catch. O n e way to find out is to look at the overall average profit factor and its
standard deviation ratio for all m a r k e t s and moving average combinations, for both
the in-sample period and the out-of-sample period. This is not shown, so you just
have to take my w o r d for it, but as it turns out the profit factor for the in-sample
period w a s 1.77, as c o m p a r e d to 1.73 for the out-of-sample period. N o t m u c h of a
difference, but what is interesting is that the ratio between the profit factor and the
standard deviation has decreased from 2.63 for the in-sample period, to a mere
1.07 for the out-of-sample period, which m e a n s that not only has the 18/8 combination b e c o m e less robust, but so has the entire logic for this trend-following
m e t h o d (and perhaps all trend-following methods).
Another interpretation is that during the out-of-sample period, most markets
simply have not been in strong enough trends, while others have been trending
extremely well. This results in a higher profit factor, but indeed also in a m u c h higher standard deviation. This can, for instance, be seen in Table 7.3, where corn and
Japanese yen have extremely high profit factorsespecially cornwhile almost all
of the others are losing money, as indicated by a profit factor below or very close to
1, which won't m a k e up for the slippage and commission on real trading.
T h u s , the system still seems to be able to catch the right types of moves, but
for m o s t markets, these moves simply have not b e e n strong enough. Even so, you
would have m a d e a profit during the out-of-sample period as well, as indicated by
an average profit factor above 2, and thanks to a couple of good moves in a few
markets. T h e ride would just have been a little choppier.
Notice also that Figure 7.3 indicates that the trends have not been strong
enough, not that they have disappeared all together. This information can be
derived from the fact that the percentage of profitable trades only decreased from
4 3 % for the in-sample period to 4 2 % for the out-of-sample period, and the 18/8
setting; and from 4 0 % to 3 9 % for the entire methodology (not shown). Again,
however, there is a higher standard deviation for the out-of-sample period, which
again confirms the fact that a few markets have m a d e up for all the others.
This also can be seen from Figures 7.6 a n d 7.7, which are derived the s a m e
way as Figures 7.2 and 7.3, except this time with data for the period October 1994
to October 1999. T h e main difference between Figure 7.2 a n d 7.6 is that during the
out-of-sample period, the profit factor has b e c o m e m o r e heterogeneous, with a few
markets m a k i n g bigger profits than previously and m a k i n g up for m a n y small losses and below previously expected average profits m a d e by other markets. This is
also confirmed by the differences between Figures 7.3 and 7.7.
O n e other interesting observation that can be m a d e from comparing the insample charts and the out-of-sample charts is how well a 12-bar Exit MA holds up
in both periods. Therefore, it can be interesting to look a little closer at the 18/12

PART 2

96

10

11

12

13

14

System Concepts

15

Exitcombinations
MA
The average profit factor for different MA
for the out-of-sample period.
FIGURE
7.6

combination as well. As it turns out, these findings are the s a m e as those for the
18/8 combination. That is, a few very profitable markets keep the overall average
profit factor very high, but with all the other markets showing very low or no profits at all, resulting in a very low stability (high standard deviation).
But this is not the most interesting observation that can be m a d e when studying all four charts. Yet another very interesting finding is the relative robustness
surrounding the area of the 6-bar Exit M A . Figures 7.2 and 7.3 show that this
entire area has a profit factor ranging from approximately 1.6 to 1.9, with a ratio
ranging from 2.0 to 3.0. Not top values by any m e a n s , but good enough, especially considering that these values do not change that m u c h in Figures 7.6 and 7.7,
w h e r e the s a m e values range from approximately 1.0 to 2.0, and from 1.5 to 2.5,
respectively. T h e s e values are perhaps a little lower than for the 18/8 and 18/12
combinations, but the relative robustness over time m a k e s this an interesting area
to consider.

CHAPTER 7

Following the Trend

FIGURE

97

7.7

The ratio between the average profit factor and the standard deviation for the out-of-sample period.

As for the 18/8 and 18/12 combinations: Just because they have not worked
as well nor been as stable on the out-of-sample data does not m e a n that they d o n ' t
w o r k in the long run, but simply that they haven't worked that well on that m a n y
m a r k e t s recently. R e m e m b e r also that the sole p u r p o s e of m a n y trend-following
systems is to catch the few winners that will m a k e up for m a n y of the losers that
occur w h e n trends aren't strong enough to generate decent profits. This is exactly
the situation we have h a d during our out-of-sample period.

DYNAMIC BREAKOUT SYSTEMS

Ever since the mid-eighties w h e n Richard Dennis and Bill Eckhardt m a d e a h u g e
success with their turtle traders, breakout systems have been on everybody's
1

A group of aspiring young traders trained by Richard Dennis. Dennis named the group "turtles" after having
visited turtle farms in Singapore and having decided to grow traders as the farms grew turtles.

PART 2

98

System Concepts

m i n d . T h e concept is easy enough: never m i n d trying to pick tops and b o t t o m s

if the m a r k e t is going up, you buy, and if it is going down, you sell. In fact, instead
of trying to sell the tops and buy the b o t t o m s , do the opposite. That is, if the market takes out the highest high for the last n days then buy, or if the m a r k e t takes
out the lowest l o w for the last n days then sell. With this basic system, y o u are in
the market all the time, either with a long position anticipating an up-trend after a
breakout on the long side, or a short position after a breakout on the short side.
To this basic system several improvements or refinements can be m a d e that
could be used either separately or together with others. A m o n g these is to treat the
long and the short side separately, assigning t h e m different lookback periods, or to
have a separate lookback period handling the exits, w h i c h then results in the system's also allowing for flat or neutral market situations in times of consolidation.
A n o t h e r refinement is to m a k e the lookback period d y n a m i c and sensitive to the
historical volatility of the market. Yet another way is to keep the lookback period
constant but add a volatility factor to the breakout level, w h i c h can be, for
instance, the highest high for the last n days, times a volatility factor.
O n e such volatility-sensitive system is the D y n a m i c Breakout System that
has been featured in several articles for Futures m a g a z i n e during 1996 and 1998.
T h e logic behind the D B S system, as it is called for short, is that in times of relative high volatility the n u m b e r of "false" breakouts increases together with "false"
trend reversals a n d deeper than expected corrections. Therefore, a volatility comp o n e n t increases the lookback period, to m a k e it harder both to enter and exit a
trade "too soon." T h e TradeStation code for the D B S system with a variable lookback period of 20 to 60 days, follows:
Inputs: MaxLB(60), MinLB(20);
Vars: HistVol(O), YestHistVol(O), DeltaHistVol(O), EntryLB(O), ExitLB(O),
YestEntryLB(O);
YestHistVol = HistVol;
HistVol = StdDev(C, 30);
DeltaHistVol = (HistVol - YestHistVol) / HistVol;
If CurrentBar = 1 Then
EntryLB = 20;
YestEntryLB = EntryLB;
EntryLB = YestEntryLB * (1 + DeltaHistVol);
EntryLB = MaxList(EntryLB, MinLB);
EntryLB = MinList(EntryLB, MaxLB);
ExitLB = EntryLB * 0.5;
Buy Tomorrow at Highest(High, EntryLB) Stop;
Sell Tomorrow at Lowest(Low, EntryLB) Stop;

CHAPTER 7

Following the Trend

99

ExitLong Tomorrow at Lowest(Low, ExitLB) Stop;

ExitShort Tomorrow at Highest(High, ExitLB) Stop;
F r o m the code (we will call this version l a ) , you can see that the lookback
period for the exits will always be half the lookback period for the entries at that
particular point in time. T h e problem with this version, however, is that the variable lookback period does not always work as intended. For instance, when traded
on the C R B index over the period June 1986 to October 1999, and the Japanese
yen over the period M a y 1972 to October 1999, the n u m b e r of trades actually
decreased somewhat for b o t h markets, if the variable lookback period was substituted for a stationary lookback period of 20 days. For the C R B index the n u m b e r
of trades decreased from 125 to 112 trades, a n d for the Japanese yen the same
n u m b e r s were 191 with the variable lookback period and 181 without it. T h e g o o d
n e w s is that, at least for the Japanese yen, the variable lookback period m a n a g e d
to p r o d u c e a few m o r e profitable t r a d e s 5 2 % as compared to 4 6 % with the 2 0 day stationary lookback period. T h e s a m e n u m b e r s for the C R B index were 26 and
2 8 % respectively. H e n c e , the first part of this little study contradicts the logic
behind the system, while the results from the second part can only be considered
inconclusive.
A n o t h e r less desirable feature with this system is that with an increasing
lookback period y o u r u n the risk of having the stops move away from the market
instead of with it, if the m a r k e t takes off in either direction. Also, usually when the
market takes off in a trend, the volatility increases with the direction of the trend,
w h i c h is a g o o d thing. Hence, there are two types of volatility, good volatility that
works for your position and bad volatility that works against it. O n e example of
this p h e n o m e n o n occurred October 7, 1998, w h e n the Japanese yen took off with
a m o v e w o r t h almost $10,000, a n d continued higher over the next few days for a
total move worth $14,500. Figure 7.8 shows h o w this affected the breakout and
exit levels for the D B S system w h e n viewed with the ordinary point-based backadjusted contract. W h e n the m a r k e t t o o k off on October 7, the system gave a buy
signal at 0.822 (the solid u p p e r line), with a stop loss placed at 0.781 (the dotted
lower line), for a total value at stake of $5,125 on a one-contract basis. During the
next couple of days, the volatility of the market increased the lookback period to
its m a x i m u m 60 days, w h i c h in turn resulted in a lowering of the stop loss to 0.764,
for a total value at stake from the entry point of $7,250, or $15,875 from the close
of 0.891 at October 13.
O n e way to change the stop-loss d i l e m m a is to apply the same logic to the
exits as to the entries. That is, if higher volatilitya sign of increasing insecurity
and higher r i s k m a k e s it m o r e difficult to enter the m a r k e t (or easier to stay out,
if you so wish), shouldn't the s a m e reasoning then m a k e it easier to exit the market (or m o r e difficult to stay in) as well? If you adhere to that theory, the
TradeStation code for the D B S system can be rewritten as follows:

PART 2

100

FIGURE

7.8

The DBS system applied on the Japanese yen.

System Concepts

CHAPTER 7

Following the Trend

101

Inputs: MaxEntryLB(60), MinEntryLB(20), MaxExitLB(30), MinExitLB(lO);

Vars: HistVol(O), YestHistVol(O), DeltaHistVol(O), EntryLB(O), ExitLB(O),
YestEntryLB(O), YestExitLB(O);
YestHistVol = HistVol;
HistVol = StdDev(C, 30);
DeltaHistVol = (HistVol - YestHistVol) / HistVol;
If CurrentBar = 1 Then
EntryLB = 20;
YestEntryLB = EntryLB;
EntryLB = YestEntryLB * (1 + DeltaHistVol);
EntryLB = MaxList(EntryLB, MinEntryLB);
EntryLB = MinList(EntryLB, MaxEntryLB);
YestExitLB = ExitLB;
ExitLB = YestExitLB * (1 - DeltaHistVol);
ExitLB = MinList(ExitLB, MaxExitLB);
ExitLB = MaxList(ExitLB, MinExitLB);
Buy Tomorrow at Highest(High, EntryLB) Stop;
Sell Tomorrow at Lowest(Low, EntryLB) Stop;
ExitLong Tomorrow at Lowest(Low, ExitLB) Stop;
ExitShort Tomorrow at Highest(High, ExitLB) Stop;
In this version of the c o d e (version l b ) , there also are situations w h e n the
l o o k b a c k p e r i o d for the entry is shorter than for the exit, w h i c h m e a n s that there
also will be several stop-and-reversal situations, as well as plain exits into a
neutral position. P e r h a p s this c o u l d b e c o n s i d e r e d slightly m o r e " d y n a m i c " than
the original s y s t e m , w h e r e t h e l o o k b a c k p e r i o d for the entry always is longer
than for t h e exit. In this c a s e , the n u m b e r of t r a d e s w h e n t r a d e d on the J a p a n e s e
yen over the period M a y 1972 t o O c t o b e r 1999 w a s 2 0 1 , a m o n g w h i c h 5 3 %
w e r e profitable. For t h e C R B index, the s a m e n u m b e r s over the period J u n e
1986 to O c t o b e r 1999 w e r e 3 1 % profitable t r a d e s out of a total of 114 t r a d e s .
To be sure, however, s o m e of these extra t r a d e s can be explained by the fact that
a faster exit also allows the s y s t e m to enter the m a r k e t a s e c o n d t i m e in the
s a m e direction. N o n e t h e l e s s , b e c a u s e these results are i n c o n c l u s i v e , as c o m p a r e d to w h a t would be d e s i r a b l e , t h e q u e s t i o n still r e m a i n s : is t h e r e a way to
u s e the historical volatility both to d e c r e a s e the n u m b e r of t r a d e s a n d increase
the n u m b e r o f w i n n e r s ?
By letting the volatility alter the lookback period, we only implicitly alter the
actual level at which a trade is triggered. A m o r e direct way to do this is to let the
volatility explicitly alter the level that triggers the trade. An attempt at such a syst e m coded in TradeStation (version 2a) can look something like this:

PART 2

102

System Concepts

Inputs: VolMethod(l), VolFactor(0.5);

Vars: HistVol(O), HalfHistVol(O), LongEntry(O), ShortEntry(O), LongExit(O),
ShortExit(O);
If VolMethod > 0 Then
HistVol = StdDev(C, 30) * VolFactor
Else
HistVol = AvgTrueRange(30) * VolFactor;
HalfHistVol = HistVol * 0.5;
LongEntry = Highest(High, 20) + HistVol;
ShortEntry = Lowest(Low, 20) - HistVol;
LongExit = Lowest(Low, 20) - HalfHistVol;
ShortExit = Highest(High, 20) + HalfHistVol;
Buy Tomorrow at LongEntry Stop;
Sell Tomorrow at ShortEntry Stop;
ExitLong Tomorrow at LongExit Stop;
ExitShort Tomorrow at ShortExit Stop;
With this system, we add half the 30-day historical volatility to the breakout
level, but only a quarter to the exit level, m a k i n g it relatively harder to get in than
to get out. In this way, we still leave the system with some slack so as not to be
stopped out too soon, but also allow for neutral, or flat, positions in case the
volatility gets too rough. Another way to achieve essentially the s a m e thing is to
substitute all the historical volatility factors, as measured with the standard deviation, with the average true range for the period. You can do this in the code by
changing the VolMethod input to zero.
To m a k e it easier for the system to exit in times of high volatility, simply
change the plus sign to a m i n u s sign when calculating the long exit, and the minus
sign to a p l u s sign w h e n calculating the short exit (this will be version 2b). In
doing so, however, the system exits a trade before the market reaches down (up)
to the lowest low (highest high) of the lookback period, which is a s s u m e d to be a
pivotal support (resistance) level.
L e t us see h o w all these systems would have performed on a portfolio of 16
markets over the period from January 1980 to October 1999. T h e 16 m a r k e t s we
will use are S & P 500, copper, crude oil, live cattle, cotton, Eurodollar, Canada dollar, Japanese yen, gold, lumber, orange j u i c e , corn, coffee, dollar index, CRB
index, and T-bonds. Because we are not interested in any percentage-based calculations or relationships, we use the ordinary point-based back-adjusted contract.
No m o n e y has been deducted for slippage and c o m m i s s i o n to avoid c o m i n g up
with any suboptimal solutions. We do this by using our export function for the
profit factor, percent profitable trades, and time in market, from Part 1.

CHAPTER 7

Following the Trend

103

Table 7.4 shows that the original system (version l a ) indeed h a d the highest
profit factor, but that this also c o m e s at the expense of a higher standard deviation,
as c o m p a r e d to version l b . This is illustrated by the first ratio column, w h i c h
depicts the ratio between the profit factor a n d its standard deviation. T h e higher
the ratio, the m o r e h o m o g e n o u s the behavior between the m a r k e t s . For instance,
with a profit factor of 1.46 and a ratio of 2.74, D B S system version la cannot be
expected, with 6 8 % certainty, to p r o d u c e a profit factor above 1 (1.46 1.46 /
2.74 = 0.93). This also holds true for version lb (1.36 - 1.36 / 3.52 = 0.97), b u t
in choosing b e t w e e n the two, version lb c o m e s out on top because of its higher
stability. Interestingly enough, all version b systems have higher ratios than their
version A counterparts, except for the t i m e spent in the market. In addition, w h e n
it c o m e s to the two m o s t important readings, the profit factor and the percentage
profitable trades, all version 2b systems also show better results. T h a t is, in m o s t
instances, not only do the version b systems show results that are m o r e profitable,
they also s e e m to be m o r e robust when c o m p a r e d b e t w e e n different markets.
L e t us take a closer look at version lb in Tables 7.5 and 7.6, and see h o w it
performed on all the individual markets. For this we also use a variation of our
trade-by-trade export function, w h i c h we also developed in Part 1. To do so, we
also m u s t switch over to the R A D contract. T h e code for this version of the export
function follows:
M

Vars: FileName(" ), TotTr(O), Prof(0), TradeStr2("");

If CurrentBar = 1 Then Begin
FileName = "DATempV + LeftStr(GetSymbolName, 2) + ".txt";
FileDelete(FileName);
TradeStr2 = "Position" + "," + "Profit" + NewLine;
FileAppend(FileName, TradeStr2);
End;
TotTr = TotalTrades;

TABLE

7.4

Trade statistics summary for different DBS systems.

Version

P Factor

Ratio

Trades

Ratio

VeMa

1.46

2.74

145

VeMb

1.36

3.52

140

Ver 2a (HV)

1.36

2.64

Ver 2b (HV)

1.38

Ver 2a (ATR)
Ver 2b (ATR)

% Prof.

Ratio %

In market

Ratio

7.39

37.24

5.64

79

28.29

7.89

40.90

8.16

80

33.65

100

7.10

39.21

5.72

88

27.67

2.65

116

7.48

40.02

6.54

76

15.95

1.35

2.71

107

6.78

39.24

5.73

90

42.50

1.36

2.92

119

7.49

39.36

.39

81

21.58

PART 2

104

System Concepts

If TotTr > TotTr[l] Then Begin.

Prof = 1 + PositionProfit(l) / (EntryPrice(l) * BigPointValue);
TradeStr2 = NumToStr(MarketPosition(l), 0) + "," +
NumToStr((Prof - 1) * 100, 2) + NewLine;
FileAppend(FileName, TradeStr2);
End;
If LastBarOnChart Then Begin
TradeStr2 = NumToStr(Close, 4) + "," + NumToStr(BigPointValue, 2) +
NewLine;
FileAppend(FileName, TradeStr2);
End;
Table 7.5 shows that, as is the case with m a n y trend-following breakout-type
systems, this version of the D B S system works very well on a handful of markets,
while others barely break even or lose money, even if at a not-too-alarming rate.
T h e question is, are these markets still worthy of being traded in a portfolio? Table
7.6 gives a m o r e detailed picture of the same p h e n o m e n o n . F r o m this table we can
see that the average losses for most markets are quite substantial in percentage
t e r m s . Copper, for instance, will on average allow for a 4 . 2 5 % move against the
trade before it allows you to exit.

TABLE

7.5

Trade statistics summary for DBS system, version lb.

Market

P Factor

Trades

% Prof.

% In market

Canada dollar

1.03

152

38.82

78

Coffee

1.47

155

39.35

79

Copper

0.90

158

39.87

80

Corn

1.72

144

42.36

81

Cotton

1.27

148

35.14

84

CRB index

0.65

114

30.70

78

Crude oil

1.56

117

47.01

82

Dollar index

1.60

95

44.21

83

Eurodollar

1.78

127

46.46

83

Gold

1.53

141

43.26

80

Japanese yen

2.05

144

50.00

78

Live cattle

0.83

157

35.03

79

Lumber

1.38

147

40.14

78

Orange juice

1.62

152

41.45

79

S&P 500

1.00

139

36.69

78

T-bonds

1.33

146

43.84

85

CHAPTER 7

Following the Trend

TABLE

105

7.6

Trade statistics summary for DBS system, version lb.

Avg. Trade

Avg. Profit

Market

(%)

Canada dollar

0.04

26

Coffee

1.51

541

Copper

($)

(%)

Avg. Loss

($)

(%)

1.61

1,092

-0.80

-545

13.46

4,830

-5.76

-2,065
-848

($)

-0.34

-67

5.91

1,178

-4.25

Corn

1.00

100

7.47

749

-3.48

-349

Cotton

0.72

194

8.76

,340

-3.81

-1,017

CRB index

-0.52

-532

2.74

2,793

-1.78

-1,814

Crude oil

2.38

545

10.74

2,463

-5.40

-1,239

Dollar index

0.67

663

3.13

3,100

-1.44

-1,427

Eurodollar

0.09

208

0.59

1,378

-0.29

-677

Gold

0.59

175

5.53

1,628

-2.98

-876

Japanese yen

1.02

1,225

4.30

5,188

-1.90

-2,294

Live cattle

0.01

3.89

1,057

-2.64

-716

Lumber

0.37

93

9.43

2,368

-5.09

-1,277

Orange juice
S&P 500
T-bonds

1.21

162

8.64

1,156

-4.43

-592

-0.33

-1,093

3.94

12,848

-2.78

-9,060

0.23

255

3.81

4,250

-2.24

-2,499

Let us also investigate how version lb would have performed over two different time periods and see if we can add to or confirm any of the conclusions we
m a d e about our directional slope system in the previous chapter. In this case, we
will look at the periods January 1980 to October 1989 (Table 7.7) and January
1990 to October 1999 (Table 7.8). Table 7.7 shows that, although this version of
the D B S system trades profitably in m o s t markets, the profit factor is somewhat
on the low side. N o t e also that the profit factor differs somewhat from our original findings, due to the different contracts used. Here I used the ratio-adjusted contract and calculated the profit factor with the help of the average weighted values
of the winners and losers. As shown in Table 7.8, for the last ten years, the results
look even less promising.
As w a s the case with the directional slope system, we can say that this version of the D B S system has not performed as well recently as in the past. This can
be seen, for instance, from the average profit factor for the two periods, which not
only is slightly lower for the later period, but also has a higher standard deviation
than in the earlier period. As for all other numbers, they too are not as good for the
later period. T h e n u m b e r of profitable trades, for example, has decreased ever so
slightly; but another indication that there are less profitable trades can be derived
from the time in market column and the n u m b e r of trades column. Usually, the

PART 2

106

System Concepts

Performance summary for version lb, January 1980-October 1989.

Market

P factor

Trades

% Winners

% In trade

Canada dollar

1.01

78

37.18

79

Coffee

1.66

76

39.47

79

Copper

0.71

80

37.5

82

Corn

1.65

71

42.25

84

Cotton

1.32

68

36.76

86

CRB index

0.84

25

28

74

Crude oil

2.31

40

55

84

Dollar index

1.04

28

35.71

82

Eurodollar

1.91

41

51.22

82

Gold

1.48

68

42.65

77

Japanese yen

2.44

71

56.34

78

Live cattle

0.97

70

40

77

Lumber

1.38

70

40

81

Orange juice

2.41

69

49.28

81

S&P 500

1.25

56

39.29

75

T-bonds

1.35

73

42.47

85

Average

1.48

61.50

42.07

80.38

Ratio

2.71

3.44

5.62

22.60

% Winners

% In trade

TABLE

7.8

Performance summary for version lb, January 1990-October 1999.

Market

P factor

Trades

Canada dollar

0.95

72

38.89

75

Coffee

1.33

76

39.47

78

Copper

1.11

74

41.89

76

Corn

1.9

68

45.59

76

Cotton

1.16

75

36

83

CRB index

0.57

83

33.73

78

Crude oil

1.24

74

43.24

80

Dollar index

2.37

58

51.72

80

Eurodollar

3.12

38

60.53

85

Gold

1.53

70

44.29

81

Japanese yen

1.83

69

43.48

77

Live cattle

0.62

81

32.1

80

Lumber

1.39

73

39.73

75

Orange juice

0.85

80

33.75

76

S&P 500

0.92

79

34.18

79

T-bonds

1.44

66

46.97

83

Average

1.40

71.00

41.60

78.88

Ratio

2.11

6.57

5.58

25.84

CHAPTER 7

Following the Trend

107

higher the profit factor, the fewer the trades and, simultaneously, the m o r e time
spent in the market. It will be interesting to see if we can improve on these n u m bers in the u p c o m i n g sections.
C o m p a r e d to the directional slope system, the profit factor also is considerably lower, while the percentage of profitable trades is approximately the same,
albeit with a lower standard deviation. This is an interesting observation that suggests that whatever system you chose, there always is a price to pay, and you have
to c o m p r o m i s e on one performance measure in favor of another.
R o t h the D B S system and the directional slope system have in c o m m o n the
fact that they do not have any fixed stop levels. Although the directional slope system lacks a natural stop level completely, the D B S system at least has a d y n a m i c
level. Even in version l b , however, this level is running the risk of moving away
from the market, which m a k e s it difficult to trade this system with a fixed fractional m o n e y m a n a g e m e n t strategy.
This takes us back to the beginning of this section where we talked about the
turtles-, because what is even m o r e important for a turtle than w h e n or what to
trade, is h o w m u c h to trade. For a turtle it is vital to cut b a c k in size during times
of high volatility and drawdown to preserve as m u c h capital as possible in expectation of the next major move, which can c o m e in any market.

STANDARD DEVIATION BREAKOUT

T h e Bollinger bands indicator is one of the m o s t versatile indicators there is. It can
be used both in short-term oscillator-type systems, looking for overbought/oversold situations or in a long-term breakout-type system. To calculate the indicator,
first calculate an appropriate-length moving average, and then add the n u m b e r of
standard deviations for the s a m e time period to that average for the upper band or
deduct the n u m b e r of standard deviations for the lower band. For a long-term
breakout system, experience has shown that you should look for a lookback period for the moving average and the standard deviation calculations somewhere
between 50 a n d 100 days. For a shorter-term system, you should aim for about 20
to 40 days. Usually the Bollinger bands indicator is charted together with the price,
as shown in Figure 7.9. W h e n you look at Figure 7.9 it is important to r e m e m b e r
that j u s t because the price penetrates the b a n d s , it d o e s n ' t have to c o m e back from
there. Equally as often it will be the b a n d that catches up with the price. In fact,
the Bollinger b a n d s indicator often m a k e s up the foundation for some excellent
breakout systems. T h e TradeStation code for this is as follows:
Inputs: BandLen(60), NoStDev(2);
Vars: BandDevi(O), MidBand(O), UpBand(O), LoBand(O);
BandDevi = StdDev(Close, BandLen) * NoStDev;
MidBand = Average(Close, BandLen);

PART 2

108

FIGURE

7.9

The Bollinger bands indicator applied on the Deutsche Mark.

System Concepts

CHAPTER 7

Following the Trend

109

UpBand = MidBand + BandDevi;

LoBand = MidBand - BandDevi;
Plotl (MidBand, "MidBand");
Plot2(UpBand, "UpBand");
Plot3(LoBand, "LoBand");
M a n y t i m e s it can be difficult to see what is actually h a p p e n i n g w h e n the
price reaches o n e of the b a n d s a n d then starts to h u g the b a n d in a trending motion.
In this case it might be better to chart the indicator as a n o r m a l i z e d oscillator
instead, w h e r e you keep the upper a n d lower bands stationary at, for instance, 100
a n d 0, respectively. T h i s is especially useful if y o u are primarily interested in
catching the short-term m o v e against the underlying trend. This is shown in Figure
7.10, w h i c h u s e s the s a m e chart as in Figure 7.9, b u t this t i m e with the Bollinger
bands oscillator added at the bottom. T h e code for the oscillator follows:
Inputs: BandLen(60), NoStDev(2);
Vars: BandDevi(O), MidBand(O), UpBand(O), LoBand(O), BandPos(O);
BandDevi = StdDev(Close, BandLen) * NoStDev;
MidBand = Average(Close, BandLen);
UpBand = MidBand + BandDevi;
LoBand = MidBand - BandDevi;
BandPos = (AvgPrice - LoBand) * 100/ (UpBand - LoBand);
Plotl (BandPos, "Position");
Plot2(100, "UpBand");
Plot3(0, "LoBand");
O n e interesting p h e n o m e n o n of the Bollinger bands indicator w h e n used as
an oscillator is h o w fast it catches the m o s t recent moves despite a relatively long
lookback period. This m e a n s that you will be able to take m u c h m o r e data into consideration but still achieve the s a m e effect as, for instance, with an ordinary R S I or
stochastic indicator. Or, alternatively, use the s a m e amount of data for a faster reacting indicator. Figure 7.11 shows a 20-day Bollinger oscillator together with a 10day RSI, charted on top of each other, depicting the same price activity on the s a m e
market. This time the upper and lower boundaries for the Bollinger bands oscillator are set to 70 a n d 30, respectively. As you can see, although the Bollinger bands
indicator takes twice as m u c h data into consideration, it still moves m u c h faster
between the overbought and oversold regions than the R S I indicator.
A l t h o u g h the Bollinger bands indicator has several advantages c o m p a r e d to
m o s t other indicators, it also shares several of their disadvantages. O n e of t h e m we
stressed w h e n w e discussed the M e a n d e r indicator: you can m a k e n o comparative
assumptions b e t w e e n two different readings, whether you look at the s a m e market
at two different points in time or at two different markets at the s a m e point in time.

110

PART 2

FIGURE

7.10

The Bollinger bands indicator also can be charted as an oscillator.

System Concepts

CHAPTER 7

FIGURE

Following the Trend

7.11

The Bollinger bands indicator reacts much faster than RSI.

111

PART 2

112

System Concepts

W h e n it c o m e s to using the Bollinger bands in a trading system, another disadvantage is that you will take the volatility into consideration twice, as compared
to what happens in a basic breakout system. For instance, if you currently do not
have a position on and the market is in a relatively w i d e trading range, with relatively high historical volatility readings, it already is relatively difficult for an ordinary highest high/lowest low breakout system to enter the market, because of the
large distance between these entry levels. But to a Bollinger band-type system you
also have to add the distance to the upper a n d lower bands which, because of the
volatility, could be placed at a considerable distance away from the highest high and
lowest low. Also, as is the case with m a n y breakout-type systems, most Bollinger
bands systems only seem to take the volatility into consideration when entering the
market but not w h e n exiting. Instead the exit usually is governed only by a crossover of the mid-band, which most of the time is a standard moving average.
Nonetheless, here is an alternative long-term system, w h e r e we m a k e it even
harder than usual to enter. T h e TradeStation code follows:
Inputs: BandLen(60);
Vars: UpBand(O), LoBand(O);
UpBand = XAverage(High, BandLen) + 2* StdDev(High, BandLen);
LoBand = XAverage(Low, BandLen) - 2 * StdDev(Low, BandLen);
Buy tomorrow at UpBand Stop;
Sell tomorrow at LoBand Stop;
ExitLong tomorrow at XAverage(Low, BandLen) Stop;
ExitShort tomorrow at XAverage(High, BandLen) Stop;
As you can see from the code, for the upper band, we have substituted the
closing price with the high, and for the lower band, we have substituted the closing price with the low. We also have m a d e it a little m o r e difficult to exit by substituting the closing price with the low, for the long exit, and the high, for the short
exit. To c o m p e n s a t e s o m e w h a t for both the slower/delayed entries and exits, we
also substituted the standard moving average for an exponential one. All in all, we
hope that these slight changes to the original setup will w e e d out a few trades that
were b o u n d to go nowhere and only catch those signals that have a high enough
m o m e n t u m to follow through and develop into a profitable trade. T h e lookback
period is set to 60 days and has not been optimized in any way. To find an optimal,
robust setting you could use the same technique as we did for the directional slope
system, w h e r e we exported the different moving average combinations and then
put together a couple of so-called surface charts.
T h e m o d e l w a s tested on data from January 1980 to D e c e m b e r 1992. The
period January 1993 to October 1999 w a s saved for some out-of-sample testing,
after we added a few additional stops and s o m e sort of filter, in section three and
four, respectively. T h e 16 markets tested were D-mark, crude oil, lumber, copper,

CHAPTER 7

Following the Trend

113

gold, dollar index, live cattle, T-bonds, cotton, Japanese yen, natural gas, wheat,
Nikkei, coffee, T-bills, and r o u g h rice. Because we used the ordinary point-based
back-adjusted contract, we added the export function for the profit factor, days in
trade, and percentage profitable trades to the code above, for further analysis in a
spreadsheet program. Table 7.9 shows the results. No m o n e y has been deducted for
slippage a n d commission.
As y o u can see from Table 7.9, this version of the Bollinger bands system
s e e m s to have several desirable features, such as a high profit factor in m o s t markets, a high percentage of profitable trades, and a fairly low percentage of the time
spent in the market. In fact, as Table 7.9 shows, all markets traded profitably, with
profit factors well above 1 in m a n y cases a n d with the average profit factor as high
as 3.82, attributable to the slight changes we m a d e to the original system. For a
long-term system, m o s t markets also h a d a very high percentage of profitable
trades, with an average of 5 1 . 7 4 % profitable trades per market.

TABLE

7.9

Long-term Bollinger bands system, January 1980-December 1992.

Market

P factor

Trades

% Winners

% In trade

Deutsche mark

3.62

34

61.76

59

Crude oil

5.41

23

73.91

61

Lumber

1.21

46

32.61

53

Copper

2.29

37

40.54

52

Gold

1.67

43

41.86

55

Dollar index

3.92

17

64.71

54

Live cattle

1.32

34

38.24

44

T-bonds

1.94

33

45.45

52

Cotton

2.32

39

43.59

59

Japanese yen

3.01

40

60

62

Natural gas

10.28

66.67

82

Wheat

2.37

38

34.21

53

Rough rice

12.65

15

66.67

59

Timber

3.41

33

60.61

62

3.3

34

47.06

50

50

63

Coffee
Nikkei index

2.42

PART

TWO

A Few Final Thoughts About Part 2

In this part, we created a few basic trading systems to take with us to the u p c o m ing sections, w h e r e we will continue to develop and e x a m i n e t h e m further.
A l t h o u g h all the systems have been very basic, with as few rules as possible, we
have shown that it is possible to exploit different types of market behavior and create a trading system that is likely to continue to work well in the future and on several different m a r k e t s s o m e t i m e s even on markets the system has not been tested
on previously.
In fact, the whole key to m a k i n g a system robust is to keep it as simple as p o s sible, while at the s a m e time m a k i n g sure it works well on several different markets.
Notice once again that the key words here are "work well," not " b e profitable." For
a well-working system to be profitable it needs to be applied to a market that can
exhibit a high enough dollar value in relation to the moves the system is designed
to catch. This does not always have to be the case, but this does not m e a n that the
system is a b a d system.
Before we go on to Part 3 , 1 would like to leave you with a few questions and
a little food for thought. Is there a way of measuring curve fitting? Is it as easy as
saying that the better the result, the m o r e curve-fitted the system; or the m o r e rules,
the m o r e curve-fitted the system? For instance, say that you just have finished
building a system with thousands of rules, which will capture every historical daily
m o v e in the S & P 500 futures contract perfectly. No doubt this system is curve fitted, but w h a t if you reverse all the rules? In other words, for every rule you had, you
will add o n e rule, ending up with twice as many. W h a t will the historical results
from this system be? T h e y will be disastrous. So is this second system m o r e or less
curve fitted that the first one?
115

116

PART 2

System Concepts

H e r e is a suggestion on h o w to measure the degree of curve fittedness: take

the market or m a r k e t s on which you would like to trade, manually pick out all
those market segments you would like for your system to catch, a n d string them
together in a fantasy equity curve or several curves. Then start building your system, adding and deducting rules in an effort to m i m i c the fantasy results. For every
change you m a k e , measure the difference between the fantasy curves and those on
which you are working by looking at the correlation between t h e m . If you have to
add a rule to increase the correlation, that is a bad thing, b u t if you m a n a g e to
increase the correlation by deducting a rule, that is a good thing.
Have I d o n e this myself? N o p e . It just c a m e to m e , but nonetheless, here is
what I consider to be a rule: every t i m e I n e e d to use any equal to ( = ) , m o r e than
( > ) , or less than ( < ) signs I a m , in fact, formulating a rule. For m e , the trick then
is to keep all these signs down to a m i n i m u m , preferably below five for the entire
strategy. As you will discover, if you are being honest about what constitutes a
sign, it is very difficult.
In the next part, we will see if we can improve each system's performance
further by adding a few stops to protect us from too severe losses and lock in profits. Although it is true that every rule you add to the system m e a n s that you probably are curve fitting it more, adding stops and exits is a necessary evil and should
be looked u p o n m o r e as a c o m m o n s e n s e proceeding than actual curve fitting.
Look at it this way: if you are long in the market and it goes against you, you lose
money, and losing m o n e y is not a bright thing to do. At least not in a systematic
fashion, and that is w h a t this is all about.

PART

THREE

Getting Out

O n c e you have a basic idea about what you would like to achieve with your system
and which type of moves you would like it to catch, it is time to start experimenting
with different types of stops a n d exits. If you wish, you can look at any basic system, like those in the previous section, as a set of entry rules only. That is, the basic
system tells you w h e n it is time to enter into a long, short, or neutral position. Of
course, entering into a short position usually also m e a n s exiting a long position. But
in essence, a basic system is a set of entry rules, which we hope signals a high-probability trading opportunity, no m o r e no less. But as so m a n y successful traders have
said, "anyone can enter the market, but to be successful in the long run, the trick is
to consistently exit with profits that are bigger than the losses."
F r o m this it follows that m o s t of the time the exits are m u c h m o r e important
than the entries. Therefore, starting out with the entry is not always the best way to
go. Of course, you might already have c o m e up with a set of excellent entry techniques without giving the exits any second thoughts. In that case, it should "only" be
a question of examining or adding a set of exits to improve performance even further.
S o m e t i m e s , however, it is an even better idea to start out w i t h the stops and
exits, and work b a c k w a r d to c o m e up with a simple entry technique that matches
your exit criteria a n d the underlying assumptions for the trade. Notice, however,
that this does not by any m e a n s m e a n that the work you do looking into different
entries is a waste of time, because m a n y times it will turn out that there will be a
better entry technique than the one with w h i c h you started. B u t to k n o w about it you
n e e d to have e x a m i n e d and tested it on a previous occasion.
In Part 4, we continue the development process by adding some sort of filter
technique. A filter efficiently keeps you away from the markets as m u c h as possible,
117

PART 3

118

Getting Out

only letting you place a trade according to your original or modified triggers after
you have examined and constructed the exit techniques. By necessity (and whether
you like it or not), filters and entries always to s o m e extent try to forecast the market. Exits, on the other h a n d never try to forecast the market, but only tell you
whether this position, as it looks so far, keeps up with what could be expected and
that it doesn't run a m o k with your account. A good exit should also let you use your
entry and filter techniques as often as possible without letting such use affect the
bottom line. T h e normal order of w o r k then, is as follows:

Start out with a basic, stable, and robust set of entries, a n d let the entry for
one position (long, short, or neutral) also signal the exit for any other type of
position.

A d d a set of appropriate, standalone exits that form a good c o m p r o m i s e

between risk and reward.

If necessary, change the entry technique and/or a d d a filter to sift out lowprobability trades.

CHAPTER

Efficient Trades

The idea is to try to m a x i m i z e the efficiency of each trade as m u c h as possible. This

is a concept popularized by R I N A systems. With the highly improved systems performance s u m m a r y in TradeStation 2000i, you can now get a feel for both the entry
and exit efficiencies, as well as the overall efficiency of your trades as R I N A syst e m s a n d O m e g a Research define t h e m .
As an initial m e a s u r e a n d starting point, TradeStation's efficiency calculations
work very well, but if you would like to expand your research and try to optimize
the possibilities these techniques offer, the TradeStation performance s u m m a r y will
no longer be enough. TradeStation again m a k e s the mistake of calculating all moves
in dollar terms, rather than in percentages. In the case of efficiency analysis, this is
OK w h e n looking at the numbers, but not if you would like to do something about
them. Let us take a look at how TradeStation does it.

For Long Trades

Total efficiency = (Exit price Entry price) / (Highest price - Lowest price)
Entry efficiency = (Highest price - Entry price) / (Highest price - Lowest price)
Exit efficiency = (Exit price - Lowest price) / (Highest price - Lowest price)

For Short Trades

Total efficiency = (Entry price Exit price) / (Highest price - Lowest price)
Entry efficiency = (Entry price - Lowest price) / (Highest price - Lowest price)
119

120

PART 3

Getting Out

Exit efficiency = (Highest price Exit price) / (Highest price Lowest price)

For All Trades

Average total efficiency = Sum of total efficiencies / Number of trades
Average entry efficiency = Sum of entry efficiencies / Number of trades
Average exit efficiency = Sum of exit efficiencies / Number of trades
Figure 8.1 shows an individual long trade. In this case we enter the market at
1,350, reach a low at 1,330, and then a high at 1,390, before we exit at 1,380.
Inserting these n u m b e r s in the above formula, we get an entry efficiency of 6 7 % ,
an exit efficiency of 8 3 % , and a total efficiency of 5 0 % . Had this instead been a
short trade, the entry efficiency would have been 3 3 % , the exit efficiency 17%,
and the total efficiency 50%.
N o t e , however, that in the case of a short trade, we would have started out
with a profit, then moved into a substantial loss, before we recaptured s o m e of that
o p e n position loss. What really happened then, was that both the entry and the exit
were quite good. It was the m i d d l e part of the trade that failed. This, however, does
not show up in the performance s u m m a r y and, therefore illustrates the importance
of dividing the trade and especially the d r a w d o w n into several subcategories, as

FIGURE

8.1

An Individual long trade.

CHAPTER 8

Efficient Trades

121

already touched upon in Part 1: the start trade d r a w d o w n ( S T D ) , the end trade
drawdown ( E T D ) , the closed trade drawdown ( C T D ) , and the total equity drawd o w n ( T E D ) . A l m o s t every trade can be divided into three distinct parts. First,
there is the STD, leading down to the low. T h e n there is the actual m o n e y - m a k i n g
phase, leading up to the high. Finally, there is the E T D , leading d o w n to the exit
level and the final profit.
A l t h o u g h this last conclusion is very important, however, it is not the m a i n
reason why these efficiency calculations are good to look at but not to tinker with.
As you can see, all values are expressed as percentages or ratios of the distance
between the low and high values for the trade. B e c a u s e they are ratios, they work
well as comparison measures, b e t w e e n both different markets and different t i m e
frames. B u t if you would like to do something about them, like adding different
exit techniques or add-on entries, you m u s t look at all the price changes and relationships in percentages as well.
For instance, if you would like to add a stop loss to this system, you cannot
calculate that stop in dollar t e r m s , but have to do it in percentage t e r m s . As we discussed, this is because we want all our trades to be affected in the s a m e way,
whether the stop is applied to a m a r k e t that has been trending heavily or to several different m a r k e t s that trade at different levels and point values. With this also
c o m e s the necessity of using the R A D contract that is p u t together using percentage differences rather than point differences, on the day for the roll from one contract to the next.

DRAWDOWNS
A h , d r a w d o w n s ! There is plenty that can be said about this intriguing subject. But
what m a n y traders do not know, is that the drawdown is highly overrated as a syst e m testing decision variable. T r u e , the d r a w d o w n should be neither neglected
completely nor investigated less than thoroughly, but you m u s t know w h a t you are
doing and w h a t you are investigating. For o n e thing, the estimated largest drawdown holds valuable information about h o w large your account size m u s t be and
can give you an indication about whether y o u have the psychological profile to
trade the system in question. Unfortunately, however, the information that can be
derived from m o s t system-testing p a c k a g e s will not be enough, because the n u m ber is given in dollars without any relation to w h e r e and w h e n this b a d sequence
of trades struck you.
Aside from not placing the drawdown in relation to the market situation at
the time, another major error m o s t system designers m a k e when they are building
a n d evaluating systems is only to look at the overall total equity drawdown ( T E D ) .
This is calculated using b o t h the open equity and the already closed out equity on
your account. For instance, say that you, at a specific point in time, d o n ' t have any
positions and that your last trade was a $3,000 winner that also took your account

122

PART 3

Getting Out

to a n e w total equity high at, let's say, $9,000. But before you m a n a g e d to exit the
position, it gave back $1,000 of its open equity.
A c c o r d i n g to m o s t analysis packages, this m e a n s that even t h o u g h you managed to a d d $3,000 to your closed out equity, you n o w are $1,000 in the hole as
c o m p a r e d to your latest total equity high. Furthermore, if your next trade turns out
to be a loser that immediately had you stopped out with a $4,500 loss before, your
closed out equity n o w is $4,500, and your d r a w d o w n is $5,500. Then you have a
$5,000 w i n n e r which, before it took off, started out going the w r o n g way with an
additional $1,000, leaving you $6,500 in the hole. T h e n it gave back $1,500 of the
open profit, and your total equity drawdown would now be a $1,500. Figure 8.2
shows what this sequence of trades could have looked like.
However, if you look closely at these n u m b e r s , they consist of three different
types of drawdowns. At point 3, we are dealing with the end trade drawdown
( E T D ) that tells us how m u c h of the open profit we h a d to give back before we
were allowed to exit a specific trade. At point 4, we are looking at the closed trade
drawdown ( C T D ) that measures the distance between the entry and exit points
without taking into consideration w h a t is going on within the trade. At point 5, we
are dealing with the start trade drawdown (STD) that measures how m u c h the trade
went against us after the entry and before it started to go our way. A n d at point 7,
finally, we again have to do with the E T D .

FIGURE

8.2

A look at different types of drawdown.

CHAPTER 8

Efficient Trades

123

Of course we would like to keep all these d r a w d o w n n u m b e r s as small as

possible. But w h e n only examining the overall drawdown n u m b e r (the T E D ) and
blindly trying to do something about it leaves us with no way of knowing w h a t it
is we're really doing and what is actually changing within the system w h e n we're
m a k i n g any changes to the input parameters.
To be sure, in recent years a few system developers and market analysts have
addressed this issue in various ways, but as far as I k n o w nobody has really nailed it
down when it comes to how one must go about examining the markets and the systems in an appropriate and scientific fashion. One of these analysts is John Sweeney,
technical analysis editor at Technical Analysis of Stocks and Commodities, who, in
his two books Campaign Trading (Wiley Finance Editions, 1996) and Maximum
Adverse Excursion (Wiley Trader's Advantage Series, 1997) came up with the concepts of maximum adverse excursion ( M A E ) and maximum favorable excursion
( M F E ) . M o r e recently, David Stendahl at RINA Systems has taken this a little further and developed a method to calculate the efficiency of a trade.
With version 2000i of O m e g a Research's TradeStation, the efficient trade
analysis also has been implemented in the systems and optimization report. But
the question still s e e m s to r e m a i n open about whether this analysis is necessary
and how one can gain from it. F u r t h e r m o r e , if it is necessary, there still remains
the question of h o w to go about m a k i n g the m o s t out of it.
At any given m o m e n t , the T E D is m a d e up of the STD, E T D , and CTD. This
m e a n s that to c o m e to grips with the T E D , you m u s t c o m e to grips with the STD,
E T D , and C T D .
Depending on what type of entry technique you are using, m a n y of your
trades will experience a start trade drawdown ( S T D ) before they start going your
way. This is especially true for short-term top and bottom picking systems, where
you enter with a limit order. In this case, the only way to avoid an S T D is to enter
at the absolute low or absolute high, and how often will that happen?
For instance, let's simplify the code for our intraday M e a n d e r system slightly to go long on the open. If we open below the VS M i d level and then stay in the
trade all the way to the close the same day, our S T D will be the distance between
the entry point (the open) and the low of the day. Figure 8.3 shows this, with the
S T D charted on the horizontal x-axis and the result of the trade on the vertical yaxis. F r o m this chart we can see that only one trade with an S T D / M A E of m o r e
than or equal to 3% m a n a g e d to produce a slight profit. M o s t of the trades that produced an S T D / M A E over 2% also ended up as losers, while m o s t of the trades
with an S T D / M A E less than 1% turned around and produced a profit. N o t e , however, that in this particular case we are assuming that the low c o m e s before the
high, which is not always the case for a system like this.
L o n g e r - t e r m b r e a k o u t systems often also experience an STD, w h e n they
enter with a stop order on the highest high or lowest low over the lookback period. Usually, this level coincides with a pivotal resistance or support level in the

124

PART 3

FIGURE

Getting Out

8.3

The final profit In relation to the STD/MAE for the intraday Meander system.

market, from which, after this level has been tested a n d y o u ' v e entered the trade,
the m a r k e t usually m a k e s yet another correction before the final penetration and
follow-through takes us away (we h o p e ) in a major trending m o v e . Yet, other
s y s t e m s might completely lack any specific levels around which to place any
natural orders. In our directional slope system, for instance (which b a s e d its
entries on the slope of the m o v i n g average rather than the crossover), getting a
feel for the S T D is crucial, because this system, as it is constructed so far, operates not only completely without any actual levels for w h e n a trade gets triggered,
but also completely without any exit rules as we define t h e m . Figure 8.4 shows
that there is no connection b e t w e e n the level of the price or the m o v i n g average,
a n d the reason w h y the trade is triggered. (Note that in this section we work with
an 18-bar Entry MA a n d a 12-bar Exit M A . )
To start looking for the S T D / M A E and E T D / M F E levels that should work
well on several different markets and time periods, we must do all our research
with the R A D contract. To export the necessary data for the directional slope syst e m into a spreadsheet program, use the following TradeStation code:
Input: EntryMA(18), ExitMA(12);
Vars: EntryAvg(O), ExitAvg(O), LongEntry(O), ShortEntry(O), LongExit(O),
ShortExit(O), LongEntryDate(O), ShortEntryDate(O), LongExitDate(O),

CHAPTER 8

FIGURE

Efficient Trades

8.4

The directional slope system applied to the Japanese yen.

125

PART 3

126

Getting Out

ShortExitDate(O), LongEntryBar(O), ShortEntryBar(O), LongExitBar(O),

ShortExitBar(O), MP(0), TotTr(O), LowestLow(Low), HighestHigh(High),
HighToLow(O), LowToHigh(O), PosProf(0), MAE(O), MFE(O), MFEBar(O),
PrelMAEfe(O), MAEfe(O), FTE(O), ETD(O), BarsForTrade(O), PriceForEntry(O),
PriceForExit(O), DateForEntry(O), DateForExit(O), TradeStrl (""), TradeStr2(""),
TradeStr3("");
Entry Avg = Average(Close, EntryMA);
ExitAvg = Average(Close, ExitMA);
Conditionl = EntryAvg > EntryAvg[l];
Condition2 = ExitAvg > ExitAvg[l];
If MarketPosition <> 1 Then Begin
LongEntry = Close;
LongEntryDate = Date;
LongEntryBar = BarNumber;
End;
If MarketPosition <> 1 Then Begin
ShortEntry = Close;
ShortEntryDate = Date;
ShortEntryBar = BarNumber;
End;
If MarketPosition = 1 Then Begin
LongExit = Close;
LongExitDate = Date;
LongExitBar = BarNumber;
End;
If MarketPosition = 1 Then Begin
ShortExit = Close;
ShortExitDate = Date;
ShortExitBar = BarNumber;
End;
If Conditionl = True and Condition2 = True and MarketPosition = 0 Then
Buy at Close;
If Conditionl = False and Condition2 = False and MarketPosition = 0 Then
Sell at Close;
If Condition2 = False Then
ExitLong at Close;
If Condition2 = True Then

CHAPTER 8

Efficient Trades

127

ExitShort at Close;
MP = MarketPosition;
TotTr = TotalTrades;
If MarketPosition = 1 Then Begin
If BarsSinceEntry = 1 Then Begin
LowestLow = LongEntry;
HighestHigh = LongEntry;
HighToLow = 0;
TradeStr2 = NumToStr(PosProf, 4);
MAE = 0;
MFE = 0;
MFEBar = BarNumber;
MAEfe = 0;
End;
PosProf = (Close LongEntry) / LongEntry;
TradeStr2 = TradeStr2 + '*," + NumToStr(PosProf, 4);
If Low < LowestLow Then Begin
LowestLow = Low;
MAE = (LowestLow LongEntry) / LongEntry;
End;
If High > HighestHigh Then Begin
HighestHigh = High;
MFE = (HighestHigh LongEntry) / LongEntry;
MFEBar = BarNumber;
HighToLow = Lowest(Low, (MFEBar MaxList(MFEBar[ 1 ], LongEntryBar)));
PrelMAEfe = (HighToLow - HighestHighfl]) / HighestHighfl];
If PrelMAEfe < MAEfe Then
MAEfe = PrelMAEfe;
End;
End;
If MarketPosition = 1 Then Begin
If BarsSinceEntry = 1 Then Begin
LowestLow = ShortEntry;
HighestHigh = ShortEntry;
LowToHigh = 0;
TradeStr2 = NumToStr(PosProf, 4);

PART 3

128

Getting Out

MAE = 0;
MFE = 0;
MFEBar = BarNumber;
MAEfe = 0;
End;
PosProf = (ShortEntry - Close) / ShortEntry;
TradeStr2 = TradeStr2 + "," + NumToStr(PosProf, 4);
If High > HighestHigh Then Begin
HighestHigh = High;
MAE = (ShortEntry - HighestHigh) / ShortEntry;
End;
If Low < LowestLow Then Begin
LowestLow = Low;
MFE = (ShortEntry - LowestLow) / ShortEntry;
MFEBar = BarNumber;
LowToHigh = Highest(High, (MFEBar MaxList(MFEBar[ 1 ], ShortEntryBar)));
PrelMAEfe = (LowestLow[l] LowToHigh) / LowestLow[l];
If PrelMAEfe < MAEfe Then
MAEfe = PrelMAEfe;
End;
End;
If TotTr > TotTrfl] Then Begin
I f M P [ l ] = 1 Then Begin
PriceForEntry = LongEntryfl];
PriceForExit = LongExit[l];
FTE = (PriceForExit - PriceForEntry) / PriceForEntry;
ETD = (PriceForExit - HighestHigh[ll) / Highesthigh[l];
DateForEntry = LongEntryDatefl];
DateForExit = LongExitDate[l];
BarsForTrade = LongExitBar[ 1 ] - LongEntryBar[l];
If MAEfe[l] > (PriceForExit - HighestHigh[l]) / HighestHigh[l]
Then
MAEfe = (PriceForExit - HighestHigh[l]) / HighestHighfl];
End;
I f M P [ l ] = -1 Then Begin
PriceForEntry = ShortEntry[1];

CHAPTER 8

Efficient Trades

129

PriceForExit = ShortExit[l];
FTE = (PriceForEntry - PriceForExit) / PriceForEntry;
ETD = (LowestLowfl] PriceForExit) / LowestLow [1];
DateForEntry = ShortEntryDatefl];
DateForExit = ShortExitDate[l];
BarsForTrade = ShortExitBarfl] - ShortEntryBar[l];
If MAEfe[l] > (LowestLow[l] - PriceForExit) / LowestLow[l] Then
MAEfe = (LowestLow[l] PriceForExit) / LowestLow[l];
End;
If FTE < MAE[1] Then
MAE = FTE Else MAE = MAE[1];
If FTE > MFE[1] Then
MFE = FTE Else MFE = MFE[1];
TradeStrl = LeftStr(GetSymbolName, 2) + "," + NumToStr(MP[l], 0) +
","

+ NumToStr(DateForEntry, 0) + "," + NumToStr(PriceForEntry, 4) +

"," + NumToStr(DateForExit, 0) + "," + NumToStr(PriceForExit, 4) +

"," + NumToStr(MAE, 4 ) + "," + NumToStr(MFE, 4) +

NumToStr(FTE, 4) + "," + NumToStr(ETD, 4) + "," +

NumToStr(MAEfe, 4) + "," + NumToStr(BarsForTrade, 0);
TradeStr3 = TradeStrl + "," + TradeStr2[l] + NewLine;
FileAppend("D:\Temp\DSS.csv",TradeStr3);
End;
Unfortunately, I am not the best p r o g r a m m e r around, which is probably
revealed by the code above. However, its length is also because of one of the stupidest features in TradeStationthe way the p r o g r a m registers an open trade and
its entry price a n d counts the n u m b e r of bars in trade. TradeStation does not note
that a trade is o p e n until after the first full bar in the trade and, consequently, does
n o t calculate the n u m b e r of bars in the trade correctly. For instance, apply the following c o d e to any market:
Buy at Close;
ExitLong ("Loss") tomorrow at EntryPrice * 0.98 stop;
ExitLong ("Profit") tomorrow at EntryPrice * 1.02 limit;
T h e result is that TradeStation will exit each trade at the o p e n of the next bar,
no matter what. B u t if you change the code to the following:
Buy at Open;
ExitLong ("Loss") tomorrow at EntryPrice * 0.98 stop;
ExitLong ("Profit") tomorrow at EntryPrice * 1.02 limit;

PART 3

130

Getting Out

T h e result is that TradeStation enters a n d exits each trade on the very same
open, leaving you with a b u n c h of zero-bar, zero-profit trades. T h e s e , by the way,
are all counted as winners in the performance summary. I do not k n o w about you,
but to me a w i n n e r is a trade that ends up with a profit; everything else is a loss.
O n e way to work around this is to rewrite the code to:
Buy at close;
If BarsSinceEntry > = 1 Then Begin
ExitLong ("Loss") tomorrow at EntryPrice * 0.98 stop;
ExitLong ("Profit") tomorrow at EntryPrice * 1.02 limit;
End;
B u t then TradeStation simply skips a bar before it starts to track the trade and
does not let you exit the trade until bar three (counting the b a r for the entry as bar
one), which w o n ' t c h a n g e even if you decide to enter at the open instead of the
close. A n d setting the BarsSinceEntry function equal to or greater than zero does
not work either.
Another example of the same flaw is when you, for instance, enter a trade intrabar on a breakout. You will not be able to exit on the close, or intrabar of that same
bar, no matter h o w m u c h the market moves against you, as suggested by the closing
price of that same bar. (That is why I have chosen to do all the research for the intraday version of the Meander system within Excel, rather than in TradeStation.)
To m e , this is the equivalent of Excel's purporting to calculate everything in
a spreadsheet correctly, without Microsoft's letting you k n o w that sometimes,
w h e n the formula get too complicated, it loses track a n d simply m a k e s up an
answer. After all, it is my (your) m o n e y we are dealing with here and we should all
expect things to work properly. Unfortunately, for that to happen we simply must
find a way around the problem. Hence, we end up with a large c h u n k of extra
spaghetti code, which obviously increases the risk for other types of errors and
logical blunders. E n o u g h said, let us work with what we have.
O n c e we are in the spreadsheet p r o g r a m it is easy to create a scatter chart like
Figures 8.3 a n d 8.5 that compare the S T D to the final profit. Although we might
be using the same techniques to derive the necessary data for the S T D as for the
M A E , it is important to understand that the two are not the same. N o r is either one
of t h e m the s a m e as R I N A S y s t e m s ' entry efficiency. T h e difference is that the
S T D should primarily be adjusted w i t h the entry technique, while the M A E should
primarily be adjusted with a specific exit technique, such as a stop loss. Together,
the M A E and S T D m a k e up the entry efficiency.
Notice also the word "primarily," because in a g o o d working system all parts
should and will intervene to form a w h o l e larger than the parts. M a n y times, it can
be difficult to distinguish between the two. T h e point is that you should at least be
aware of the differences and exactly what it is you are trying to achieve. It is still
your m o n e y we are talking about here.

CHAPTER 8

Efficient Trades

FIGURE

131

8.5

The final profit in relation to STD/MAE for the directional slope system.

Figure 8.5 shows the S T D / M A E for the directional slope system traded on
the Japanese yen. This chart shows that of all trades with an S T D / M A E of m o r e
than or equal to 4% only two did not end up as losers, while most of the trades with
a n S T D / M A E o f 2 % o r less e n d e d u p a s winners.
In the case of the M e a n d e r system, after you have m a d e sure that you have
optimized your exit techniques and taken every other step necessary, perhaps all
that m u s t be d o n e to the entry is to wait a couple of ticks before you enter, or at
least not enter immediately at the o p e n w h e n the volatility can be high. In the case
of the directional slope system, o n e solution could be to wait for a pullback and
then enter on a b r e a k through the price level that causes the Entry MA to change
its direction.
Just as the S T D is not the same as the M A E , the end trade drawdown ( E T D )
is not the s a m e as the m a x i m u m favorable excursion ( M F E ) , although we might be
using essentially the s a m e techniques to derive the necessary information. M F E is
the m a x i m u m open profit of the position. E T D is the difference between the M F E
a n d the exit point, and the amount you are giving back to the market before the syst e m signals the time to exit. T h e M F E and the final profit should therefore be m a x imized with the amount invested and the n u m b e r s of contracts traded. T h e ETD, on
the other hand, should be managed with different types of exit techniques and stops,

132

PART 3

Getting Out

such as a trailing stop, time-based stop, or profit target stop. The only way to completely avoid any E T D is to exit all trades with a limit order that is triggered at the
top of the equity curve. Figure 8.6 shows the relationship between the E T D and the
final profit for the simplified M e a n d e r system, while Figure 8.7 shows the relationship between the M F E and the final profit for the directional slope system.
From Figure 8.6, we can see that an E T D of m o r e than 0 . 5 % very often resulted in
a loss greater than 1%. This happens when the trade goes immediately against us,
and the S T D and E T D b e c o m e s the same. With a stop loss or trailing stop at an
appropriate level (to address E T D ) , or a m o r e conservative entry technique (to
address S T D ) most of these losses could have been reduced considerably.
Figure 8.7 illustrates the final profit in relation to the M F E for the directional
slope system traded on the Japanese yen. It shows that not until we get an M F E above
approximately 5% does the system manage to signal an exit before we have given away
all the open equity. Looking at Figures 8.6 and 8.7, it is not too difficult to see that it
is easier to handle the information from Figure 8.6. Sure, we realize that something
needs to be done somewhere around the 5% level for the directional slope system, but
what? To get a better feel for this, we must look at the E T D in relation to the MFE
instead of in relation to the final profit. This has been done in Figures 8.8 and 8.9.
Because this modified version of the M e a n d e r system operates with a
d y n a m i c profit target that triggers an exit with a limit order, the higher the M F E ,
the lower the E T D . Figure 8.8 shows that for all trades with an M F E above 0.5%,

FIGURE

8.6

The final profit in relation to ETD for the Meander system.

CHAPTER 8

133

Efficient Trades

FIGURE

8.7

The final profit in relation to the MFE for the directional slope system.

FIGURE

8.8

The ETD in relation to MFE for the Meander system.

134

PART 3

Getting Out

however, there are still several that have a larger E T D than M F E . With a profit protection or breakeven stop, m a n y of these losses can be avoided.
As is the case with so m a n y other long-term trend-following systems, the
directional slope system s e e m s to give away a substantial part of the profit before
it allows us to exit. For instance, Figure 8.9 shows that both trades with an M F E
immediately above 2 0 % did not allow us to exit until the market had taken back
about a third of our open profit. N o t e that in this case the E T D is denoted with
negative values.
T h e closed trade drawdown ( C T D ) is the drawdown you experience on your
account from one trade to another for all your losing trades. It also is important to
distinguish the C T D from the T E D , as the differences can be quite substantial,
especially for a longer-term trend-following system that is p r o n e to give back a
substantial part of the open profit for a trade before it allows you to exit. Because
m a n y times the C T D also is the s a m e as the STD, you m u s t c o m e to t e r m s with
the C T D by limiting the n u m b e r of trades with an overriding filter, which lets
through only those setups and long-term market situations that show the highest
likelihood of producing a w i n n i n g trade, given the entry trigger you use.
Especially for long-term trend-following systems, the C T D also can be a function
of the ETD, which m e a n s it also should be matched with an appropriate exit tech-

FIGURE

8.9

The ETD in relation to MFE for the directional slope system.

CHAPTER 8

135

Efficient Trades

nique, such as a trailing stop. To get a feel for the C T D , you can chart the final
profit, separated into w i n n i n g and losing trades, against both the S T D / M A E , and
against the M F E for all losing trades. Figure 8.10 shows that if a trade within the
directional slope system turned out to be a loser, only on a few occasions w a s the
final p r o f i t / C T D significantly different from the S T D / M A E . This suggests that
w h e n a trade goes bad, m o s t of the time it does so right off the bat. Figure 8.11
shows that for the w i n n i n g trades, the relationship between the final profit and
S T D / M A E isn't as clear-cut as for the losers, w h i c h indicates that to c o m e to grips
w i t h the C T D , it is important to analyze it separated from w i n n i n g trades, a n d not
clutter t h e analysis w i t h u n n e c e s s a r y information.
From Figures 8.10 and 8.11, we also can see that of all winning trades, only
six had an S T D of m o r e than 2 % , of which only three m a n a g e d to produce a profit of m o r e than 10%. At the other e n d of the spectrum there were 16 losing trades
with an S T D of m o r e than 2 % , which in m o s t instances also produced a losing trade
of m o r e than 2 % . In fact, if the trade turned out to be a loser, m o r e often than not,
it also m a n a g e d to be closed out very close t o , or at, the lowest low of the trade.
T h e s a m e p h e n o m e n o n also can be seen in Figures 8.12 a n d 8.13, w h i c h
show the C T D respectively, the final profit for the M e a n d e r system, in relation to
the S T D / M A E . As you can see from Figure 8.12, if the trade turned out to be a
loser, only on a few occasions w a s the final p r o f i t / C T D significantly different

FIGURE

8.10

The final profit for all losing trades in relation to STD/MAE for the directional slope system.

PART 3

136

FIGURE

Getting Out

8.11

The final profit for all winning trades In relation to STD/MAE for the directional slope system.

from the S T D / M A E . This suggests that w h e n a trade goes bad, m o s t of the time it
does so right off the bat. F r o m Figure 8.13 you again can see that the similarity
between the long-term directional slope system and the short-term M e a n d e r system is striking. T h e relationship between the winning trades and their S T D / M A E
aren't as clear-cut as for the losers, which confirms that to c o m e to grips with the
C T D it is important to analyze it separated from winning trades, with no unnecessary information.
F r o m Figures 8.12 and 8.13 we also can see that very few trades with an STD
of m o r e than 2% m a n a g e d to p r o d u c e a profit above 0 . 5 % , and only one came
close to 2 % . On the other hand, there are plenty of trades with an S T D of more
than 1% that also produced a final loss greater than 1%. A n d just as w a s the case
for the directional slope system, m a n y of these trades seem to have been closed out
at or very near their lowest open equity. T h e question is, what was the current market situation like to produce these bad trades, and could this have been dealt with
before entering the trade in the first place?
W h e n we are comparing the final profit for the trades within the directional
slope system with the M F E we can see that there are a few trades that seem not to
be stopped out until there has been an adverse move in the neighborhood of 5% or
more. Obviously, this is too m u c h to give back, which m a k e s it a good idea to

CHAPTER 8

Efficient Trades

FIGURE

137

8.12

The CTD in relation to STD/MAE for the Meander system.

FIGURE

8.13

The final profit for all winning trades in relation to STD/MAE for the Meander system.

PART 3

138

Getting Out

e x a m i n e the possibility of adding s o m e form of a trailing stop. Figure 8.14 also

shows that a total of 17 out of 24 losing trades for the directional slope system only
m a n a g e d to produce an M F E of 2% or less. With a good working filter m a n y of
these trades could have been weeded out as well. Again, this indicates that most
losers go b a d from the point of inception or very soon thereafter. This indicates
that w h e n a trend gets hold of the market, it is likely to be persistent and lead to a
profitable trade. Also, the larger the M F E , the smaller the final loss.
T h e s a m e holds true for the M e a n d e r system in Figure 8.15, which has quite
a few trades that start out with a profit but then allow for an adverse move of more
than 1.5% or m o r e before w e ' r e allowed to exit on the close of the bar. There also
are plenty of trades for which m o s t of the losers failed to p r o d u c e an M F E of more
than 0.2%). In the former case, we could have avoided some of the losses with a
stop loss of s o m e sort. In the latter case, m a n y trades could probably have been
avoided completely with some sort of a filter. Again, the larger the M F E , the
smaller the final loss. A l s o , whether we're trading long t e r m or intraday, when a
trend gets hold of the market it is likely to be persistent and lead to a profitable
trade, no matter in w h a t time perspective w e ' r e trading.
A n o t h e r way to s u m m a r i z e the entire life s p a n of a trade is to chart the
S T D / M A E , the M F E , and the E T D / f i n a l profit in o n e s e q u e n c e , as they are
a s s u m e d to h a v e taken place. F i g u r e 8.16 attempts to s h o w the s u m m a r i z e d life

FIGURE

8.14

The CTD in relation to MFE for the directional slope system.

ER 8

Efficient Trades

FIGURE

139

8.15

The CTD in relation to MFE for the Meander system.

for a trade in the directional slope system. T h e a s s u m p t i o n we m a k e is that the

further we get from the point of origin, the m o r e divergent the p a t h s the trade
could have taken, w h i c h leads to a h i g h e r standard deviation of the o u t c o m e s .
Therefore, it is r e a s o n a b l e to a s s u m e that, in the case of the directional slope system, m o s t M F E s h a p p e n e d after the S T D s . T h i s is not entirely correct b e c a u s e
not all t r a d e s register their M F E s after their M A E s . B u t a s s u m i n g this is the
c a s e , we can see from Figure 8.16 that the average trade for the directional slope
s y s t e m on the J a p a n e s e yen first starts out by going against us with about 2%
(the m i d d l e line), before it takes off in the right direction to register an M F E of
about 7 . 5 % . F r o m then on, however, it is downhill again, before we are stopped
out for a final profit of approximately 3 % , after we have given b a c k m o r e than
5 0 % o f the o p e n profit.
T h e s a m e p h e n o m e n o n can be seen again in Figure 8.17, which shows the
summarized life of a trade in the M e a n d e r system. It is reasonable to assume that
the further we get from the point of origin, the m o r e divergent paths the trade could
have taken, which leads to a higher standard deviation of the outcomes. This seems
not to be the case for the M e a n d e r system, which m e a n s that we have to be a little
m o r e careful interpreting our S T D and E T D findings. Figure 8.17 shows that the
average trade starts out with a wrong move of approximately 0.6% before things
start to go our way, and we can register an M F E of about 0 . 6 % and a final profit of
about 0 . 1 % . Again, however, this assumes that the course of events is as outlined.

140

PART 3

FIGURE

8.16

The summarized life for a trade in the directional slope system.

FIGURE

8.17

The summarized life for a trade in the Meander system.

Getting Out

CHAPTER 8

Efficient Trades

141

That this is not always the case is noticeable in the standard deviation interval (the
upper and lower lines), which are likely to m o v e further apart from each other the
further away from the entry point we get; this does not happen in this chart.
Nonetheless, individually examining each point, with its standard deviation
interval, also can give us some clues to where and how to place our stops. For
instance, in the case of the directional slope system we can expect that 6 8 % of all
trades should produce an S T D / M A E somewhere between - 3 . 7 to 0 % , an M F E in
the interval -2 to 17%, and a final profit between - 5 . 5 and 11.5%. Trades that do
not match these criteria could then be assumed to be better or worse than acceptable. R e m e m b e r from Part 2 that a trade that is too good to be true is not necessarily a good thing. Different methods for handling these trades, as well as better ways
to depict the life of a trade, are dealt with in the next chapter, where we take a closer look at Sweeney's M A E and M F E . Although the old adage "your worst drawd o w n is still to c o m e " is likely to c o m e true sooner or later, it does not have to h a p pen first thing tomorrow, provided you have done your homework correctly.

CHAPTER

Sweeney's MAE/MFE

Look at it this way: W h e t h e r your trading style resembles a gunslinger shooting

from the hip or a well-aimed sniper lying in ambush, k n o w i n g where your trades
are heading could m e a n the difference b e t w e e n riding into the sunset or lying fatally w o u n d e d on a dusty street at high n o o n . To stay alive you m u s t k n o w when to
draw a n d when to run.
For a gunslinger or a sniper, the ultimate advantage c o m e s from being able to
calculate the trajectory of the bullet. Although this hardly is possible in a shootout
that d e m a n d s split second decisions, it can be d o n e in the financial markets where
you can estimate the trajectory of your trades in relation to their targets.
T h e first thing you m u s t do, is find w h e r e your targets have been and how you
have performed in the past. To do this you must take a deep look into your population of trades and examine each individual trade closely. W h e n you do, you will discover that there are several similarities between them, but that every trade also has
a unique set of characteristics and traits that distinguish the winners from the losers. All of these characteristics often can be spotted w h e n you trade using John
Sweeney's M A E and M F E techniques, as illustrated in Figure 9 . 1 . This chapter
illustrates one way of interpreting and m a k i n g better use of Mr. Sweeney's excellent work.
The M A E is defined as the most negative intraday price movement against your
position, which corresponds to the lowest open equity during the lifespan of the trade.
The M A E does not take into consideration whether this happens immediately at the
inception of the trade, or just before exiting, after the trade might have lost all its open
profit and turned into a loser. The M F E is defined as the most positive price movement
for your position, which corresponds to the highest open equity, during the lifespan of
143

144

FIGURE

PART 3

9.1

MAE and MFE levels for two trades in the Japanese yen.

Getting Out

CHAPTER 9

Sweeney's MAE/MFE

145

the trade. The M F E does not take into consideration whether this happens early on in
the trade or just before exiting with or without a profit. Notice, for instance, that in
Figure 9 . 1 , both M A E s occurred after the M F E s , and for the short trade also in conjunction with the exit of the trade. If we look at the long trade and refer back to the
different drawdowns just discussed, the distance between the entry and the M A E level
would in fact make up the S T D and as such should preferably have been avoided completely with the help of some sort of a filter. But once in the trade anyway, the adverse
move following the M F E should have been limited by a trailing stop or a stop loss.
With the M A E and M F E techniques, we can examine our systems for high-probability exit techniques and even for situations when it could be a good idea to add to a
trade, or even fade (to trade against the trend) a previously given signal.
As a c o m p l e m e n t and a starting point to the M A E / M F E analysis we also take
a closer look at the entire lifespan and development of a trade. (A sort of trajectory analysis if you like.) This helps us examine our systems for additional trading
rules, depending on whether the trade stays within the boundaries set by the
M A E / M F E analysis. As an analogy, albeit with a little backwards reasoning, c o m pare the M A E / M F E values with the pop-up targets at a shooting range. In this
case, the M F E targets are the bad guys to be aimed at, while the M A E targets are
the "innocent p e o p l e " w h o should be avoided.
T h e trajectory analysis helps us determine if we should "freeze" the bullet
(the trade) mid-air if we mistakenly shoot at one of the good guys (or even take a
shot in the completely opposite direction), or if it could be a good idea to throw
away a couple of additional shots in the same direction w h e n we have o n e of the
bad guys in our sights. Additionally, with the trajectory analysis, each curve of the
open equity is a mirror image of the price development of the market, which
m a k e s it possible to compare each trade's ongoing price action with what actually
happened during similar situations in the past. If technical analysis is the study of
the price action from the past on the assumption that it will repeat itself in the
future, isn't this the ultimate, practical, and beneficial form by which this can be
done? To export the entire necessary data into a spreadsheet, we continue to use
the code from the previous chapter.
Let us apply this code to all markets, over the time period January 1980 to
October 1999, in our directional slope system from Part 2. T h e markets tested are
T-bonds, live cattle, Japanese yen, corn, Canada dollar, crude oil, dollar index,
lumber, orange j u i c e , S & P 500, copper, Eurodollar, C R B index, cotton, gold, and
coffee. If you have this workspace saved already, do not forget to change from the
ordinary point-based back-adjusted contract to the R A D contract. And, as before,
do not deduct any m o n e y for slippage and commission, because this will only
result in a suboptimal solution.
In this case, we start by looking at all trades (winners and losers, longs and
shorts) for all markets, and then divide our analysis into winners and losers, respectively. O n c e in your spreadsheet program, m a k e three copies of the spreadsheet. If

PART 3

146

Getting Out

you also would like to examine short and long trades separate from each other, it is
r e c o m m e n d e d that you m a k e separate copies for this as well. With the data from the
spreadsheet, we can create a chart similar to Figure 9.2, which shows the development for a set of trades.
Although just charting all trades on top of each other m a k e s little sense and
looks m o r e like the inside of a magpie's nest, already at this point we can distinguish a few differences between the winners and the losers. For o n e thing, it is fairly easy to see that the life span for several of the losers is fairly short, while the winners seem to live on for much longer. There also seem to be a bias (trend) to the
upside, the longer the trade lasts. To emphasize and m a k e it easier to see this tendency we can channel in the entire bouquet, as in Figure 9.3. Although we already
can see s o m e tendencies and form some basic guidelines, like "a winner should be
in positive territory, approximately by bar 1 8 " or "around bar 30, most trades
should show a profit between 10 to 3 0 % , " this is not enough to formulate any
systematic and mechanical rules that m o v e us away from "trading with the gut."
To be able to do this, we must simplify things. T h e first thing to do is calculate the average percentage m o v e for each bar, together with its standard deviation
boundaries. At the b o t t o m of the spreadsheet, type in the following formula and
fill t h e m out to the right for as long as you have any open trades:

FIGURE

9.2

The development for a set of trades for the directional slope system.

CHAPTER 9

147

Sweeney's MAE/MFE

FIGURE

9.3

The development for a set of trades with manually added trend lines.

=AVERAGE(H1:H1240)
w h e r e H denotes the column w h e r e we stored the percentage move for bar o n e .
= H 1 2 4 6 + S T D E V(H 1 :H 1240)
=H1246 - STDEV(H 1: H1240)
W h e n d o n e , y o u s h o u l d be able to p r o d u c e a chart like Figure 9.4, w h i c h
s h o w s the d e v e l o p m e n t of the average trade (the m i d d l e squiggly line) together
w i t h the p l u s (upper line) a n d m i n u s (lower line) standard deviation b o u n d a r i e s .
T h e s t a n d a r d deviation b o u n d a r i e s d e n o t e the area in w h i c h 6 8 % of all trades
can be expected to fall. To this y o u also could a d d a least square regression line
for t h e average trade. Instead of l o o k i n g at h u n d r e d s of lines that m a k e little
sense, we are n o w d o w n to four, m a k i n g for m u c h easier interpretation and rule
making.
T h e further away from the entry point we get, m o r e and the m o r e trades are
closed out. However, the fewer the trades, the less reliable our results. T h e vertical
dashed line m a r k s out the point w h e r e we have fewer than 20 open trades. In this
case, this h a p p e n s at b a r 4 8 . To this chart, I also have added the regression line
(middle straight line) for the average trade. T h e regression equation for this line is

148

PART 3

FIGURE

Getting Out

9.4

The development of the average trade (the middle squiggly line) together with the plus (upper line) and
minus (lower line) standard deviation boundaries.

0.0074X, which m e a n s that the average trade increases in value by approximately

0 . 7 4 % per day.
With a chart like this we can start to set up some m o r e definite guidelines
like "if we cross the average line from below, the trade is about to b e c o m e better
than average and can be worth adding t o " or "if we cross the u p p e r standard deviation line from below the trade is too good to be true and we h a d better take some
profit." If we also divide all the trades into groups of winners and losers, we get
an even better feel for our rule making. Figure 9.5 shows the development of the
average losing trade, which s e e m s to go nowhere fast. T h e dashed line at bar 14
m a r k s out the day after which we have less than 20 trades still alive. This is too
few to be able to m a k e any statistical conclusions. If a trade is not at least marginally profitable at this time, your m o n e y is probably better used elsewhere.
As you can see, the average loser hardly m a n a g e s to produce an open profit
of m o r e than 2 % , and no m o r e than an estimated 1 6 % of all losers ((1 - 0.68)/2)
m a n a g e to produce an open profit of m o r e than approximately 7%, as indicated by
the u p p e r 1 standard deviation line around bar 16. Together, Figures 9.4 and 9.5
can give rise to guidelines such as "if we're not profitable by b a r 14, the trade is
probably a loser and should be ditched."

CHAPTER 9

149

Sweeney's MAE/MFE

FIGURE

9.5

The development of the average losing trade.

By the way, Figure 9.4 also supports our a s s u m p t i o n from the previous
chapter about the typical look of a trade, with the M A E occurring before the
M F E . W h a t this looks like for all of our 1,240 trades in the directional slope syst e m can be seen in Figure 9.6, w h i c h also gives us s o m e additional insight into
the system, like "a trade that quotes an M A E b e l o w 7 . 5 % (which is the lower 1
standard deviation b o u n d a r y for the M A E ) is t o o far away from the average trade
and therefore b o u n d to be a loser." T h e standard deviation interval is narrower for
the exit than for the point of the M F E because the exit for m a n y losers h a p p e n s
m u c h sooner after the entry than t h e t i m e it takes for m o s t w i n n e r s to p r o d u c e an
M F E reading in the first place. B e c a u s e of the short distance traveled from the
entry for m a n y of these trades, the dispersion of the o u t c o m e s has not yet had a
c h a n c e to b e c o m e that large. By n a r r o w i n g d o w n our analysis as m u c h as this, we
n o w can s u m m a r i z e the life of our trades using only three lines with only four
data points each.
T h a t there definitely is a difference b e t w e e n winners and losers can be seen
from Figures 9.7 a n d 9.8, w h i c h show the same thing as Figure 9.3, but this time
with the w i n n e r s and the losers separate from each other. T h e m o s t obvious thing
w h e n c o m p a r i n g these charts is h o w difficult it is for the losers to take off in any
direction, while the winners, on the other hand, clearly have a u p w a r d bias; there

150

PART 3

F I G U R E

Getting Out

9.6

The summarized life for the average trade in the directional slope system.

are plenty m o r e winners taking off to higher grounds than there are losers taking
the express elevator to the basement. In fact, very few of the losers really take off
in any direction, and those that m a n a g e to m o v e into positive territory quickly end
their lives with a swift m o v e in the w r o n g direction. As for the winners, we can see
that m o r e often than not, they quickly take off with a bold m o v e in the right direction. Too often, however, they give back way too m u c h of the open profit before
the system allows us to exit with s o m e of the profits. A n o t h e r compelling difference is the lengths of the trades. Although most w i n n e r s last well above 30 bars,
m o s t losers do not last half that long. In essence, a winning trade usually starts out
very well right off the bat and continues to develop favorably for quite some time,
while m o s t losers usually go nowhere fast.
The purpose of the M A E / M F E analysis is to find those points in the trade
where, with the highest likelihood based on previous experience, we can estimate the
outcome of the trade, if we just let it run without any stops or exits. To do this we
must perform the same type of analysis as we did in the previous chapter about different types of drawdowns. For the directional slope system this may resemble the
charts in Figures 9.9 and 9.10 which show the final profit from all trades in all 16
markets (1,240 trades in total) in relation to each trade's M A E and M F E readings,
respectively. Figure 9.9 shows that the least square regression line for this M A E

CHAPTER 9

Sweeney's MAE/MFE

FIGURE

9.7

The development for a set of losing trades.

FIGURE

9.8

The development for a set of winning trades.

151

152

PART 3

Getting Out

analysis hints that the larger the M A E , the larger the loser. T h e true relationship,
however, seems m o r e to resemble an exponential one, as suggested by the handdrawn, dotted trend line. That is, as long as the M A E stays above a certain level, the
final outcome of the trade is very likely to be profitable. T h e least square regression
line for the M F E analysis in Figure 9.10 hints that the larger the M F E , the larger the
winner. This also can be seen from the regression equation, which shows that the
final profit can be expected to reach approximately 5 7 % of the value of the MFE.
That is, the directional slope system regularly gives back as m u c h as 40 to 5 0 % of
the m a x i m u m open profit in the trade before it allows us to exit. Surely, there must
be a way to keep more of that open profit?
F r o m Figure 9.9 we can see that only a handful of trades with an M A E larger than 1 0 % m a n a g e d to end up as winners. By setting a stop loss at 10%, we
would have m a d e all these trades into 1 0 % losers. T h e exact point where the cumulative value of the losers will be greater than that of the w i n n e r s can be calculated
by subtracting the value of all stopped-out losing trades from the value of all winning trades. As you tighten the stop, this value eventually gets below zero, which
is w h e n the value of losing trades has gotten larger than the value of all winning
trades. H e n c e , you are no longer gaining anything by tightening the stop further.
To do this, type in the following formula at the b o t t o m of your spreadsheet:

FIGURE

9.9

The final profit in relation to MAE for 16 markets in the directional slope system.

CHAPTER 9

153

Sweeney's MAE/MFE

FIGURE

9.10

The final profit in relation to MFE for 16 markets in the directional slope system.

=SUMIF(G$1:G$1240;"< = -0.005";I$1:I$1240)/
COUNTIF(G$ 1 :G$ 1240;" < = - 0.005")
w h e r e column G denotes the value for the M A E , and column / denotes the value
of the final profit.
N o t e that this is the arithmetically calculated value. Just calculating the M A E
point where the losers are likely to b e c o m e larger than the winners is not enough (in
this case the point will happen with an M A E larger than or equal to 1.5%). The next
step is to calculate the difference between what would have happened had these
trades been stopped out and what would have happened had they not. Obviously, had
they been stopped out at the M A E level, the average profit for all those trades would
have equaled that level as well. Therefore, to calculate the difference between the
results at a specific M A E level, simply rewrite the above formula to:
=SUMIF(G$ 1 :G$ 1240;" < = - 0.005 ";I$ 1 :I$ 1240)/
COUNTIF(G$ 1 :G$ 1240;" < = - 0.005")+0.005
Using this method, you can put together a table resembling Table 9.1, which
shows the differences in the average profits from all trades corresponding to a certain
M A E , depending on whether or not you have assumed the trades to be stopped out.

154

PART 3

TABLE

Getting Out

9.1

MAE levels, original profits and differences in outcome for 16 markets in the directional slope system.
MAE

Original profit

Difference

-0.50%

0.84%

1.34 points

-1.00%

0.25%

1.25 points

-1.50%

-0.24%

1.26 points

-2.00%

-0.56%

1.44 points

-2.50%

-0.95%

1.55 points

-3.00%

-1.05%

1.95 points

-3.50%

-1.54%

1.96 points

-4.00%

-1.66%

2.34 points

-4.50%

-2.47%

2.03 points

-5.00%

-3.07%

1.93 points

-5.50%

-3.75%

1.75 points

-6.00%

-3.94%

2.06 points

-6.50%

-4.08%

2.42 points

-7.00%

^1.16%

2.84 points

-7.50%

-4.65%

2.85 points

-8.00%

-5.29%

2.71 points

-8.50%

-6.12%

2.38 points

-9.00%

-6.53%

2.47 points

-9.50%

-6.69%

2.81 points

-10.00%

-6.74%

3.26 points

With these values in place, we now can compare the original final profits for
all trades corresponding to a certain M A E level to the s a m e set of trades as if they
h a d been stopped out at that level. For instance, for all trades with an M A E of at
least 3% the arithmetically calculated average profit for all trades would have been
1.05%, but with the stop it obviously would have been 3.0%. Hence, we would
have done worse with a 3% stop loss than without. But, as can be seen from Table
9 . 1 , in this case, no matter where you place the stop you are worse off than had
you not used a stop at all. An interesting observation is that generally the larger the
M A E , the larger the percentage point difference a m o n g the results from using the
stop w h e n c o m p a r e d to not using the stop.
Such is the nature of this g a m e . By m a k i n g sure that you know exactly how
m u c h you are expected to lose in a worst-case scenario you decrease the standard
deviation of outcomes for the system and increase your feeling of certainty and
well-being for that particular trade. T h e price you pay for this is the higher likelih o o d for a lower accumulated result over a longer sequence of trades.
As Table 9.2 shows you, this holds true at the other end of the spectrum as
well. That is, by taking profits w h e n a trade has reached a certain level we once

CHAPTER 9

155

Sweeney's MAE/MFE

again lower the standard deviation of o u t c o m e s , m a k i n g the system "less risky."

T h e price we pay is a lower accumulated profit over a longer sequence of trades,
no matter where we place the profit target level.
" S o w h a t is the point w i t h all t h i s ? " you m i g h t ask. T h e point is, by keeping
the standard deviation of the outcomes lower, while at the s a m e time also decreasing the ratio b e t w e e n the open a n d closed out profits, we can be m o r e aggressive
w h e n it c o m e s to the n u m b e r of contracts traded in our portfolio, according to an
optimal / or fixed fractional trading strategy.
So, w h e r e to place the stops, if we even should have any? That is pretty m u c h
a question only you can answer depending on h o w you evaluate the risk/reward
relationships between different stop or profit target levels for the particular system
on w h i c h you currently are working. In this case, we m a k e things as " b a c k w a r d s "
as possible and place a stop loss at 4% a n d a profit target at 6 0 % , simply because
those are the levels (within reason) w h e r e we lose the m o s t in the long run, as suggested by Tables 9.1 and 9.2. This should also m e a n that these levels will help us
gain the m o s t w h e n we apply the system to a fixed fractional m o n e y m a n a g e m e n t
strategy.

TABLE

9.2

MFE levels, original profits, and differences for 16 markets In the directional slope system.
MFE

Original profit

Difference

0.05

8.08%

3.08 points

0.10

12.93%

2.93 points

0.15

18.72%

3.72 points

23.98%

3.98 points

0.25

30.93%

5.93 points

0.30

36.45%

6.45 points

0.35

43.62%

8.62 points

0.40

50.34%

10.34 points

0.45

60.44%

15.44 points

0.50

63.98%

13.98 points

0.55

71.44%

16.44 points

0.60

77.94%

17.94 points

0.65

80.47%

15.47 points

0.70

86.19%

16.19 points

0.75

86.19%

11.19 points

0.80

89.53%

9.53 points

0.85

97.00%

12.00 points

0.90

97.00%

7.00 points

0.95

97.00%

2.00 points

1.00

102.05%

2.05 points

0.20

PART 3

156

Getting Out

BEYOND MAE/MFE
W i t h a stop loss and a profit target in place, we n o w turn our attention to the trailing stop. We would like to find the m a x i m u m adverse excursion ( M A E ) from the
latest favorable excursion (FE) within e a c h trade. N o t e that this does not have to
be the m a x i m u m favorable excursion point, w h i c h is likely to be the very last
favorable excursion, leading into the end trade drawdown, but rather any of the
previous m a x i m u m favorable excursions preceding the largest retracement in the
opposite direction that did or did not result in an exit. We call this point M A E f e .
T h e difficulty w i t h t h e M A E f e is that we do not k n o w w h e r e the m a r k e t
w a s trading before the M A E f e (at least n o t in this analysis). Therefore to c o m e
u p w i t h s o m e meaningful results w e m u s t p r o c e e d differently f r o m the way w e
w o r k the profit target a n d the stop loss analysis. In this c a s e , we c o m p a r e the
M A E f e w i t h the average c o r r e s p o n d i n g M F E for all trades a n d then calculate the
theoretical result as if the M A E f e h a d b e e n the end trade d r a w d o w n . T h e basic
Excel formulae are t h e s a m e as for the stop loss a n d the profit target. T h e results
can be seen in Table 9 . 3 .

TABLE

9.3

MAEfe levels, average MFE levels, and theoretical profits for 16 markets In the directional slope system.
MAEfe

Avg. MFE

Profits

-2.50%

9.78%

7.04%

-3.00%

10.28%

6.98%

-3.50%

10.85%

6.97%

-4.00%

11.39%

6.94%

^.50%

12.02%

6.98%

-5.00%

12.66%

7.03%

-5.50%

13.34%

7.10%

-6.00%

13.71%

6.89%

-6.50%

14.31%

6.88%

-7.00%

14.86%

6.82%

-7.50%

15.33%

6.68%

-8.00%

16.02%

6.74%

-8.50%

16.96%

7.02%

-9.00%

17.24%

6.69%

-9.50%

18.06%

6.84%

-10.00%

19.24%

7.32%

-10.50%

20.14%

7.53%

-11.00%

20.92%

7.62%

-11.50%

21.51%

7.54%

-12.00%

22.08%

7.43%

CHAPTER 9

Sweeney's MAE/MFE

157

F r o m Table 9.3 we can see that w h e n the M A E f e is at 4 % , there is a bottom

for the theoretical profits. With a trailing stop at this level, and a s s u m i n g our reasoning m a k e s sense, the average profit from all stopped out trades should now be
slightly below 7 % . All trades that did not produce a retracement of 4% will continue to produce the same total profit as they would have before.
A trailing stop of 4% from the latest favorable excursion also m a k e s sense
from a symmetrical point of view w h e n c o m p a r e d to the stop loss and what could
be a general market behavior, if we assume that the breakout and initial entry usually coincides with a test of a previous high (low) that acts as a pivotal resistance
(support) level. N o t e also that for both the stop loss and the trailing stop, the next
worse level s e e m s to be very close to 8%, w h i c h further confirms the theory that
there s e e m s to be some kind of symmetrical behavior that results in, on average,
4% retracements (or multiples thereof) after the market has set a new high/low in
the direction of the underlying trend.
A n o t h e r way of lowering the standard deviation of the o u t c o m e is to work
with time-based stops as a c o m p l e m e n t to any of the stops discussed so far. If it is
reasonable to assume that a market that first takes out the latest resistance/support
level signals that it is ready to take off in o n e direction, and then goes nowhere
from there, is a likely reversal candidate (perhaps after a prolonged m o v e in the
same direction), it could be a g o o d idea to c o m p l e m e n t the initial stop loss with a
time-based stop, which is triggered if the trade hasn't reached a certain profit within the stipulated n u m b e r of bars.
Table 9.4 shows that if we would like to take the expected end trade drawd o w n ( E T D ) into consideration to calculate the estimated final profit for a trade,
it will take an average o p e n profit of 8.59% before we can expect the average trade
to e n d with a small profit at s o m e bar in the future. For the average trade, this open
profit will, on average, be reached on b a r 14.
This finding, in conjunction with what we already have discussed, m a k e s it
seem a good idea to put a time-based stop loss at this bar. That is, if a trade stays or
turns negative after this bar, we exit with a market order at the close. Note that it is
the final profit in which we are interested, which should be positive after bar 14.
This m e a n s that although the average open profit for bar 14 is 8.58, we let the trade
live on if it is below this level, as long as it does not m o v e into negative territory.
At the other end of the trade, a time-based stop also could work as a c o m plement to the profit target, to streamline the entire group of trades and lower even
further the standard deviation of the outcomes of all trades. Also, if we let a trade
go on for too long, we only start giving back large parts of the open profit before
we're allowed to exit in accordance with the original system. To investigate this,
we employ a strategy similar to that we used for the stop loss and profit target and
put together a table like Table 9.5.
From Table 9.5 we can see that all trades that last for 40 bars or longer have
an average open profit of 2 5 . 7 4 % , but p r o d u c e an average final profit of 31.96%)

158

PART 3

TABLE

Getting Out

9.4

Bars in trade, open profits, ETD, and differences for 16 markets in the directional slope system.
Bars in Trade

Open profit

ETD

Difference

0.17%

-5.74%

- 5 . 5 8 points

1.16%

-7.20%

- 6 . 0 4 points

2.43%

-7.78%

- 5 . 3 4 points

3.75%

-7.94%

- 4 . 1 8 points

10

5.07%

-8.11%

- 3 . 0 4 points

12

6.31%

-8.22%

- 1 . 9 1 points

14

8.59%

-8.24%

0.35 points

16

10.36%

-8.20%

2.16 points

18

11.98%

-7.90%

4.08 points

20

13.25%

-7.96%

5.29 points

22

14.42%

-8.06%

6.35 points

24

16.66%

-8.20%

8.46 points

26

19.37%

-8.59%

10.79 points

28

22.67%

-8.91%

13.76 points

30

23.16%

-8.66%

14.50 points

32

22.90%

-9.04%

13.86 points

34

24.83%

-9.28%

15.55 points

36

26.66%

-9.55%

17.10 points

38

27.78%

-8.60%

19.18 points

40

25.74%

-8.11%

17.63 points

at some future date. As we can see, there is no such thing as a free lunch at the
markets. Cutting all trades short lowers the total result over a longer string of
trades. W h a t is gained in increased certainty regarding the individual outcome and
lower standard deviation, is lost in a lower expected final profit.
Finally, the time-based stop also could be implemented into the trailing stop
technique. O n e way of doing this is to m a k e use of the findings from a chart
resembling Figure 9.4. For instance, if the current trade is doing better than what
could be expected, as suggested by the least square regression line, then the regression line from the basic system with no stops added could work as a trailing stop
level. For a trade that currently is not doing as well as the average trade, but has
not yet been stopped out for any other reasons, we could use the regular trailing
stop. Alternatively, a combination of several techniques could be used that, for
instance, stops out all trades that retrace m o r e than what is allowed by the trailing
stop, but only if the retracement also m e a n s that the trade has moved below the
regression line, as suggested by the original unadjusted trade. If you also are using
a stop loss, or a time-based stop, as we do at bar 14, then you m u s t think about giving these stops ample leeway as well. If you do not, the trailing stop will probably

CHAPTER 9

159

Sweeney's MAE/MFE

TABLE

9.5

Bars in trade, final profit open profits, and differences for 16 markets in the directional slope system.
Bars in Trade

Final profit

Open profit

Difference

12

8.84%

6.31%

- 2 . 5 3 points

14

11.35%

8.59%

- 2 . 7 6 points

16

12.69%

0.36%

- 2 . 3 3 points

18

14.30%

11.98%

- 2 . 3 2 points

20

15.94%

13.25%

- 2 . 6 9 points

22

17.67%

14.42%

- 3 . 2 5 points

24

19.94%

16.66%

- 3 . 2 8 points

26

22.73%

19.37%

- 3 . 3 5 points

28

24.70%

22.67%

- 2 . 0 3 points

30

25.18%

23.16%

- 2 . 0 2 points

32

27.69%

22.90%

- 4 . 7 9 points

34

28.63%

24.83%

- 3 . 7 9 points

36

31.08%

26.66%

- 4 . 4 3 points

38

27.97%

26.66%

- 1 . 3 1 points

40

31.96%

25.74%

- 6 . 2 2 points

42

33.96%

28.90%

- 5 . 0 6 points

44

35.84%

31.37%

- 4 . 4 7 points

46

37.50%

34.26%

- 3 . 2 4 points

48

42.98%

38.81%

- 4 . 1 8 points

50

44.33%

42.00%

- 2 . 3 3 points

have you stopped out too soon during the initial stage of the trade, w h e n it is the
most vulnerable.
Let us c o m p a r e the results for the original directional slope system with
s o m e of the findings we have m a d e in this section. T h e rules for the modified syst e m a n d the corresponding TradeStation code are:

Enter long/short w h e n the 18-bar average turns up/down.

Exit with a loss if the m a r k e t moves the w r o n g way by 4 % .

Exit with a small loss/breakeven if the market moves into (stays in) negative
territory at or after bar 14.

Exit with a trailing stop, after bar 14, if the market moves the w r o n g way by
4% or m o r e and ends up below the regression line for the original average
trade.

Exit with a profit target if the market moves by 6 0 % or m o r e .

Exit with a profit if the trade lasts for 40 bars or m o r e .

Enter into a neutral position if the 12-bar average changes direction.

160

PART 3

Getting Out

T h e TradeStation code for our additional exit techniques is:

If MarketPosition = 1 Then Begin
ExitLong at EntryPrice * 1.60 Limit;
If BarsSinceEntry >= 40 Then
ExitLong at Close;
If BarsSinceEntry >= 14 Then Begin
If Highest(High, BarsSinceEntry) * 0.96 < EntryPrice * ( 1 + 0.0074 *
BarsSinceEntry) Then
ExitLong at Highest(High, BarsSinceEntry) * 0.96 Stop;
ExitLong at EntryPrice Stop;
End;
ExitLong at EntryPrice * 0.96 Stop;
End;
If MarketPosition = 1 Then Begin
ExitShort at EntryPrice * 0.40 Limit;
If BarsSinceEntry >= 40 Then
ExitShort at Close;
If BarsSinceEntry >= 14 Then Begin
If Lowest(Low, BarsSinceEntry) * 1.04 > EntryPrice * ( 1 - 0.0074 *
BarsSinceEntry) Then
ExitShort at Lowest(Low, BarsSinceEntry) * 1.04 Stop;
ExitShort at EntryPrice Stop;
End;
ExitShort at EntryPrice * 1.04 Stop;
End;
To this code, we also m u s t attach the trade-by-trade export function from
Part 1. We m u s t use the trade-by-trade function because we are now dealing with
ratio adjusted data and can, therefore, no longer rely on the performance summary from TradeStation.
Finally (because of another little TradeStation feature), there will be a few
trades that we m u s t adjust manually within the spreadsheet program. This happens
w h e n the trade moves immediately and swiftly against us and, therefore, should
have b e e n stopped out sooner than TradeStation allows. We also run the risk of
inadvertently changing the results to the better for a few trades, which truly did
have a loss greater than 4% because of legitimate opening gaps. We also must
recalculate the cumulative profit, latest equity high, and drawdown. In this case,
we have a s s u m e d that all losses larger than 5% are erroneous and consequently
adjusted t h e m down to 4 % . Obviously, I am cheating a little bit here and obvious-

CHAPTER 9

Sweeney's MAE/MFE

161

ly, you will n e e d to be m o r e careful in your o w n research, but there's not r o o m

e n o u g h here to go into details.
To recalculate the cumulative profit in Excel, type in the following formula
on top of the values in the cumulative profits column:
In cell G 2 :
=M2
In cell G 3 :
=((1+G2/100)*(1+F3/100)-1)*100
then drag d o w n to the b o t t o m of the sheet, to fill all the cells in the column.
To recalculate the equity top, type:
In cell H 2 :
=MAX(G2,0)
In cell H 3 :
=MAX(H2,G3)
then drag d o w n to the b o t t o m of the sheet, to fill all the cells in the c o l u m n .
To recalculate the drawdown, type into cell L 2 :
=((1 + G2/100)/( 1 + H2/100) -1 )* 100
then drag d o w n to the b o t t o m of the sheet, to fill all the cells in the c o l u m n .
W h e n you are d o n e exporting and calculating in Excel, you can put together
a set of tables resembling Tables 9.6 a n d 9.7 for all of the individual markets, and
Tables 9.8 and 9.9 with a composite of the m o s t important m e a s u r e m e n t s . In Table
9.10, we s u m m a r i z e the differences for each market.
Table 9.6 shows that with the original system, copper had a profit factor of
1.60. T h e largest w i n n e r would have b e e n 7 1 . 4 % , or $13,692 in today's market
value. T h e largest loser would have b e e n 1 6 . 8 % , or $3,231 in today's m a r k e t
value. Table 9.7 shows that with the modified system, copper had a profit factor
of 2 . 0 1 , calculated on the average percentage w i n n e r and loser, and then trans-

TABLE

9.6

Example of performance summary.

Copper, before exits (original system)
Total trades

96

Profit factor

1.60

Lrg winner

1.49%

286.05

Avg winner

12.67%

2,429.91

Avg profit
St Dev

Winners

C u m profit

37

38.54%

59

61.46%

71.36%

13,692.20

Losers
Lrg loser

-16.84%

-3,231.18

10.32%

1,980.46

Avg loser

-4.05%

-776.54

128.72%

24,698.15

Drawdown - 4 3 . 0 2 %

-8,254.46

PART 3

162

TABLE

Getting Out

9.7

Individual market performance summary.

Copper, after exits (modified system)
Total trades

113

Profit factor

2.01

Avg profit

1.76%

337.90

St Dev

9.96%

1,944.14

Winners
Lrg winner
Avg winner
C u m profit

46

40.71%

60.00%

11,512.50

8.62%

1,653.17

363.64%

69,772.90

67

59.29%

Lrg loser

-4.69%

-899.89

Avg loser

-2.95%

-565.11

Drawdown - 4 0 . 8 8 %

-7,843.57

Losers

formed into today's market value. T h e largest winner would have been 6 0 %
(thanks to the profit target), or $11,513 in today's market value. T h e largest loser
would have been 4 . 7 % (thanks to the stop loss), or $900 in today's market value.
From Table 9.8 we can see that all markets but one had a profit factor above
1 in the original system. A high profit factor, however, does not necessarily mean
that the market in question is worth trading. For this conclusion, we m u s t also look
at the value of the average trade, in today's market, and deduct a proper amount for
slippage and c o m m i s s i o n . A rule of t h u m b could be that the remaining value of the
average trade should be at least two times the estimated slippage a n d commission,

TABLE

9.8

Performance measurements for all markets with the original system.

Before exits (original system)
Market

P. factor

Avg. trade

2 St. dev.

Drawdown

Corn

2.43

285.7

62,254.95

-3,604.93

S&P 500

1.01

31.63

37,680.83

-136,839.38

Orange juice

1.58

218.63

3,126.66

-6,671.28

Live cattle

1.34

149.33

3,285.72

-8,007.16

Lumber

2.20

816.09

7,019.19

-14,383.49

Coffee

2.00

1,443.41

19,797.92

-22,991.64

Japanese yen

3.88

3,462.79

19,158.31

-11,891.25

Copper

1.60

286.05

4,859.82

-8,254.46

Gold

1.51

243.21

4,174.29

-9,413.32

Eurodollar

1.74

239.40

3,375.29

-5,953.72

Dollar index

2.22

1,004.14

7,502.97

-7,436.68

Cotton

2.47

681.00

5,227.15

-4,309.28

CRB index

0.70

-390.33

7,135.96

-38,390.04

Crude oil

2.46

584.00

5,540.84

-6,996.26

Canada dollar

1.59

194.31

2,617.40

-^,331.60

T-bonds

1.34

521.24

10,826.72

-37,845.11

CHAPTER 9

Sweeney's MAE/MFE

TABLE

163

9.9

Performance measurements for all markets.

After exits (modified system)
Market

P. factor

Avg. trade

2 St. dev.

Drawdown

Corn

2.62

266.72

2,036.38

-2,384.17

S&P 500

0.67

-1,947.59

26,215.82

-177,550.12

Orange juice

-3,544.32

2.04

234.21

2,346.97

Live cattle

1.23

93.94

2,707.90

-6,913.48

Lumber

2.38

758.44

6,289.95

-11,438.38

Coffee

2.08

917.67

10,271.95

-15,742.83

Japanese yen

2.72

2,000.98

11,427.63

-8,938.80

Copper

2.01

337.90

3,822.28

-7,843.57

Gold

1.54

248.07

3,882.21

-10,208.43

Eurodollar

1.74

238.78

3,373.71

-6,416.68

Dollar index

2.00

829.42

7,282.08

-8,200.28

Cotton

2.02

462.77

4,669.08

-6,965.14

CRB index

0.78

-258.75

5,991.32

-33,982.87

Crude oil

4.21

1,363.21

7,898.06

-5,875.78

Canada dollar

1.36

116.00

2,204.25

-5,551.04

T-bonds

1.28

399.42

9,802.15

-28,461.80

to m a k e r o o m for a decrease in performance and so that the system will continue

to be profitable in case the m a r k e t declines and, consequently, the value of the
average trade b e c o m e s lower as well. With a slippage and c o m m i s s i o n of $ 7 5 ,
Tables 9.8 and 9.9 contain 11 and 12 markets, respectively, that are w o r t h trading.
T h e five m a r k e t s not worth trading, in Table 9.8, are S & P 500, orange j u i c e , live
cattle, C R B index, and C a n a d a dollar. Of these five markets, three also h a d a lower
profit factor a n d average trade, after the exits w e r e implemented. M o s t remarkable
is that orange j u i c e goes from being a non-tradable market to a tradable one.
Of the remaining 11 markets that were tradable with the original system, six
h a d a better profit factor with the modified system, but only three had a better average trade. In essence, this m e a n s that we will not m a k e as m u c h m o n e y per contract
traded with the modified system as with the original system, but at the same time,
the cost of doing business will also be m u c h less. A l s o , just to look at the average
trade by itself is not always that useful. Instead, it should be c o m p a r e d to the standard deviation of the o u t c o m e ; in this case, 15 out of 16 markets h a d a lower standard deviation, which suggests that the system has b e c o m e m o r e stable a n d less
risky to trade. In most cases the drawdown also has decreased dramatically.
From Table 9.10 we also can see that out of the 12 tradable markets with the
modified system (including orange juice), seven (corn, orange juice, lumber, copper,

164

TABLE

PART 3

Getting Out

2 St. dev.

Drawdown

Better
3

9.10

Summarized differences, before/after exits.

Differences
Market
Corn

P. factor

Avg. trade

7.92%

-6.66%

-9.69%

-33.86%

-33.35%

-6257.25%

-30.43%

29.75%

Orange juice

29.11%

7.13%

-24.94%

-46.87%

Live cattle

S&P 500

-8.70%

-37.09%

-17.59%

-13.66%

Lumber

8.59%

-7.06%

-10.39%

-20.48%

Coffee

4.03%

-36.42%

-48.12%

-31.53%

-29.99%

-42.21%

-40.35%

-24.83%

25.58%

18.13%

-21.35%

-4.98%

1.68%

2.00%

-7.00%

8.45%

3
2

Japanese yen
Copper
Gold
Eurodollar

0.47%

-0.26%

-0.05%

7.78%

Dollar index

-10.11%

-17.40%

-2.94%

10.27%

Cotton

-18.19%

-32.05%

-10.68%

61.63%

CRB index

11.48%

-33.71%

-16.04%

-11.48%

Crude oil

71.19%

133.43%

42.54%

-16.02%

-14.68%

-40.30%

-15.78%

28.15%

-4.11%

-23.37%

-9.46%

-24.79%

15

10

Canada dollar
T-bonds
Better

Eurodollar, crude oil and T-bonds) also show a positive relationship when it comes
to how m u c h the average trade has changed in relation to the standard deviation. The
drawdown, finally, has decreased for eight (corn, orange juice, lumber, coffee,
Japanese yen, copper, crude oil, T-bonds) out of the 12 worth trading. That is, with
the modified system the cost of doing business has decreased, while at the same time
we can feel more sure about the outcome and sleep better thanks to a lower expected drawdown. T h e drawback is that the value of the average trade has decreased; but
because of all the benefits, we still should be able to make more money over any
given period of time. This is thanks to the fact that a m o r e stable system allows us to
trade more aggressively with more contracts, in accordance with a fixed fractional
money management strategy.
Finally, please r e m e m b e r that this is a sample system only and that we used
all the exit techniques previously discussed. This might not be the most optimal
solution, and it might be better, excluding one or several of the exits. In this case
(and all the following), I added all the exits directly on top of the results from the
original system. Because the statistical traits for the system will change with every
change we m a k e , this is not necessarily the best way to do it. Instead, a better way
could be to examine only one exit technique at a time and then rerun the system
before each new implementation. Or, set up the modified system as we have done

CHAPTER 9

Sweeney's MAE/MFE

165

here a n d work b a c k w a r d s from there, modifying or excluding one exit type at a

time. A l s o , in this and the other long-term systems to follow, because of a variety
of u n w a n t e d TradeStation "features" I have had to m a k e a few modifications to the
TradeStation c o d e and simplifications in the Excel spreadsheet that might have
diluted the results somewhat.

CHAPTER

10

Adding Long-term Exits

There are probably several w a y s of locking in a profit or taking a loss, b u t one

technique that has great scientific value is J o h n Sweeney's m a x i m u m adverse
( M A E ) and m a x i m u m favorable excursion ( M F E ) . E v e n this m e t h o d can p r o d u c e
catastrophic results if not u n d e r s t o o d and i m p l e m e n t e d properly. As I have stressed
throughout this b o o k , one of the m o s t important things you m u s t do is m a k e sure
that you understand exactly w h a t it is you are m e a s u r i n g a n d h o w to go about
doing it. In the following chapters, I will r u n d o w n w h a t we have learned and
i m p l e m e n t it on the other two long-term systems with w h i c h we have worked so
far. In doing so, I am painfully aware that this m i g h t not be the optimal solution,
and you m i g h t be better off excluding one or several of the exits we discuss here.
T h e important thing to r e m e m b e r is w h a t it is you w a n t your system to d o . If for
instance, you want it to be a long-term trend-following system, you must treat it
as such a n d allow p r o p e r leeway for each trade, so that you do not inadvertently
c h a n g e it into a shorter-term system.
With the M A E / M F E technique it also is possible to add to trades that show
tendencies of being better than w h a t could be expected. However, call me a b a d syst e m s developer, if you like, but I have yet to c o m e across such a system that is capable of delivering results and is thoroughly robust and stable. Therefore, I have c h o sen not to look into this technique in this book.
A n o t h e r possibility I have not explored is to fade the original system if the
trade moves against us by a certain amount. I believe that such a technique,
although it h a s its merits, starts to c o m e awfully close to curve fitting. This
b e c o m e s apparent if you formulate the rules in plain English. For instance, first we
w a n t the system to trigger a trade in one direction, w h i c h is a very specific request
167

PART 3

168

Getting Out

in the first place. T h e n we want the market to move against us by a certain amount
within a limited t i m e frame, w h i c h are two other very specific requests.
F u r t h e r m o r e , to take all the other exits a n d risk levels into consideration, we might
even add another criterion or two that might stipulate that the m a r k e t is allowed or
not allowed to behave in a certain fashion. A little too m a n y rules, ifs, ands, and
buts for my comfort.

THE DYNAMIC BREAKOUT SYSTEM

In Part 2 we discussed several different versions of the D y n a m i c Breakout System
(the D B S system) before we settled for version l b , which h a d a "reversed" volatility relationship for the exits. That is, w h e n the historical volatility increased, the
lookback period for the exit decreased. T h e underlying reasoning w a s that if a
higher volatility, as an indication of higher risk, m a k e s it m o r e difficult to enter the
market, then a higher volatility also should m a k e it m o r e difficult to stay in (or easier to get out, if you so wish).
In this chapter we apply that system (version l b ) to the s a m e 16 markets as
in Part 2, to examine the possibility of increasing its performance by adding a few
stops and exits in accordance with our findings regarding John Sweeney's
M A E / M F E methods. T h e t i m e period covered is from January 1980 to October
1999. A n o t h e r reason I chose to work with this system is simply that the original
system is closer in resemblance to the standard deviation breakout system that follows. T h e export function from TradeStation into Excel is essentially the s a m e for
the D B S system as for the directional slope system.
Figure 10.1 shows the average open profit for all trades in relation to the
n u m b e r of bars in trade. Instead of looking at h u n d r e d s of lines that m a k e little
sense, we are n o w down to four, for m u c h easier interpretation and rule making.
From this chart we can see that the average profit is expected to increase with
about 0 . 2 3 % per bar. To the right of the dotted line we have fewer than 20 trades
that remain open, which is too few to m a k e any meaningful statistical conclusions.
T h e first things we will examine are the stop loss in accordance with the
M A E , the trailing stop in accordance with the M A E f e , and the least square regression line from Figure 1 0 . 1 . To find the m o s t logical level for w h e r e to place a stop
loss or trailing stop you can put together a set of tables resembling Tables 10.1 and
10.2. As you can see from Table 10.1, for this version of the D B S system, we
would have benefited from placing a stop loss at j u s t about any level. In fact, the
greater the M A E , the m o r e we would gain h a d we decided to place a stop at a specific level. Because there seem to be no local m i n i m a , indicating s o m e kind of
market inefficiency that we could have tried to exploit, we simply m u s t place the
stop w h e r e it m a k e s m o s t sense.
Table 10.3 shows that, basically the s a m e reasoning goes for the trailing stop
as for the stop loss. In this case, the greater the M A E f e value, the less we benefit

CHAPTER 10

169

Adding Long-term Exits

FIGURE

10.1

Tthe average open profit for all trades in relation to the number of bars In trade.

TABLE

10.1

MAE levels, original profits, and differences in outcome for 16 markets in the DBS system.
MAE

Original profit

Difference

-0.50%
-1.00%

-0.92%

- 0 . 4 2 points
- 0 . 3 5 points

-1.50%
-2.00%
-2.50%
-3.00%
-3.50%
^1.00%
-4.50%
-5.00%
-5.50%
-6.00%
-6.50%
-7.00%
-7.50%
-8.00%
-8.50%
-9.00%
-9.50%
-10.00%

-1.35%
-1.98%
-2.64%
-3.43%
^.20%
^.67%
-5.29%
-5.89%
-6.58%
-7.27%
-7.76%
-8.46%
-9.00%
-9.65%
-10.08%
-10.39%
-11.08%
-11.71%
-12.56%

-0.48
-0.64
-0.93
-1.20
-1.17
-1.29
-1.39

points
points
points
points
points
points
points

-1.58
-1.77
-1.76
-1.96
-2.00

points.
points
points
points
points

- 2 . 1 5 points
- 2 . 0 8 points
-1.89
-2.08
-2.21
-2.56

points
points
points
points

170

PART 3

TABLE

Getting Out

10.2

MAEfe levels, average MFE levels, and theoretical profits for 16 markets in the DBS system.
MAEfe

Avg. MFE

Profits

-2.50%

6.68%

4.02%

-3.00%

7.08%

3.87%

-3.50%

7.51%

3.75%

-4.00%

8.02%

3.70%

-4.50%

8.53%

3.65%

-5.00%

8.98%

3.54%

-5.50%

9.44%

3.42%

-6.00%

9.75%

3.17%

-6.50%

10.07%

2.91%

-7.00%

10.49%

2.75%

-7.50%

10.84%

2.52%

-8.00%

11.01%

2.13%

-8.50%

11.59%

2.11%

-9.00%

11.83%

1.76%

-9.50%

12.11%

1.46%

-10.00%

12.49%

1.24%

-10.50%

12.49%

0.68%

-11.00%

13.13%

0.69%

-11.50%

13.53%

0.47%

-12.00%

14.29%

0.58%

from a stop at a specific level. No local m a x i m a indicate any market inefficiency

that we could have tried to exploit, so again, you simply must place the stop where
it m a k e s the m o s t sense to you. For the sake of simplicity, then, we place a stop
loss and a trailing stop at 4 % . T h e trailing stop c o m e s into play at the same time
as the time-based breakeven stop and only kicks in if the decrease in open profit
also takes us below the regression line from Figure 10.1. Exactly h o w deep this
retracement m u s t be before we exit we d o n ' t know, but we do k n o w that as long as
the retracement does not bring us below the regression line, the trade is still doing
better than average and could be worth staying in.
A n o t h e r way of analyzing the expected profits is to look at the average open
profit at a specific bar a n d then deduct the expected E T D for trades lasting longer.
This is shown in Table 10.3, which shows that final profit is not expected to turn
positive until after b a r 30. With the information from Figure 10.1 and Table 10.3,
we can formulate our third exit rule, which says that if the open profit for a trade
stays or turns negative at or after bar 30, we exit with a market order at the close.
Table 10.4 shows that we would n o t have benefited from placing a profit target anywhere. However, using the s a m e b a c k w a r d s reasoning as we did for the stop

CHAPTER 10

Adding Long-term Exits

TABLE

171

10.3

Bars in trade, open profits, ETD, and differences for 16 markets in the DBS system.
Bars in trade

Final profit

Open profit

Difference

0.12%

-5.31%

- 5 . 1 9 points

0.26%

-5.33%

- 5 . 0 7 points

0.38%

-5.33%

- 4 . 9 4 points

0.65%

-5.36%

- 4 . 7 1 points

10

0.97%

-5.32%

- 4 . 3 5 points

12

1.34%

-5.30%

- 3 . 9 7 points

14

1.72%

-5.31%

- 3 . 5 9 points

16

2.14%

-5.28%

- 3 . 1 3 points

18

2.69%

-5.30%

- 2 . 6 1 points

20

3.25%

-5.31%

- 2 . 0 7 points

22

3.75%

-5.27%

- 1 . 5 2 points

24

4.21%

-5.28%

- 1 . 0 8 points

26

4.61%

-5.36%

- 0 . 7 5 points

28

5.15%

-5.38%

- 0 . 2 3 points

30

5.54%

-5.48%

0.06 points

32

6.13%

-5.52%

0.61 points

34

6.48%

-5.58%

0.90 points

36

7.02%

-5.59%

1.44 points

38

7.50%

-5.67%

1.83 points

40

7.92%

-5.61%

2.31 points

loss and profit target in the directional slope system, the " b e s t " result s e e m s to be
around the 35 to 4 0 % area. Because this s e e m s to be a little too close to the entry,
we will instead choose to go with the second "best" level at around 7 0 % . That is,
as soon we reach an open profit of 7 0 % or m o r e , we exit with a limit order. T h e
p u r p o s e of this is to streamline all trades as m u c h as possible and keep the standard deviation as low as possible. By keeping the standard deviation down, we are
decreasing the overall risk of the system and should be able to trade it m o r e
aggressively with m o r e contracts within a fixed fractional m o n e y m a n a g e m e n t
strategy. Therefore, what we lose with the profit target stop, we h o p e to regain
from m o r e aggressive trading.
A n o t h e r way to achieve parts of the s a m e effect is to apply a time-based stop.
Just as w a s the case for the profit target, Table 10.5 shows that the D B S system
s e e m s to have two different trade lengths at which this would be the least opportune m o m e n t . T h e first o n e c o m e s after about 50 bars and the second after about
100 bars, which, by the way, again indicates that there is some sort of s y m m e t r y
in h o w the markets behave. U s i n g the s a m e reasoning as for the profit target, a n d
not to place the stop too close to the entry point, we place it at bar 100. R e m e m b e r ,

PART 3

172

TABLE

Getting Out

10.4

MFE levels, original profits, and differences for 16 markets in the DBS system.

WIFE

Original profit

Difference

5.00%

6.14%

1.14 points

10.00%

11.61%

1.61" points

15.00%

16.27%

1.27 points

20.00%

22.86%

2.86 points

25.00%

28.20%

3.20 points

30.00%

35.08%

5.08 points

35.00%

44.54%

9.54 points

40.00%

48.30%

8.30 points

45.00%

51.77%

6.77 points

50.00%

53.96%

3.96 points

55.00%

55.96%

0.96 points

60.00%

61.49%

1.49 points

65.00%

70.85%

5.85 points

70.00%

78.62%

8.62 points

75.00%

81.24%

6.24 points

80.00%

81.24%

1.24 points

85.00%

95.07%

10.07 points

90.00%

95.07%

5.07 points

95.00%

95.07%

0.07 points

100.00%

150.62%

50.62 points

however, that this is very b a c k w a r d reasoning and for demonstration purposes

only, just to c o m e up with s o m e motivation to use these stops at all.
Hence, the suggested exit levels for the D B S system are as follows:

Exit with a loss if the market moves the w r o n g way by 4 % .

Exit with a small loss/breakeven if the m a r k e t moves into (stays in) negative
territory at or after bar 30.

Exit with a trailing stop, after bar 30, if the m a r k e t moves the w r o n g way by
4% or m o r e and ends up below the regression line for the original average
trade.

Exit with a limit order if the open profit exceeds 7 0 % .

Exit with a profit if the trade lasts for m o r e than 100 bars.

W h e n you are done exporting and calculating in Excel, you can put together a
set of tables like Tables 10.6 and 10.7 for all of the individual markets, and Tables
10.8 and 10.9 with a composite of the most important measurements. In Table 10.10,
finally we summarize the differences for each market. Tabic 10.6 shows that with the

CHAPTER 10

Adding Long-term Exits

TABLE

173

10.5

Bars in trade, final profit, open profits, and differences for 16 markets in the DBS system.
Bars in trade

Final profit

Open profit

Difference

30

5.56%

5.54%

- 0 . 0 2 points

35

6.70%

6.74%

0.03 points

40

7.97%

7.92%

- 0 . 0 5 points

45

9.62%

9.73%

0.10 points

50

10.74%

11.32%

0.58 points

55

11.72%

12.12%

0.41 points

60

12.43%

11.38%

- 1 . 0 5 points

65

12.56%

11.84%

- 0 . 7 3 points

70

12.91%

12.36%

- 0 . 5 5 points

75

13.91%

13.42%

- 0 . 4 9 points

80

14.43%

14.12%

- 0 . 3 1 points

85

16.95%

17.56%

0.61 points

90

16.66%

17.29%

0.63 points

95

18.52%

19.12%

0.60 points

100

18.35%

21.25%

2.90 points

105

22.76%

23.05%

0.29 points

110

27.06%

28.14%

1.07 points

115

28.67%

26.91%

- 1 . 7 5 points

120

24.36%

21.29%

- 3 . 0 7 points

125

24.36%

22.70%

- 1 . 6 6 points

original system, Japanese yen had a profit factor of 2.00. The largest winner would
have been 2 1 . 5 % , or $25,263 at today's market value. T h e largest loser would have
been 5 . 7 1 % , or - $ 6 , 7 0 9 at today's market value. Table 10.7 shows that with the
modified system, Japanese yen h a d a profit factor of 2 . 2 1 , calculated on the average
percentage winner and loser, and then transformed into today's market value. T h e
largest winner would have been 17.6%, or $20,668 at today's market value. T h e
largest loser would have been 4.20%, or $4,935 at today's market value.

TABLE

10.6

Individual market performance summary.

Japanese yen, before exits (original system)
Total trades

140

Profit factor

2.00

Winners
Lrg winner

65

46.43%

21.50%

25,262.50

Avg profit

1.00%

1,173.83

Avg winner

4.31%

5,060.09

St Dev

4.41%

5,178.98

C u m profit

254.60%

299,155.00

75

53.57%

Lrg loser

-5.71%

-6,709.25

Avg loser

-1.87%

-2,194.27

Drawdown - 1 2 . 0 4 %

-14,147.00

Losers

PART 3

174

TABLE

Getting Out

10.7

Individual market performance summary.

Japanese yen, after exits (modified system)
Total trades

143

66

46.15%

Profit factor

2.21

Winners
Lrg winner

17.59%

20,668

Avg profit

1.05%

1,236

Avg winner

4.17%

4,898

St Dev

4.13%

4,848

Cum profit

298.86%

351,159

Losers

77

53.85%

Lrg loser

-4.20%

-4,935

Avg loser

-1.62%

-1,902

Drawdown - 1 1 . 6 6 %

-13,701

F r o m Table 10.8 we can see that c o m p a r e d to the directional slope system,

the D B S system has several markets with considerably lower profit factors and
several m a r k e t s with an average trade too low to m a k e it worthwhile trading after
we have deducted the necessary a m o u n t for slippage and c o m m i s s i o n . Of all 16
m a r k e t s traded with the original system, all but three ( S & P 500, copper, and CRB
index) h a d a profit factor above one. For a system to be profitable on a specific
market, however, the value of the average trade also must be high enough. To stay
on the safe side and to m a k e r o o m for future changes in the value of the average
trade, which is dependent on the current level at which the m a r k e t is trading and

TABLE

10.8

Performance measurements for all markets.

Before exits (original system)
Market

P. factor

Avg. trade

2 St. dev.

Drawdown

Corn

1.51

113.34

1,549.46

-2,865.10

S&P 500

0.80

-1,220.96

29,455.57

-166,159.78

Orange juice

1.50

173.24

2,808.58

-9,319.17

Live cattle

1.06

26.23

2,388.86

-9,217.80

Lumber

1.07

55.70

5,022.72

-16,904.45

Coffee

1.44

509.15

11,095.92

-16,446.50

Japanese yen

2.00

1,173.83

10,357.98

-14,147.00

Copper

0.91

-46.29

2,705.02

-12,075.74

Gold

1.34

174.87

3,470.76

-9,157.20

Eurodollar

1.61

226.78

3,057.64

-8,576.59

Dollar index

2.04

737.82

5,972.58

-9,281.08

Cotton

1.27

178.81

5,137.79

-14,027.04

CRB index

0.59

-517.88

5,032.53

-56,549.59

Crude oil

1.84

569.80

5,974.32

-10,282.72

Canada dollar

1.09

28.86

1,931.03

-7,990.00

T-bonds

1.11

161.90

8,898.80

-30,230.08

CHAPTER 10

Adding Long-term Exits

175

the point value of the market, a rule of t h u m b is that the value of the average trade,
net after expected costs, should be two times the expected costs. For instance, if
you expect slippage a n d c o m m i s s i o n s to be $ 7 5 , the gross value of the average
trade should be $ 2 2 5 . U s i n g this rule of t h u m b , this version of the D B S system
holds five markets that are worth trading both with the original system and w i t h
the modified version.
Table 10.9 shows that after we have added the stops, the changes in the syst e m are not that large in the profit factor and average trade. However, w h e n looking at the standard deviation of the trades and the drawdown, we can see that the
improvements for the portfolio as a whole are quite dramatic.
On a separate note, it is w o r t h m e n t i o n i n g that j u s t b e c a u s e a m a r k e t s e e m s
not to be profitable in itself, it still could be w o r t h trading in a portfolio of m a r kets, with a p r o p e r m o n e y m a n a g e m e n t strategy: it m i g h t be uncorrelated e n o u g h
with all or s o m e of the other m a r k e t s to p e r f o r m satisfactorily w h e n the other
m a r k e t s are p e r f o r m i n g badly, thus, k e e p i n g the equity level u p , as well as
smoother, for the future w h e n the other m a r k e t s start to p e r f o r m better. W h e n
tested on historical c o m m o d i t y futures data, such t e n d e n c i e s can m a n y t i m e s be
found in such m a r k e t s as short-term interest rates, lumber, live cattle, a n d p r e cious metals.

TABLE

10.9

Performance measurements for all markets.

After exits (modified system)
Market

P. factor

Avg. trade

2 St. dev.

Drawdown

Corn

1.38

82.76

1,431.65

-3,203.35

S&P 500

0.83

-886.14

26,848.42

-167,873.90

Orange juice

1.49

141.62

2,442.96

-8,684.48

Live cattle

1.02

7.54

2,289.34

-10,738.53

Lumber

1.13

74.29

3,581.20

-12,035.95

Coffee

1.01

10.65

4,635.90

-23,150.08

Japanese yen

2.21

1,236.13

9,696.50

-13,700.98

Copper

0.94

-25.61

2,249.53

-10,131.14

Gold

1.45

194.71

2,964.58

-8,556.81

Eurodollar

1.69

248.93

3,018.58

-8,578.04

Dollar index

2.10

702.19

5,733.17

-7,862.49

Cotton

1.24

131.13

3,843.48

-10,841.46

CRB index

0.60

-502.97

5,123.84

-56,819.25

Crude oil

2.23

544.72

4,662.18

-7,504.91

Canada dollar

1.12

40.08

1,938.18

-7,202.11

T- bonds

1.24

332.91

8,688.05

-25,087.24

176

TABLE

PART 3

Getting Out

Better

10.10

Summarized differences, before/after exits.

Differences
Market

P. factor

Corn

Avg. trade

2 St. dev.

Drawdown

-8.31%

-26.98%

-7.60%

11.81%

4.47%

-27.42%

-8.85%

1.03%

Orange juice

-0.04%

-18.25%

-13.02%

-6.81%

Live cattle

-4.29%

-71.23%

-4.17%

16.50%

5.79%

33.38%

-28.70%

-28.80%

-29.65%

-97.91%

-58.22%

40.76%

10.41%

5.31%

-6.39%

-3.15%

Copper

3.46%

-44.68%

-16.84%

-16.10%

Gold

7.61%

11.34%

-14.58%

-6.56%

Eurodollar

5.14%

9.77%

-1.28%

0.02%

Dollar index

2.81%

-4.83%

-4.01%

-15.28%

-2.33%

-26.67%

-25.19%

-22.71%

0.90%

-2.88%

1.81%

0.48%

21.16%

-4.40%

-21.96%

-27.01%

3.34%

38.90%

0.37%

-9.86%

11.90%

105.63%

-2.37%

-17.01%

14

10

S&P 500

Lumber
Coffee
Japanese yen

Cotton
CRB index
Crude oil
Canada dollar
T-bonds
Better

11

Table 10.10 shows that of all 16 markets examined, we m a n a g e d to lower the

d r a w d o w n for 10 and lower the standard deviation of the trades for 14. A l s o , of all
markets, only four increased their performance in only one category, while as
m a n y as nine m a n a g e d to improve at least three categories. Of all m a r k e t s tested,
6 ended up with a higher average trade after the exits were added, while a total of
14 also m a n a g e d to end up with lower standard deviations. This is all g o o d news,
indicating that the system has b e c o m e m o r e stable and easier to trade. This is further confirmed by the profit factor, which has increased for a total of 11 markets
after the exits have been added. This m e a n s that the cost of doing business has
decreased considerably.
A l t h o u g h this system overall seems to be less profitable than the directional
slope system, and although the average trade did not improve as m u c h as we would
have liked, we can see that the exits are still doing their j o b w h e n it comes to
streamlining the trades, w h i c h should m a k e it possible to trade it m o r e aggressively with the proper m o n e y m a n a g e m e n t . Other things that could increase performance might be to change b a c k to the original D B S system (version l a ) , which
should give the individual trades a little m o r e leeway, and/or continue to experim e n t with different types of stops other than the profit target stop and the timeb a s e d stop.

CHAPTER 10

Adding Long-term Exits

177

THE STANDARD DEVIATION BREAKOUT SYSTEM

Let us do one m o r e long-term system in this meticulous m a n n e r before we m o v e
on to the shorter-term systems and yet another way of examining h o w best to exit
a trade. As we mentioned in a previous chapter, the Standard Deviation Breakout
( S D B ) system is very similar to the original D B S system (version l a ) , but despite
the similarities it also has a few specific features that m a k e s it very suitable to
form the base in a portfolio of systems, or as a starting point for diversifying out
into other types of systems, such as short-term systems that trade both with and
against the prevailing trend.
O n e major feature of this system is that it is incredibly robust and trades profitably on a w i d e variety of markets, almost regardless of the lookback period for
the indicators and breakout levels. A less than ideal feature is a relatively low n u m ber of profitable trades. This is, however, of lesser concern for a professional trader, w h o usually is m o r e interested in the n u m b e r of profitable m o n t h s for a system
traded on a portfolio of markets. A n d because the basic d y n a m i c breakout system
usually does a g o o d j o b in letting the profits r u n while cutting the losses short, it
often ends with close to 7 0 % profitable m o n t h s a n d fairly short flat times (time
spent between two equity highs). For our p u r p o s e s , I settled for a system with a
lookback period of 60 bars. This w a s for no particular reason other than that I
k n o w it works fairly well. There w a s no optimization influencing this decision.
W h e n traded on 16 different markets, using the R A D contract, on data from
J a n u a r y 1980 to October 1992, it traded profitably on all 16 markets. T h e 16 m a r kets traded were crude oil, T-bonds, T-bills, r o u g h rice, N i k k e i index, natural gas,
live cattle, lumber, coffee, copper, gold, dollar index, Japanese yen, D m a r k (as a
p r o x y for the E u r o ) , cotton, a n d wheat. T h e export function from TradeStation into
Excel is essentially the s a m e for the S D B system as for the directional slope syst e m . T h e data for the period N o v e m b e r 1992 to October 1999 were saved for some
out-of-sample testing in later sections. Figure 10.2 shows that the o p e n equity for
the average trade is exceptionally smooth all the way up to over 150 bars, and that
we can expect the average trade to increase its o p e n profit with about 0 . 1 8 % per
bar. We will m a k e use of this finding later, w h e n we put together a trailing stop.
T h e dotted line denotes the n u m b e r of bars from w h i c h point on there are fewer
than 20 open trades. This h a p p e n s at b a r 132. As you can see, there are a few trades
that go on for m u c h longer than that.
O n e disadvantage with this system is that it b a s e s its exits on a moving average, w h i c h might start m o v i n g away from the market somewhat before you are
allowed to exit. This is especially important to r e m e m b e r at the time of the entry,
because the m a x i m u m loss, as indicated by the system at that point, might not be
the s a m e as after a few days; you never can be completely sure about h o w m u c h
to risk a n d h o w m a n y contracts to trade in a fixed fractional m o n e y m a n a g e m e n t
strategy.

178

PART 3

rluURc

Getting Out

10.2

The average profit for all open trades in relation to the number of bars in trade.

Therefore, it is very important to examine the system for any natural stop loss
levels based on percentage data. Table 10.11 shows that all trades with an M A E
equal to or greater than 5.5% would have ended with an average profit of 1.90%.
Obviously, had we decided to put a stop in at this level, the average profit would
have been 5 . 5 % , which would have resulted in an overall decrease of performance
of 3.60%) points. For all trades with an M A E of 2 % , the overall decrease in performance would have been 0 . 9 9 % points for these trades. It also seems that the 2%
M A E forms s o m e sort of local m i n i m i u m , with higher values on each side of it. The
5 . 5 % M A E level, on the other hand, seems to form a local m a x i m i u m , with lower
values on each side of it. T h e natural choice, and definitely one that should be
investigated further, would then be to place the stop at 2 % . However, in accordance
with our previous discussion about everything in trading being a trade-off between
good and bad, wanted a n d unwanted features, we decide to go with the 5 . 5 % stop,
simply because this seems to be the most obvious place for w h e r e not place a stop,
hoping that what we might lose in a few individual trades on a one-contract basis
we will win b a c k with the proper m o n e y management. Furthermore, the most obvious solution doesn't necessarily have to be the best one in the long run; what we
lose on the swings, we might m a k e up on the merry-go-round.
Table 10.12 s h o w s t h a t the larger t h e M A E f e , the larger the e x p e c t e d profit, a s s u m i n g that the M A E f e c o r r e s p o n d s to the E T D , w h i c h is t h e last retrace-

CHAPTER 10

Adding Long-term Exits

TABLE

179

10.11

MAE levels, original profits, and differences in outcome for 16 markets in the SDB system.
MAE

Original profit

Difference

-0.50%

-1.00%

1.50 points

-1.00%

-0.52%

1.52 points

-1.50%

-0.23%

1.27 points

-2.00%

-1.01%

0.99 points
1.33 points

-2.50%

-1.17%

-3.00%

-1.52%

1.48 points

-3.50%

-1.89%

1.61 points

-4.00%

-2.40%

1.60 points

-4.50%

-2.45%

2.05 points

-5.00%

-2.24%

2.77 points

-5.50%

-1.90%

3.60 points

-6.00%

-3.38%

2.62 points

-6.50%

-2.64%

3.86 points

-7.00%

-4.16%

2.84 points

-7.50%

-4.73%

2.77 points

-8.00%

-4.15%

3.85 points

-8.50%

-3.53%

4.97 points

-9.00%

-8.02%

0.98 points

-9.50%

-7.65%

1.85 points

-12.57%

- 2 . 5 7 points

-10.00%

m e n t d o w n from t h e M F E level. B e c a u s e o f this w e p r o b a b l y w o u l d b e better

off w i t h o u t a trailing stop, but b e c a u s e we w a n t to m a k e u s e of o u r t e c h n i q u e s ,
we d e c i d e to p l a c e o n e a n y w a y at the 5 . 5 % level, p r o v i d e d that a r e t r a c e m e n t
this large also results in a d o w n s i d e b r e a k t h r o u g h of t h e r e g r e s s i o n line for the
average trade.
Because of the expected E T D , a trade m u s t accumulate a certain open profit before it can be expected to close with a profit. W h e n it c o m e s to the time-based
stop, Table 10.13 shows that the average profit on b a r 40, for all trades that go on
for m o r e than 40 bars, is 7.74%, but before a trade that goes on for this long is
allowed to exit it will, on average, give back 7.63. Thus not until this b a r can we
expect the average trade to e n d with a profit. Therefore, we will cut all trades that
are not profitable at this point on the assumption that they will only turn even
worse because of the E T D that is still likely to come.
At the other end of the spectrum, we do not w a n t our trades to go on for too
long either, because the larger the o p e n profit, the larger the resulting retracement
is likely to be as well, resulting in a less-than-stellar final profit. It j u s t so h a p p e n s
that o n e unsatisfactory feature of the S D B system is its tendency to give back a

PART 3

180

TABLE

Getting Out

10.12

MAEfe levels, average MFE levels, and theoretical profits for 16 markets in the SDB system.

MAEfe

Avg. MFE

Profits

-2.50%

9.96%

7.21%

-3.00%

10.47%

7.15%

-3.50%

11.11%

7.22%

-4.00%

11.79%

7.32%

-4.50%

12.68%

7.61%

-5.00%

13.44%

7.77%

-5.50%

14.07%

7.80%

-6.00%

14.65%

7.77%

-6.50%

15.52%

8.01%

-7.00%

16.42%

8.27%

-7.50%

17.35%

8.55%

-8.00%

18.65%

9.16%

-8.50%

20.06%

9.85%

-9.00%

20.81%

9.94%

-9.50%

21.50%

9.96%

-10.00%

22.86%

10.58%

-10.50%

23.15%

10.22%

-11.00%

24.39%

10.71%

-11.50%

26.12%

11.62%

-12.00%

27.05%

11.80%

substantial part of the o p e n profit before it signals time to exit. T h i s can be seen
from Table 10.14 which, for instance, shows that all trades with an M F E of 6 0 %
or m o r e , are likely to p r o d u c e only an average profit of 4 5 . 6 3 % . For the S D B system, we seem to be better off no matter w h e r e we place a profit target. There seems
to be no local m a x i m a or m i n i m a on w h i c h to base our reasoning (backwards or
otherwise), so just for the heck of it we place it at 7 5 % , which should result in a
performance increase of about 25.58 percentage points. ( R e m e m b e r that these systems are for demonstration p u r p o s e s only.)
A l s o , notice the differences between this system and the not too different
D B S system in the previous chapter. With the S D B system, we seemingly do
worse using a stop loss and a trailing stop, but do better with a profit targetthe
exact opposite of what we found for the D B S system, although we are dealing with
two trend-following strategies, which, one would think, would act pretty m u c h the
same. W h e n you put together your own systems, y o u must think about all these
subtleties and m a k e sure that they w o r k the way y o u want t h e m to and in a way
that m a k e sense to you. N o b o d y else can m a k e these decisions for you. I am merely pointing t h e m out and using t h e m as examples for you to learn from.

CHAPTER 10

Adding Long-term Exits

TABLE

181

10.13

Bars in trade, open profits, ETD, and differences for 16 markets in the SDB system.
Bars in trade

Open profit

ETD

Difference

12

1.56%

-6.91%

- 5 . 3 5 points

14

2.01%

-6.92%

4.91 points

16

2.42%

-6.97%

- 4 . 5 5 points

18

2.64%

-7.02%

- 4 . 3 8 points

20

3.23%

-7.10%

- 3 . 8 7 points

22

3.81%

-7.21%

- 3 . 4 1 points

24

4.29%

-7.28%

- 2 . 9 8 points

26

4.49%

-7.44%

- 2 . 9 5 points

28

4.89%

-7.47%

- 2 . 5 9 points

30

5.26%

-7.53%

- 2 . 2 7 points

32

5.72%

-7.60%

- 1 . 8 8 points

34

6.30%

-7.61%

- 1 . 3 0 points

36

6.88%

-7.60%

- 0 . 7 2 points

38

7.24%

-7.63%

- 0 . 4 0 points

40

7.74%

-7.63%

0.11 points

42

7.86%

-7.62%

0.24 points

44

8.25%

-7.69%

0.55 points

46

8.83%

-7.76%

1.07 points

48

9.13%

-7.83%

1.30 points

50

9.68%

-7.70%

1.98 points

W h e n it c o m e s to n o t letting a w i n n i n g t r a d e go o n e for t o o long we also

c o u l d d e c i d e to cut all w i n n e r s at a certain bar, no m a t t e r what. A t r a d e that
g o e s on for t o o long not only i n c r e a s e s t h e s t a n d a r d deviation of o u t c o m e s for
the entire s y s t e m , but also b u i l d s up an o p e n profit that it is likely to give b a c k
a substantial part of before the s y s t e m signals an exit. T h e r e f o r e , it could be a
g o o d idea to cut all profitable t r a d e s that go on p a s t a certain n u m b e r of b a r s .
To c o m e to a c o n c l u s i o n for this we can c o m p a r e the average o p e n profit for a
specific b a r with the average final profit for all t r a d e s that last longer. In this
case, Table 10.15 s h o w s that t h e r e s e e m s to be s o m e k i n d of local m a x i m u m
s u r r o u n d i n g the 105-day region, so we d e c i d e to go w i t h a m a x i m u m t r a d e
length of 105 days.
With all the suggested exit levels in place, the exit techniques for the S D B
system are as follows:

Exit with a loss if the market moves the w r o n g way by 5 . 5 % .

Exit with a small loss/breakeven if the market moves into (stays in) negative
territory at or after bar 4 0 .

PART 3

182

TABLE

Getting Out

10.14

MFE levels, original profits, and differences for 16 markets in the SDB system.
MFE

Original profit

Difference

5.00%

7.34%

2.34 points

10.00%

12.03%

2.03 points

15.00%

16.16%

1.16 points

20.00%

20.02%

0.02 points

25.00%

24.88%

- 0 . 1 2 points

30.00%

32.25%

2.25 points

35.00%

34.60%

- 0 . 4 0 points

40.00%

37.65%

- 2 . 3 5 points

45.00%

41.65%

- 3 . 3 5 points

50.00%

42.97%

- 7 . 0 3 points

55.00%

45.63%

- 9 . 3 7 points

60.00%

45.63%

- 1 4 . 3 7 points

65.00%

47.10%

- 1 7 . 9 0 points

70.00%

47.52%

- 2 2 . 4 8 points

75.00%

49.42%

- 2 5 . 5 8 points

80.00%

49.42%

- 3 0 . 5 8 points

85.00%

45.69%

- 3 9 . 3 1 points

90.00%

45.69%

- 4 4 . 3 1 points

95.00%

49.41%

- 4 5 . 5 9 points

100.00%

49.41%

- 5 0 . 5 9 points

Exit with a trailing stop, after bar 40, if the market moves the w r o n g way by
5 . 5 % or m o r e and ends up below the regression line for the original average
trade.

Exit with a limit order if the open profit exceeds 7 5 % .

Exit with a profit if the trade lasts for m o r e than 105 bars.

Tables 10.16 and 10.17 show the hypothetical results for wheat traded with
and without the exits. Without the exits, wheat had a profit factor of 1.66 and an
average profit p e r trade of 1.27%, equal to $277 in today's marketplace. With the
exits, wheat had a profit factor of 1.85 and an average profit of 1.48%, equal to
$307 in today's marketplace. Tables 10.16 and 10.17 also show that the added exits
have decreased both the drawdown and the standard deviation, which is good
news. For the modified system, the largest winner would have been 3 2 . 1 3 % , or
$6,665 at today's market value. T h e largest loser would have been 5 . 5 0 % (thanks
to the stop loss), or $1,141 at today's market value.
T h a t the S D B system is very robust can be seen from Table 10.18, which
shows that no market ended up with a profit factor below 1 or an average trade less

CHAPTER 10

Adding Long-term Exits

TABLE

183

10.15

Bars in trade, final profit, open profits, and differences.

Bars in trade

Final profit

Open profit

55

10.92%

10.93%

0.02 points

60

11.90%

11.51%

- 0 . 3 9 points

Difference

65

12.36%

12.62%

0.26 points

70

13.00%

13.44%

0.43 points

75

13.87%

14.49%

0.62 points

80

14.21%

14.85%

0.64 points

85

15.35%

16.12%

0.77 points

90

16.50%

17.57%

1.07 points

95

17.71%

18.76%

1.04 points

100

18.10%

19.75%

1.65 points

105

18.71%

20.05%

1.34 points

110

20.41%

21.93%

1.52 points

115

20.37%

21.34%

0.97 points

120

22.78%

23.00%

0.22 points

125

23.09%

24.46%

1.37 points

130

23.81%

24.69%

0.88 points

135

23.47%

23.71%

0.24 points

140

23.43%

23.16%

- 0 . 2 7 points

145

27.17%

28.02%

0.85 points

150

27.17%

28.99%

1.82 points

than three times the expected costs of $75 p e r trade. S o m e markets, like rough rice
a n d natural gas, even showed amazingly g o o d results with profit factors well
above 10.
Table 10.19 shows that with the exits added, all markets still have a profit
factor above 1, but unfortunately the increased security about each trade's individual o u t c o m e also results in a lower average trade for m o s t markets. However,
because of no disaster markets, the overall result confirms the stability of the S D B

TABLE

10.16

Individual market performance summary.

Wheat, before exits (original system)
Total trades

Profit factor

1.6

Avg profit

1.27%

276.5

St Dev

7.73%

1,678.7

Winners
Lrg winner
Avg winner
C u m profit

15

39.47%

31.11%

6,762.54

Lrg loser

8.09%

1,759.29

Avg loser

46.50%

10,107.94

Drawdown

Losers

23

60.53%

-7.59%

-1,649.88

-3.18%

-690.40

-25.76%

-5,599.58

PART 3

184

TABLE

Getting Out

10.17

Individual market performance summary.

Wheat, after exits (modified system)
Winners

Total trades

Profit factor

1.8

Lrg winner
Avg winner

Avg profit

1.48%

307.3

St Dev

7.52%

1,559.5

C u m profit

16

41.03%

23

58.97%

32.13%

6,665.05

Losers
Lrg loser

-5.50%

-1,140.92

7.84%

1,626.46

Avg loser

-2.94%

-610.23

61.39%

12,734.93

Drawdown - 2 3 . 0 6 %

-4,784.24

system, even after we have "polluted" it with all the exits. Interestingly enough,
the markets that did get a higher average trade a n d profit factor were markets that
usually do well in almost any trend-following system. These markets were Tb o n d s , Nikkei index, Japanese yen, and wheat; dollar index and D m a r k only managed to increase their average trades, and crude oil and cotton their profit factors.
F r o m Table 10.20 we can see that a total of six markets m a n a g e d to increase
their performance in at least three categories, but also that as m a n y as seven only
m a n a g e d to increase one category or less. T h e one market that did not improve its
performance at all w a s natural gas, but even so, it was highly profitable, both with

TABLE

10.18

Performance measurements for all markets.

Before exits (original system)
Market

P. factor

Avg. trade

2 St. dev.

Drawdown

Crude oil

3.08

987.94

5,524.19

-4,275.06

T-bonds

1.23

378.11

10,932.67

-28,771.85

T-bills

2.65

917.62

6,513.73

-5,750.25

Rough rice

10.07

1,678.34

6,042.87

-2,181.27

Nikkei index

2.16

2,847.55

20,787.00

-6,747.19

Natural gas

17.11

2,200.72

4,964.46

-546.27

1.45

228.30

3,596.36

-7,212.76
-6,816.53

Live cattle
Lumber

1.49

333.37

5,089.97

Coffee

2.95

639.42

3,918.96

-3,958.94

Japanese yen

2.29

1,675.54

14,463.96

-11,781.00

Copper

1.89

535.64

5,825.51

-6,583.29

Gold

1.85

662.89

5,732.22

-7,241.86

Dollar index

2.51

1,444.03

7,693.95

-4,357.46

D mark (Euro)

3.11

1,818.71

9,699.02

-8,303.75

Cotton

2.35

755.11

5,557.00

-4,430.03

Wheat

1.66

276.58

3,357.54

-5,599.58

CHAPTER 10

Adding Long-term Exits

185

and without the exits. Of all 16 markets tested, we m a n a g e d to decrease the standard deviation for 13 and the drawdown for 8. Unfortunately, we did not m a n a g e
to increase the profit factor and the average trade for as m a n y as we wanted.
In conclusion, for the S D B system, we could c o m e to the same conclusion as
we did for the D B S system. That is, although the average trade did not improve as
m u c h as we would have liked, we can see that the exits are still doing their j o b
w h e n it c o m e s to streamlining the trades, which is indicated by the fact that we
m a n a g e d to lower the drawdown for eight markets and the standard deviation for
13 markets. This should m a k e it possible to trade it m o r e aggressively with p r o p er m o n e y m a n a g e m e n t ; to possibly increase performance, optimize the lookback
a n d breakout period, and/or continue to experiment with different types of stops
a n d exit levels.

TABLE

10.19

Performance measurements for all markets.

After exits (modified system)
Market

P. factor

Avg. trade

Crude oil

3.44

747.03

4,182.98

-1,803.36

T-bonds

1.29

440.87

10,925.97

-19,886.70

T-bills

2.15

676.68

6,451.11

-5,626.32

Rough rice

1.26

125.77

3,042.76

-5,176.72

Nikkei index

4.09

6,118.78

24,009.93

-4,904.47

Natural gas

6.98

1,557.54

6,441.10

-1,041.45

Live cattle

1.09

48.38

3,018.49

-8,610.89

Lumber

1.42

267.40

4,768.68

-5,572.99

Coffee

1.25

87.68

2,458.44

-4,157.51

Japanese yen

2.41

1,711.84

12,882.49

-10,575.98

Copper

1.15

96.01

4,049.64

-6,919.10

Gold

1.27

248.28

5,376.98

-8,558.86

Dollar index

2.43

1,447.94

8,019.24

^1,357.46

D mark (Euro)

2.96

1,829.22

9,472.23

-12,074.73

Cotton

2.72

670.49

4,529.95

-3,890.78

Wheat

1.85

307.38

3,119.04

-4,784.24

2 St. dev.

Drawdown

186

PART 3

Getting Out

Drawdown

Better

Summarized differences, before/after exits.

Differences
Market

P. factor

Avg. trade

2 St. dev.

Crude oil

11.49%

-24.39%

-24.28%

-57.82%

4.68%

16.60%

-0.06%

-30.88%

T-bills

-18.76%

-26.26%

-0.96%

-2.16%

Rough rice

-87.45%

-92.51%

-49.65%

137.33%

Nikkei index

89.12%

114.88%

15.50%

-27.31%

Natural gas

-59.20%

-29.23%

29.74%

90.65%

Live cattle

-24.96%

-78.81%

-16.07%

19.38%

-4.77%

-19.79%

6.31%

-18.24%

-57.55%

-86.29%

-37.27%

5.02%

4.95%

2.17%

-10.93%

-10.23%

Copper

-39.40%

-82.08%

-30.48%

5.10%

Gold

-31.25%

-62.55%

-6.20%

18.19%

Dollar index

-3.24%

0.27%

4.23%

0.00%

T-bonds

Lumber
Coffee
Japanese yen

D mark (Euro)

-4.80%

0.58%

-2.34%

45.41%

Cotton

15.70%

-11.21%

-18.48%

-12.17%

Wheat

11.57%

11.14%

-7.10%

-14.56%

Better

13

CHAPTER

11

Working with Random Entries

For our shorter-term systems, we take a slightly different approach. We look m o r e

closely into the world of descriptive statistics and learn about a few new measurers
that can c o m e in handy, not only w h e n building systems but also when analyzing
markets in general.
With descriptive statistics, we can estimate and m a k e use of the likelihood of
a certain occurrence, as we did by m e a s u r i n g m e n ' s height for the M e a n d e r system
in Part 2. That is, we can start saying things like " u n d e r n o r m a l circumstances, and
assuming the height of m e n is normally distributed; there is only a 2 . 2 7 % chance
that the next m a n I b u m p into will be very tall a n d only a 0 . 0 5 % chance that I will
b u m p into two such m e n in a row." W h e n doing so, we are relating the r a n d o m variable directly to its m e a n a n d standard deviation.
So far, w h e n we have used several of our statistical m e a s u r e m e n t s , such as the
average profit and the standard deviation of the outcomes, we have a s s u m e d that all
our variables, such as the percentage profit per trade, have b e e n normally distributed a n d statistically independent, continuous r a n d o m variables. The beauty of this
assumption is that the familiar bell-shaped n o r m a l distribution is both easy to
understand and, at the s a m e time, lends itself to approximate other, close to normally distributed variables as well. Figure 11.1 shows what such a distribution can
look like.
N o t e , however, that for a variable to be r a n d o m it does not have to be normally
distributed, as Figure 11.2 shows. This chart is simply created with the help of the
r a n d o m function in Excel. Figure 11.1, in turn, is created by taking a 10-period
average of the r a n d o m variable in Figure 11.2. M o r e often than not, a r a n d o m variable (like that in Figure 11.2) that is an average of another r a n d o m variable (like
187

PART 3

188

FIGURE

Getting Out

11.1

The bell-shaped curve of a normal distribution.

that in Figure 11.1) is distributed approximately as a n o r m a l r a n d o m variable,

regardless of the distribution of the original variable.
However, a normal distribution can very well be both higher and narrower, or
lower and fatter, than that in Figure 11.1. Most important, it has a single m o d e (one
value that appears m o r e frequently than the others) and is symmetrical, with equally as m a n y observations to the left as to the right of the mean. For any normally distributed variable, it holds that 6 8 . 2 7 % of all values lie within 1 standard deviation
of the mean; 9 5 . 4 6 % of the values lie within 2 standard deviations; and 9 9 . 7 3 % of
the values lie within 3 standard deviations of the mean. To calculate the standard
deviation of a sample you can use the standard deviation formula in Excel.
For the longest time, the n o r m a l distribution has b e e n the statistical distribution m o s t analysts used to calculate market returns, although it has long been evident that the m a r k e t does not follow a n o r m a l distribution. This can be seen from
Figure 11.3, which shows the daily percentage returns for the S & P 500 index over
the period April 1982 to October 1999.
As you can see from Figure 11.3, this curve is not entirely symmetrical, and
it s e e m s to have m u c h fatter tails than the n o r m a l distribution curve in Figure 11.1.
If a distribution has a relatively peaked and narrow body a n d fatter tails than a normal distribution, it is said to be leptokurtic. If the opposite hold true, it is said to

CHAPTER 11

Working with Random Entries

FIGURE

189

11.2

A random variable, not normally distributed.

be platykurtic. To calculate the kurtosis of a variable you can use the kurt function
in Excel. In the case of the daily distributions of returns for the S & P 500 index,
kurtosis equals 56. A positive value m e a n s that the distribution is leptokurtic; a
negative value m e a n s that it is platykurtic. W h e n building a trading system we
would prefer for the distribution of returns to be platykurtic, so that we m a k e sure
that the system does not stand or fall with any large out Iyer trades. This is a twoedged sword, however, in that a positive value for the kurtosis also might indicate
that we are doing a g o o d j o b of keeping all our trades as similar as possible to the
average trade. If that is the case, the kurtosis should increase with the n u m b e r of
trades from w h i c h we can draw any statistical conclusions, while the dispersion of
these trades stays the same. That is, we are not changing the levels at which we
take our profits a n d losses.
W h a t probably cannot be seen from Figure 11.3 is that it also is slightly
skewed to the left (a negative skew) with a m e a n smaller than the m e d i a n (the middle value, if all values are sorted from the smallest to the largest). T h e tail to the
left of the b o d y stretches out further than the tail to the right. To calculate the
degree of skew of the distribution y o u can use the skew function in Excel. In this
case, the skew equals - 2 . 0 1 , with the average return p e r day equal to 0 . 0 2 9 6 % and
the m e d i a n equal to 0 . 0 3 1 1 % . W h e n building a trading system we would prefer for

190

PART 3

FIGURE

Getting Out

11.3

A random variable skewed to the left (a negative skew) with a mean smaller than the median.

the returns to be positively skewed (to the right) with the tail to the right stretching out further from the b o d y than the tail to the left, so that we m a k e sure that
w e ' r e cutting our losses short and letting our profits run.
A n o t h e r way to m a k e sure that a system is not too dependent on any out-lyers
is to exclude t h e m w h e n calculating the average return. You can do this in Excel
with the t r i m m e a n function, which excludes a certain percentage of observations
from each end of the distribution. Ideally, w h e n building a trading system we
would prefer for the t r i m m e a n to remain above zero, so that we m a k e sure that the
bulk of our trades are behaving as they should. If the t r i m m e a n is larger than the
average return, it m e a n s that historically the positive out-lyers have been smaller
than the negative out-lyers. If the t r i m m e a n is smaller than the average return, it
m e a n s that historically the positive out-lyers have been larger than the negative
out-lyers. In this case, whatever you think is best depends on what you expect and
how you would like the system to perform in the future. After a few bad trades, it
could be comforting to k n o w that the positive out-lyers historically have a tendency to be larger than the negative ones; but on the other hand, having to rely on a
set of not-yet-seen out-Iyer trades to find your way out of a drawdown, seems to
be awfully close to gambling with your life and the need to rely on Lady Luck.
If we d o n ' t k n o w for sure whether the distribution is normal, we cannot use
our n o r m a l distribution m e a s u r e m e n t s that, for instance, say that 6 8 . 2 7 % percent
of all observations will be within 1 standard deviation of the m e a n . Instead we

CHAPTER 11

191

Working with Random Entries

have to use Chebychef's theorem, which says that for a k greater than or equal to
one (k 2= 1), at least (1 - 1 / k 2 ) will lie within k standard deviations of their
m e a n value. Note that this method cannot give an exact value.
For instance, if we have a n o r m a l distribution, we k n o w that 9 5 . 4 6 % of all
observations will lie within 2 standard deviations, but if we d o n ' t k n o w what the
distribution looks like we can only say that at least 7 5 % (1 1 / 2 2 ) will lie within 2 standard deviations. F u r t h e r m o r e , without k n o w i n g the exact distribution, we
cannot say if they will be equally distributed around the mean. We have to estimate
this from our kurtosis and skew measurements. Similarly, to encapsulate at least
50, 6 7 , and 9 0 % of all observations, the standard deviations m u s t be 1.41, 1.74,
and 3.16, respectively.
Let us take a closer look at how we can rebuild and better the exit techniques
for our short-term systems using this newly found knowledge.
A

THE GOLD DIGGER SYSTEM

T h e second version of our data m i n i n g system, Gold Digger II, told us to go long
as soon as we had two down days and two down weeks in a row. To go short, we
needed two up days and two up weeks. T h e reasoning behind this strategy w a s that
a move in a certain direction cannot go on forever and that a short-term rebound
is likely to happen sooner or later. T h e exit techniques also were very simple. If
the market proved us right by going our way for two consecutive days, we would
exit a long position, or if we were short, we would exit at the first close that went
our way. T h e danger with this strategy is, however, that it lacks any type of stops
other than the entry in the opposite direction, which can be quite a substantial
amount of time and m o n e y away.
T h e advantage with this system, as compared to Gold Digger I, is that Gold
Digger II is m o r e symmetrical in nature, but because of the different exit techniques, not completely so. Let us see if we can m a k e it completely s y m m e t r i c and
at the s a m e time e x a m i n e the possibility of adding some sort of stop-loss techniques as well. For a system to be completely symmetrical in both the long and
short entries, the long and short exit should be m i r r o r images of each other. A partly symmetrical system, such as Gold Digger II, from Part 2, has either symmetrical entries or exits. You will find the original TradeStation code for this system in
Part 2. Because we want this to continue to be a short-term strategy, we also start
out with the assumption that no trade will be allowed to go on for longer than five
days. T h e first step is to modify the TradeStation code to the following:
Conditionl = CloseW(2) > CloseW(l) and CloseW(l) > C and C[2] > C[l] and
C [ l ] > C;
C o n d i t i o n = CloseW(2) < CloseW(l) and CloseW(l) < C and C[2] < C [ l ] and
C[1]<C;

192

PART 3

Getting Out

If Conditionl = True and MarketPosition = 0 Then

Buy ("Go long") at open;
If Condition2 = True and MarketPosition = 0 Then
Sell ("Go short") at open;
If BarsSinceEntry 5= 5 Then Begin
ExitLong ("Exit long") at close;
ExitShort ("Exit short") at close;
End;
Of course, we might have decided on a different m a x i m u m trade length and,
for instance, looked for an optimal value in the s a m e m a n n e r as we did with the
Black Jack system in Part 2. If that had been the case, we also could have used the
measures for skew and kurtosis that seemed to be producing the m o s t stable
results. B u t for now, let us settle for the five-day limit to a trade. As w a s the case
with the original system, we will test this strategy on the R A D contract for the
S & P 500 index futures contract, covering the period January 1995 to October
1999.
U s i n g the trade-by-trade export function from Part 1 p r o d u c e d the results in
Table 11.1, If we c o m p a r e this system to the original version of Gold Digger II,
we can see that the standard deviation for this version is m u c h higher, while the
average profit is m u c h lower, which m e a n s that this version is m u c h riskier to
trade. A lower profit factor also indicates that the cost of doing business has
increased. T h e drawdown also has increased considerably. In all, this new version
of Gold Digger is not as g o o d as the original version. Overall, this totally symmetrical version of Gold Digger does not look as good as its predecessor, although
it has its benefits and perhaps also a few strengths that can be improved upon even
further using the right analysis tools.
As can be deduced from Table 11.2 a n d Figure 11.4, these results are not
normally distributed at all. Instead, we have a m e d i a n that is smaller than the
m e a n and a positive skew, suggesting that the right-hand tail stretches out further
than the o n e on the left-hand side. T h i s is good, b e c a u s e despite the fact that we
are cutting all trades at day five the system by itself is still letting the profits run
while cutting losses short. O n e not so g o o d thing, however, is the positive value

TABLE

11.1

Performance summary for the modified version of Gold Digger II, January 1995-October 1999.
Total trades

80

Profit factor

1.18

Lrg winner
Avg winner

1.65

5,559

C u m profit

9.18

30,983

Avg profit

0.13%

453

StDev

2.22%

7,483

Winners

53.75%

8.54

28,823

Losers
Lrg loser
Avg loser
Drawdown

46.25%

-4.33

-14,614

-1.62

-5,480

-17.47

-58,961

TER 11

Working with Random Entries

TABLE

193

11.2

Descriptive statistics for Gold Digger II (modified version).

Mean

0.13

Median

0.21

Kurtosis

2.49

Skew

0.83

Trimmean (20%)

0.04

for the kurtosis, w h i c h suggests that so far the s y s t e m is leptokurtic and, hence,
does not p r o d u c e stable returns. However, with the p r o p e r set of exit techniques
we should be able to lower the kurtosis into negative territory while k e e p i n g a
positive skew.
T h e trick then is to see if we can improve on the results somewhat, while also
m a k i n g the distribution of the outcomes less leptokurtic and perhaps slightly m o r e
skewed to the right. T h a t this should be possible is suggested by the 2 0 % trimm e a n , w h i c h is smaller than the m e a n , indicating that, historically, the positive outlyers have been larger than the negative ones. F r o m Figure 11.4, we also can see

FIGURE

11.4

The distribution of returns for the modified Gold Digger II system.

194

PART 3

Getting Out

that the distribution of returns indicates that the right-hand tail is stretching out
further than the left-hand side. This is generally good and an indication that we are
letting the profits run. T h e positive kurtosis, however, suggests that the results also
are a little too toppy, which indicates instability.
Let us say that we would like to cut all trades that go against us with more than
two standard deviations. Because we know that the returns from this version of Gold
Digger are not normally distributed, we cannot simply cut off all trades that go
against us by a certain value, as suggested by the regular standard deviation formula. Instead, we have to use Chebychev's theorem that says that the two standard deviation boundaries will hold for at least 7 5 % of all observations. But because we do
not know the exact distribution, we also do not know how many of the remaining
2 5 % will be located in the right-hand tail or the left-hand tail, respectively. This we
can only estimate with the help of the other measures. T h e reasoning, then, is that
because of the positive skew, a larger quantity of the 2 5 % remaining trades should
be located to the left of the mean. For argument's sake then we could say that 20 percentage points should be located in the right tail with the remaining 5 percentage
points in the left tail.
An alternative m e t h o d could be to look for a different a m o u n t of trades that
the standard deviation boundaries should encapsulate. For instance, if we would
like to encapsulate 6 8 % of all trades, we should leave out about 1 6 % of all trades
on each side. Just for argument's sake, let us say that a m o n g these we should cut
four times as many losers as winners, which in this case would mean approximately 21 losers (80 X 0.33 X 0.80) and 5 winners (80 X 0.33 X 0.20). Doing so
in the Gold Digger system would result in a stop loss at about 1.0% and a profit
target at about 3 . 8 % . Let us see w h e r e this takes us, by adding these stops to the
code. (It also might be a good idea to add a m i n i m u m profit objective, but because
the average profit from the original system is somewhat low, we will, instead, aim
for half the distance between the stop loss and the profit target. This will be at an
open profit of 1.4%, which will be triggered if the market first closes with a larger profit but then moves down below this level again.) T h e additional TradeStation
c o d e follows:
If MarketPosition = 1 Then Begin
ExitLong ("Long profit") tomorrow at 1.038 * EntryPrice limit;
ExitLong ("Long loss") tomorrow at 0.99 * EntryPrice stop;
If Close > 1.014 * EntryPrice Then
ExitLong ("Long prot") tomorrow at 1.014 * EntryPrice stop;
End;
If MarketPosition = -1 Then Begin
ExitShort ("Short profit") tomorrow at 0.962 * EntryPrice limit;
ExitShort ("Short loss") tomorrow at 1.01 * EntryPrice stop;

CHAPTER 11 Working with Random Entries

195

If Close < 0.986 * EntryPrice Then

ExitShort ("Short prot.") tomorrow at 0.986 * EntryPrice stop;
End;
Table 11.3 shows the result with these stops in place. As you can see, these
stops most certainly did not improve results. T h i s looks really bad, with an average profit of a m e r e 0 . 0 7 % and a cumulative profit of 4.72% after five years of
trading. T h e only positive thing is that we m a n a g e d to decrease the drawdown
quite a bit, although the standard deviation stayed approximately the same. But
because we believe in the original concept (please say "yes") we will not give u p .
Perhaps we might be able in the end to increase the profits s o m e w h a t by substituting these exits with m o r e generic and less curve-fitted ones, and perhaps the
entry criteria m i g h t even be too strict? Let's take a look.
F r o m Table 11.4 a n d F i g u r e 11.5 we also can see that the b a d n e w s c o n t i n u e s w h e n we l o o k at t h e statistical m e a s u r e s . N o t only did the stops, exits,
a n d profit targets h e l p t o c o m p l e t e l y s c r e w u p the actual results, they also
h e l p e d t o w o r s e n t h e statistical m e a s u r e s , m o s t n o t a b l y w h e n i t c o m e s t o the
still-positive k u r t o s i s a n d the j u s t b a r e l y positive skew. As we can see from
F i g u r e 11.5 it is o b v i o u s that m o s t t r a d e s m o v e against us right off the bat, a
fact m o r e evident b e c a u s e of the way we h a v e split the final profit into fewer
p o s s i b l e o u t c o m e s . As t h i n g s are right now, m o s t t r a d e s will be s t o p p e d out by
the stop loss. A profit p r o t e c t i o n exit also m a k e s sure that m o r e t r a d e s will have
their o u t c o m e s in the 1.5% r e g i o n . B e c a u s e m o s t t r a d e s s e e m to be m o v i n g
a g a i n s t us immediately, however, the profit p r o t e c t i o n stop a n d profit target
s e e m to be c o m p l e t e l y u s e l e s s . T h e trick is to f i g u r e o u t why. T h a t is a q u e s t i o n
for u p c o m i n g c h a p t e r s .
Before we m o v e on with the next strategy, let us see h o w this system would
have fared on older data, leading up to the optimization period. Specifically, we
will look at the period January 1985 to D e c e m b e r 1994. T h e results can be found
in Tables 11.5 and 11.6 and Figure 11.6.
W h a t do you k n o w ! ? All of a s u d d e n the system seems to be working fine,
with an average profit of 0 . 2 5 % and a drawdown no higher than 1 5 % . C o m p a r e d
to the original system, tested over the period January 1995 to October 1999 (see
Table 5.6), though, m o s t n u m b e r s are slightly worse in Table 11.5. O n e simple

TABLE

11.3

Performance summary for the second modified version of Gold Digger II, January 1995-October 1999.
Total trades
Profit factor
Avg profit
StDev

0.07%
2.21%

106
1.08
229
7,446

Winners
Lrg winner
Avg winner
Cum profit

39
5.14%
2.51%
4.72%

36.79%
17,348
8,462
15,930

67
-8.30%
Avg loser
-1.35%
Drawdown -12.01%
Losers

Lrg loser

63.21%
-28,013
-4,564
-40,534

196

PART 3

TABLE

Getting Out

11.4

Descriptive statistics for Gold Digger II (second modified version).

Mean

0.07

Median

-1.00

Kurtosis

1.14

Skew

0.15

Trimmean

0.00

reason for this is the test period, w h i c h is m o r e than twice as long and consequently leaves a lot m o r e r o o m for b a d things to happen.
A positive value for the kurtosis, as c o m p a r e d to Table 11.4, is a positive
because it m e a n s that m o s t o u t c o m e s now are centered around the m e a n . That is,
the system is doing a good j o b doing the s a m e thing over and over again. It is just
not profitable enough. A n o t h e r negative reflection is that the skew is now negative,
indicating that the system is doing a good j o b catching those situations when the
market is poised to take off in a swift m o v e o n l y thing, these moves s e e m to be
in the w r o n g direction. This also is the conclusion to be m a d e from studying the
distribution of returns chart in Figure 11.6.

FIGURE

11.5

The distribution of returns for the second modified version of Gold Digger II system.

CHAPTER 11

TABLE

Working with Random Entries

197

11.5

Performance summary for the second modified version of Gold Digger II, January 1985-December 1994.
Total trades

270

Winners

Profit factor

1.38

Lrg winner
Avg winner
Cum profit

Avg profit

0.25%

859

St Dev

2.00%

6,745

11

42.96%

7.62

25,718

Losers
Lrg loser

2.16

7,283

Avg loser

87.86

296,528

Drawdown

15

57.04%

-12.61

^42,559

-1.18

-3,981

-15.11

-50,996

L o o k i n g at the distribution of trades, we can m a k e two interesting observations. First, most losers are stopped out by the stop loss, and only on occasion does
a trade end up as a loser otherwise. Second, very few trades are stopped out by the
profit limit. T h e big discrepancy in n u m b e r of trades, when looking at trades
stopped out with a loss and trades stopped out at the m a x i m u m profit, indicates
that the m o m e n t u m , w h e n the trade was taken, simply w a s n ' t working in favor of
the trade. We will try to do something about that later. For simplicity's sake, from
n o w on I simply refer to this system as the Gold Digger.

THE MEANDER SYSTEM (WEEKLY DATA)

So far, with the M e a n d e r system we have only looked at daily data. N o w we take
a closer look at the s a m e system using a combination of weekly and daily data.
That is, we will build the indicator and its trigger levels using weekly data and then
try to trade and monitor it using daily data. T h e best way to go about achieving this
is to use the weekly data series as data 2 in TradeStation. A s s u m i n g we are once
again trying to pick.the tops and bottoms, stay with the trade for five days (no matter w h a t ) , and with the trigger levels set to two standard deviations away from the
m e a n , the TradeStation code looks as follows:
Input: VSStd(2);
Vars: SumVS(O), AvgVS(O), DifrYS(O), StdVS(O), SetArr(O), SumArr(O),
DifTArr(O), VSLow(O), VSMid(O), VSHigh(O);
Array: VS[20](0);

TABLE

11.6

Descriptive statistics for Gold Digger II, January 1985-December 1994 (second modified version).
Mean
Median
Kurtosis
Skew
Trimmean

0.25
-1.00
6.00
-.019
0.06

198

PART 3

Getting Out

FIGURE 11.6
The distribution of returns for the second modified version of Gold Digger II, between January 1985-December 1994.

For SetArr = 0 To 4 Begin

VSfSetArr * 4 + 0] = (OfSetArr] Data2 - CfSetArr + 1] Data2) /
CfSetArr + 1] Data2;
VSfSetArr * 4 + 1] = (H[SetArr] Data2 - CfSetArr + 1] Data2) /
C[SetArr + 1] Data2;
VS[SetArr * 4 + 2] = (L[SetArr] Data2 - CfSetArr + 1] Data2) /
CfSetArr + 1] Data2;
VSfSetArr * 4 + 3] = (CfSetArr] Data2 - CfSetArr + 1] Data2) /
CfSetArr + 1] Data2;
End;
For SumArr = 0 To 19 Begin
If SumArr = 0 Then
SumVS = 0;
SumVS = SumVS + VSfSumArr];
If SumArr = 19 Then
AvgVS = SumVS / 20;
For DiffArr = 0 To 19 Begin
If DiffArr = OThen

CHAPTER 11

Working with Random Entries

199

DifTVS = 0;
DiffVS = DiffVS + Square(VS[DiffArr] - AvgVS);
IfDiffArr = 19 Then
StdVS = SquareRoot(DiffVS / 20);
End;
End;
VSLow = C Data2 * (1 + (AvgVS - StdVS * VSStd));
VSMid = C Data2 * (1 + AvgVS);
VSHigh = C Data2 * (1 + (AvgVS + StdVS * VSStd));
If MarketPosition = 0 Then Begin
Buy ("Go long") tomorrow at VSLow limit;
Sell ("Go short") tomorrow at VSHigh limit;
End;
If BarsSinceEntry >= 5 Then Begin
ExitLong ("Long time") at close;
ExitShort ("Short time") at close;
End;
Tables 11.7 and 11.8 and Figure 11.7 show the results for this version of the
Meander, w h e n tested on data for the period January 1995 to October 1999, and
with the trade-by-trade export function from Part 1 attached to the code.
F r o m Table 11.7 we see that this is not a b a d start at all, c o m p a r e d to Gold
Digger at the s a m e stage. A high profit factor is c o m p l e m e n t e d by a high average
trade. T h e standard deviation is somewhat high, but a slightly higher ratio between
the average profit a n d the standard deviation shows that the Meander, at this stage,
m a n a g e s to produce a higher risk-adjusted return than Gold Digger. T h e drawd o w n is a little too high, but could probably be worked on with a well-working stop
loss and some type of filtering technique (to be added later). A n o t h e r positive factor is the positive value for the skew, in Table 11.8, which m e a n s that the Meander,
too lets its profits r u n while cutting the losses short. T h e very high positive value
for the kurtosis is, however, a negative factor at this point, indicating that the individual outcomes are a little too centered around the mean.
TABLE

11.7

Performance summary for the modified version of Meander, January 1995-October 1999.
Total trades

55

Profit factor

1.90

Winners
Lrg winner

29

52.73%

26

47.27%

13.93%

47,014

Lrg loser

-4.65%

-15,694

Avg loser

-1.59%

-5,365

Drawdown - 1 5 . 3 3 %

-51,739

Avg profit

0.67%

2,270

Avg winner

2.70%

9,116

St Dev

3.13%

10,569

Cum profit

41.02%

138,443

Losers

200

PART 3

TABLE

Getting Out

11.8

Descriptive statistics for Meander (modified version).

Mean

0.67

Median

0.10

Kurtosis

5.55

Skew

1.78

Trimmean

0.37

T h e clustering of individual outcomes around, or j u s t below, the mean also

can be seen in Figure 11.7, which shows the distribution of trades. Trades ending
up in this area are likely not to be stopped out by a stop loss or a profit target. The
only way to limit these losses even further is to either exit the trade with a small
profit (if and w h e n that h a p p e n s ) , or to use a time-based stop that forces us to exit
before profits turn even worse. Hopefully, a profit protection stop will do it, as the
high n u m b e r of very profitable trades, at the 3.4 and 5% levels, indicates that the
M e a n d e r is doing a good j o b of working with the m o m e n t u m of the market.
U s i n g the s a m e reasoning as for Gold Digger, but this time assuming that all
o u t c o m e s are evenly distributed around the m e a n , we place a stop loss approxi-

FIGURE

11.7

The distribution of returns for the modified version of Meander.

CHAPTER 11

201

Working with Random Entries

mately 12 trades in both on the left-hand tail and on the right-hand tail. This m e a n s
that we place a stop loss (using the same type of code and principles as for G o l d
Digger) at about 1.8% and a profit target at about 3.2%. Furthermore, using the
profit for the average trade as a guide we also add a m i n i m u m profit level at
0.64%. Tables 11.9 and 11.10, and Figure 11.8 show the results of using these
stops and exits in accordance with the following TradeStation code:
If MarketPosition = 1 Then Begin
ExitLong ("Long profit") tomorrow at 1.032 * EntryPrice limit;
ExitLong ("Long loss") tomorrow at 0.982 * EntryPrice stop;
If Close > 1.0064 * EntryPrice Then
ExitLong ("Long prot") tomorrow at 1.0064 * EntryPrice stop;
End;
If MarketPosition = -1 Then Begin
ExitShort ("Short profit") tomorrow at 0.968 * EntryPrice limit;
ExitShort ("Short loss") tomorrow at 1.018 * EntryPrice stop;
If Close < 0.9936 * EntryPrice Then
ExitShort ("Short prot.") tomorrow at 0.9936 * EntryPrice stop;
End;
A r e we good, or what? As can be seen from Table 11.9, although the average
trade and profit factor declined somewhat, we still m a n a g e d to lower the standard
deviation to 2 . 1 9 % and the drawdown to 1 0 . 8 1 % . T h e n u m b e r of profitable trades
also increased dramatically to 6 5 . 7 5 % . T h e good news continues with a negative
kurtosis. A negative kurtosis m e a n s that we have spread out (or perhaps "plateaued
o u t " is a better expression) the individual outcomes a little m o r e , which could be
interpreted as our target area has grown and, therefore, should be easier to hit.
With the skew in negative territory, however, we are no longer cutting our losses
as short as we probably should. From Figure 11.8, we see that we have transformed
the distribution of returns to resemble Gold Digger. Although this system is even
m o r e distinct, with three very different and clear-cut scenarios that have crystallized: o n e highly profitable, o n e modestly profitable, and one with a negative
return, and with only a few trades scattered around. Either the trade b e c o m e s a
1.8% loser, or a 0.64 or 3.2% winner.
TABLE

11.9

Performance summary for the second modified version of Meander, January 1995-October 1999.
Total trades

73

Profit factor

1.78

Winners
Lrg winner

Avg profit

0.52%

1,764

Avg winner

St Dev

2.19%

7,388

Cum profit

48

65.75%

25

34.25%

4.88%

16,470

Losers
Lrg loser

-5.74%

-19,373

1.81%

6,113

Avg loser

-1.95%

-6,587

43.78%

147,758

Drawdown - 1 0 . 8 1 %

-36,484

202

PART 3

TABLE

Getting Out

11.10

Descriptive statistics for Meander (second modified version).

0.52

Mean
Median

0.64

Kurtosis

-0.56

Skew

-0.03

Trimmean

0.49

T h e interesting thing is that these different scenarios are so distinctly separate from each other that we almost can start calculating the likelihood for a certain scenario to happen. For instance, from Figure 11.8, we can see that there are
a total of 21 out of 73 trades (or 28.8 %) that ended up with a profit of 3.2% or
m o r e , 2 8 . 8 % that ended up with a profit of 0.64%, 3 2 . 9 % that ended up as 1.8%
losers or worse, and only 9 . 5 % that ended up s o m e w h e r e in between. That is, we
k n o w that on close to every third trade, or at least every fourth trade, we should
end up with a 3 . 2 % winner, on average. With the average trade being a 0 . 5 2 % winner, we also can calculate the expected average return for all the trades that fall
outside of the three major scenarios, which will c o m e out to approximately

FIGURE

11.8

The distribution of returns for the second modified version of Meander.

CHAPTER 11

Working with Random Entries

203

0 . 0 4 8 % (or approximately $161 in today's market value). N o t , too bad for what is
falling through the net.
Before we move on with the next system, let us see h o w this strategy would
have performed on data covering the period January 1985 to D e c e m b e r 1994, as
shown in Table 11.11. C o m p a r e d to the optimization period, we did not do quite
as well, b u t even so, we definitely did well enough, with an average profit of
0.52% and a standard deviation of 3 . 0 3 % . Nonetheless, although the out-of-sample data did not completely stand up against those for the optimization period, they
did c o m e in remarkably close. This indicates that the system is very robust and
therefore very likely to hold up in the future as well.
T h e good n e w s also continues w h e n we look at Table 11.12 which once again
shows that the skew has moved into positive territory, indicating that we are once
m o r e letting our profits r u n while cutting the losses short. T h e high value for the
kurtosis, as c o m p a r e d to Table 11.12, also is positive in that it indicates that the
system is doing a very good j o b in producing a very similar set of trades. T h e m e a n
and m e d i a n values for the out-of-sample data are exactly the s a m e as for the insample data, which indicates that the system is working basically the s a m e over
different time periods.
T h i s can also be seen from Figure 11.9, which shows the distribution of
returns. W h e n tested on out-of-sample data, the m i n i m u m profit trades n o w m a k e
up the majority of all trades, while not quite as m a n y trades m a n a g e to be closed
out with a limit order at the profit target level. It is obvious the market has changed
its behavior somewhat, but because the basic characteristics are still there, the
results look very robust and promising. That there are plenty m o r e 1.8% losers
than there are 3.2% winners does not matter as long as the winners are worth so
m u c h m o r e in dollar t e r m s .
W h a t do the M e a n d e r and Gold D i g g e r experiences teach us? It is very p o s sible to c o m e up with a profitable trading system that is likely to w o r k equally as
well on both the optimization period a n d the out-of-sample data, as long as we do
not aim for the stars. A l s o , that if something s e e m s to be too g o o d to be true during optimization, it likely is. But even with such a m o d e s t system as Gold Digger,
we still can work wonders after we improve it by adding a filter that helps us w e e d
out the m o s t obvious losers. A n d there is plenty of h o p e for the M e a n d e r as well.

TABLE

11.11

Performance summary for the second modified version of Meander, January 1985-December 1994.
Total trades

121

Profit factor

1.71

Winners
Lrg winner

Avg profit

0.52%

1,745

Avg winner

St Dev

3.03%

10,232

Cum profit

78

64.46%

43

35.54%

21.92%

73,980

Lrg loser

-11.21%

-37,834

1.93%

6,507

Avg loser

-2.04%

-6,894

69.28%

233,820

Drawdown - 1 3 . 2 8 %

-44,820

Losers

204

PART 3

TABLE

Getting Out

11.12

Descriptive statistics for Meander.

Mean

0.52
0.64

Median
Kurtosis

21.27

Skew

2.62

Trimmean

0.33

THE BLACK JACK SYSTEM

In Part 2 we built a system, which we called Black Jack, the n a m e implying that the
system should be able to exploit minute market discrepancies and that if we only stay
at it long enough we should be able to make a profitnot m u c h from each trade, but
slowly and surely build up a stake worth retiring on. We started with a couple of preset entry and filter criteria, based upon c o m m o n sense and market knowledge.
This time, we will reverse our procedure and start with no such criteria at
all, but instead enter each trade randomly, using TradeStation's random number
generator. By entering randomly, we can test all markets several times and each

FIGURE

11.9

The distribution of returns for the second modified version of Meander between January 1985-December 1994.

CHAPTER 11

Working with Random Entries

205

time c o m e up with a new and unique sequence of trades. For this version of the
Black Jack system, we use the s a m e 16 markets as we did for the Standard
Deviation Breakout ( S D B ) system, covering the time period January 1980 to
D e c e m b e r 1992. To m i m i c approximate general market conditions that keep the
market in a trading range approximately 6 0 % of the time, and in an u p - or downtrend approximately 2 0 % of the time in each direction, we set the r a n d o m n u m b e r
generator to signal a long or short position approximately every fifth day, if w e ' r e
not in a position already.
Basically, there are two slightly different ways to go about building a system
like this. You could examine for one type of exit at a time, starting out with the
time-based stop, which later could be removed. (With all trades entered randomly
there invariably will be situations w h e n we need s o m e type of exit to take us out
of the market no matter what the underlying market conditions look like at the
time.) Or you could examine all exits in pairs, starting out with the time-based stop
and the stop loss. This is what we do in this example. To do this, we must know
the value of the average trade for each market and each stop loss/trade length c o m bination, for every time we run that particular combination on that particular m a r ket. R e m e m b e r , thanks to the fact that each trade will be taken randomly, we can
run the s a m e test on the s a m e market as m a n y times as we wish and always c o m e
up with a new and different result.
It is a g o o d idea to start out with the stop loss/trade length combination
because if it happens that we enter exactly w h e n the market takes off in one direction and we happen to be right, the m a x i m u m trade length shows us for how m a n y
bars (days, hours, etc.) the market r e m e m b e r s a certain event or whatever external
factor that triggered the move. This is the amount of time we would like to stay
with the trade, no more, no less. W h e n the market starts to discount other factors
as m o r e important, we want to exit and wait for the next move that might c o m e our
way, because at the point of exit, the reason for our entry is no longer there. T h e
stop loss is obviously necessary to tell us when we are most likely to be wrong,
whatever the reason for entering in the first place.
Because we want this to be a short-term system, we restrict our search to the
area between two to ten days. I chose to go with two days instead of o n e day
because of the s a m e TradeStation b u g I already have described and will rant further about below. As for the stop loss, we k n o w from experience that we probably
are best off looking in the 0.5 to 2% region, so this is what we will d o .
T h e quickest way to do the search is to declare both the trade length and the
stop loss as inputs, run t h e m through the optimization procedure and, for each run
through the data, export the necessary information into a spreadsheet program for
further analysis. Do not bother using the optimization report in TradeStation,
because it is tailored to run on point-based back-adjusted contracts only. In the
optimization dialog boxes we tell TradeStation to r u n through every combination
of our input variables, from two to ten, in steps of one, for the m a x i m u m trade

206

PART 3

Getting Out

length, a n d from 0.5 to two, in steps of 0 . 1 , for the stop loss.

We also declare a third input variable, which we can use to r u n the same stop
loss/trade length combination as m a n y times as we wish. I run each combination
ten times. This results in 1,440 loops for each market. With 16 markets to test, each
one holding about ten years' w o r t h of data, this produces something like 230,000
u n i q u e yearly trading sequences. I have not bothered to count how m a n y trades we
will do, but hopefully this will be e n o u g h to form s o m e solid statistical conclusions. Note: This takes a hell of a lot of time, even on the fastest of computers. I
work on a dual processor, 5 0 0 M H z Pentium III, 2 5 6 M b memory, a n d I still have
a few hours to go do other things when I do runs like this.
If you perform the analysis by exporting the result for each trade separately,
the testing b e c o m e s extremely slow (now we're talking days), but you also generate so m a n y data that a single Excel spreadsheet w o n ' t be large e n o u g h to hold it
all. (One Excel spreadsheet has close to 17 million cells [65,536 rows X 256
columns].) T h e TradeStation c o d e for this and all other runs follows:
Inputs: Counter(O), TradeLength(7), StopLoss(l.l),TrailingStop(O.l),
ProfitTarget(2);
Vars: PositionGenerator(O), TotTr(O), MP(0), LowestLow(O), HighestHigh(O),
MAE(0), MFE(0), FTE(0), ETD(0), SumMAE(O), SumMFE(O), SumFTE(O),
,m

SumETD(O), FName( ), TradeStrl("");

PositionGenerator = IntPortion(Random(5));
If MarketPosition = 0 Then Begin
If PositionGenerator = 3 Then
Buy at Close;
If PositionGenerator = 4 Then
Sell at Close;
End;
If BarsSinceEntry > = 1 Then Begin
{ ExitLong ("Long Target") at Entry Price * (1 + (ProfitTarget * 0.01)) limit;
ExitShort ("Short Target") at EntryPrice * (1 - (ProfitTarget * 0.01)) limit;
If Close > EntryPrice * (1 + (TrailingStop * 0.01)) Then
ExitLong ("Long Trailing") at EntryPrice * (1 + (TrailingStop * 0.01))
stop;
If Close < EntryPrice * (1 - (TrailingStop * 0.01)) Then
ExitShort ("Short Trailing") at EntryPrice * (1 - (TrailingStop * 0.01))
stop;}
ExitLong ("Long Loss") at EntryPrice * (1 - (StopLoss * 0.01)) stop;
ExitShort ("Short Loss") at EntryPrice * (1 + (StopLoss * 0.01)) stop;
End;

CHAPTER 11

Working with Random Entries

If BarsSinceEntry >= TradeLength Then Begin

ExitLong ("Long Time") at Close;
ExitShort ("Short Time") at Close;
End;
TotTr = TotalTrades;
MP = MarketPosition;
If MarketPosition = 1 Then Begin
If BarsSinceEntry = 1 Then Begin
LowestLow = EntryPrice;
HighestHigh = EntryPrice;
MAE = 0;
MFE = 0;
End;
If Low < LowestLow Then Begin
LowestLow = Low;
MAE = (LowestLow EntryPrice) / EntryPrice;
End;
If High > HighestHigh Then Begin
HighestHigh = High;
MFE = (HighestHigh - EntryPrice) / EntryPrice;
End;
End;
If MarketPosition = 1 Then Begin
If BarsSinceEntry = 1 Then Begin
LowestLow = EntryPrice;
HighestHigh = EntryPrice;
MAE = 0;
MFE = 0;
End;
If High > HighestHigh Then Begin
HighestHigh = High;
MAE = (EntryPrice HighestHigh) / EntryPrice;
End;
If Low < LowestLow Then Begin
LowestLow = Low;
MFE = (EntryPrice LowestLow) / EntryPrice;

207

PART 3

208

Getting Out

End;
End;
If CurrentBar = 1 Then Begin
FName = "D:\Temp\BJS.csv";
End;
If TotTr > TotTr[l] Then Begin
I f M P [ l ] = I T h e n Begin
FTE - (ExitPrice(l) - EntryPrice(l)) / EntryPrice(l);
ETD = (ExitPrice(l) - HighestHigh[l]) / Highesthigh[l];
End;
I f M P [ l ] = -1 Then Begin
FTE = (EntryPrice(l) - ExitPrice(l)) / EntryPrice(l);
ETD = (LowestLc-wfl] - ExitPrice(l)) / LowestLowfl];
End;
If FTE < MAE[1] Then
MAE = FTE Else MAE = MAE[1];
If FTE > MFE[1] Then
MFE = FTE Else MFE = MFE[1];
SumFTE = SumFTE + FTE;
SumETD = SumETD + ETD;
SumMFE = SumMFE + MFE;
SumMAE = SumMAE + MAE;
End;
If LastBarOnChart Then Begin
FTE = SumFTE / TotalTrades;
ETD = SumETD / TotalTrades;
MFE = SumMFE / TotalTrades;
MAE = SumMAE/TotalTrades;
TradeStrl = LeftStr(GetSymbolName, 2) + "," + NumToStr(MAE, 4)+
V + NumToStr(MFE, 4) + "," + NumToStr(FTE, 4) + "," +
NumToStr(ETD, 4) + "," + NumToStr(TradeLength, 2) + "," +
NumToStr(StopLoss, 2) + "," + NumToStr(TrailingStop, 2) + "," +
NumToStr(ProfitTarget, 2) + NewLine;
FileAppend(FName, TradeStrl);
End;
This code exports the M A E , M F E , a n d C T D values, together with the average profit, and the value of the input parameters for that particular run, each time

CHAPTER 11

Working with Random Entries

209

the p r o g r a m loops through the market. For our p u r p o s e s , we only need the average
profit and the value for the inputs to keep track of all combinations. T h e others are
there for other, but similar, analysis techniques, such as w h e n exporting each individual trade or examining one exit technique at a time. O n c e we are done testing,
we substitute for the r a n d o m trade generator a high probability entry technique,
k n o w i n g that we n o w should have a system that lets us trade marginally profitably
or at least does not force us into too large disasters, no matter w h a t the underlying
market conditions look like.
T h e code above also illustrates another TradeStation feature. As you can see,
for the stop loss we use the criteria If BarsSinceEntry 3= 1. This really should not
be necessary, but without this line TradeStation sometimes exits the long on the
first o p e n following the entry, whether the trade is (or might have been) a w i n n e r
or loser.
Because of this extra piece of code, we encounter another problem, which I
already ranted about. That is, on the day you wait for this condition to c o m e true
(the first whole day in the trade, if you entered at the close the day before) the syst e m does not recognize that y o u have a trade going. O n e way to work around this
is to declare your o w n entry price variables, but that causes you a multitude of
other p r o b l e m s further d o w n in the code, like in the export function and on, and
on, a n d o n . . . .
Okay, we m u s t move on. After you have exported all the necessary data for
the trade length and the stop loss into your spreadsheet program, you can use the
s a m e charting technique as we did in Part 2, w h e n we examined different moving
average length combinations. A surface chart is an excellent way to get a feel for
how the output variable reacts to two different input variables a n d how the inputs
are interacting with each other. T h e first chart, Figure 11.10, shows the average
percentage profit as a function of the stop loss in percent and the trade length in
days. For instance, with a stop loss of 1.3% and a trade length of eight days, the
average profit c o m e s out to somewhere between 0.1 and 0%. T h e trick is to try
to find a squared area that is as large as possible, with a value for the profitable
trade that is as high as possible. In Figure 11.10 there are two areas that look particularly promising. T h e first one is around the 0 . 7 % stop loss a n d the other one
around the 1.1 to 1.2% stop loss. T h e area surrounding the eight-day trade length
and the 1.1% stop loss looks very interesting.
Figure 11.11 shows the standard deviation of the returns. In this chart we
would like to find an area that is as large as possible, but has a value that is as low
as possible. In this case we can see that, generally, the longer the trade length and
the w i d e r the stop loss, the higher the standard deviation. Obviously, we also need
the standard deviation that we choose to coincide with an area of positive returns
in Figure 11.10. T h e two charts in combination tell us that if we have two similar
alternatives to choose from w h e n it c o m e s to the average profit, we should select
the one with the narrowest stop and shortest trade length.

210

PART 3

FIGURE

11.10

The percentage return from random entries as a function of the trade length and stop loss level.

FIGURE

11.11

The standard deviation of the percentage returns as a function of the trade length and stop loss level.

Getting Out

CHAPTER 11

Working with Random Entries

211

Figure 11.12 shows the kurtosis for each stop loss/trade length combination.
H e r e we also would like to find an area as large as possible with a value as low as
possible and preferably also in negative territory. In this case too, we can see that,
generally, the longer the trade length and the w i d e r the stop loss, the higher the
kurtosis. Finally, Figure 11.13 shows the skew, w h i c h should be as high as possible. In this case, it o n c e again seems that the streak around the 1.2% stop loss looks
m o s t promising. In this case, it b e c o m e s difficult to keep it negative. It also s e e m s
that the kurtosis generally behaves the opposite way from the standard deviation.
My rule-of-thumb is to give a higher consideration to the standard deviation and
give the kurtosis the absolute lowest consideration.
To find the individual stop loss or trade length that is likely to be the most
robust input parameter, no matter what h a p p e n s with all the other input c o m b i n a tions, we also can look at separate kurtosis and skew tables for each input. Tables
11.13 and 11.14 show w h a t these measures look like for each trade length applied
to all the different stop loss values, a n d each stop loss value applied to every trade
length, respectively. F r o m Table 11.13 we can see that the skew increases with the
trade length, which m e a n s that the longer the trade, the better off we seem to be.
A similar relationship does not to exist for the kurtosis. W h e n it c o m e s to the stop
loss both the kurtosis and the skew p e a k around the 1.2% level. This is good w h e n
it c o m e s to the skew, but bad from a kurtosis point of view.
F r o m Figures 11.10 through 11.13 and Tables 11.13 and 11.14 it seems as
t h o u g h a stop loss/trade length combination of 1.1 to 1.2% and seven to eight days

FIGURE

11.12

The kurtosis of the returns as a function of the trade length and stop loss level.

PART 3

212

FIGURE

Getting Out

11.13

The skew of the returns as a function of the trade length and stop loss level.

has a lot going for itself. R e m e m b e r these values when we m o v e on to c o m p a r e the

stop loss in relation to the profit target. N o t e that despite all the fancy techniques
and statistical m e a s u r e m e n t s , this is w h e r e the system-building profession still is
m o r e of an art form than a science and where previous experience, both as a systems designer and as a trader, plays a vital role for the end result. It also is important to always keep in mind w h o you are and if you actually are psychologically
able to trade and cope with the system that you are about to put together.

TABLE

11.13

Kurtosis and skew for the trade length.

Trade length

Kurtosis

Skew

2.0000

3.5569

-0.7034

3.0000

3.8781

-0.7326

4.0000

4.8803

-0.3035

5.0000

4.3825

-0.3015

6.0000

3.1974

0.6934

7.0000

7.0289

1.0748

8.0000

2.6979

1.0727

9.0000

4.4220

0.7154

10.0000

3.8914

1.2584

CHAPTER 11

213

Working with Random Entries

TABLE

11.14

kurtosis and skew for the stop loss.

Stop loss

Kurtosis

Skew

0.5000

4.0508

1.0517

0.6000

3.8028

1.1058

0.7000

2.7710

0.9926

0.8000

3.7190

1.1055

0.9000

5.0752

1.0647

1.0000

7.9998

2.0985

1.1000

10.0930

0.3240

1.2000

11.6853

2.4169

1.3000

4.2248

1.0409

1.4000

3.1491

0.8757

1.5000

2.1659

0.9603

1.6000

9.7810

2.0154

1.7000

2.9646

0.5511

1.8000

4.1007

-0.4604

1.9000

8.6028

1.2669

2.0000

4.8537

0.4595

Figure 11.14 shows the average return as a function of the profit target and the
stop loss. As you can see, the profit target was tested for all values between 1 and
4% in increments of 0.2%. T h e profit target/stop loss tests were m a d e the same way
as the stop loss/trade length tests, but this time for a total of over 400,000 unique
yearly trade sequences, which took forever and a day. From Figure 11.14 y o u can
see that generally, the m o r e distant the profit target and the wider the stop loss, the
better the results. B u t there also are two smaller areas around the 1% stop loss that
could be worth looking into: o n e with a profit target of about 2 . 8 % , and the other
with a profit target around 3.4%. The average return seems to benefit from keeping
the distance between the stop and the profit target as w i d e as possible. To reach
these returns, however, we m u s t take on proportionally m u c h more risk that if we
choose to keep the profit target down to around 2 . 8 % . For instance, to increase the
return for the profit target by 2 1 % , from 2.8 to 3.4%, we must increase the risk by
6 0 % , from 1 to 1.6%>, which is not such a desirable trade-off.
With several alternatives like this, y o u m u s t sit d o w n and figure out which
one will give you the m o s t b a n g for the buck. That is, which combination will give
y o u the highest average profit and at the s a m e time keep the profit potential open
for large positive out Iyer trades, while also keeping the lid on the losses? O n e
advantage of the riskier combination, with the 3 . 4 % profit target and 1.6% stop
loss, is that it is surrounded by several other high-value areas, which m e a n s that it
probably will continue to produce good results even if the true best values happen

214

PART 3

FIGURE

Getting Out

11.14

The percentage return from random entries as a function of the profit target and stop loss levels.

to move away from w h e r e we think they are. But, as noted, there is a price to everything in the market and only you can m a k e the choices most suitable for you.
To m a k e a choice like this, it also could be a good idea to look at the standard deviation chart which, in this case, shows that the standard deviation also
increases the w i d e r the stop loss and the m o r e distant the profit target (chart not
shown). Other charts and tables, such as those for kurtosis and skew also indicate
that we probably would be better off with a stop loss s o m e w h e r e in the 0.9 to 1.1%
region, which also confirms our findings from the stop loss/trend length examination. Let us go with a stop loss of 1.1% and a profit target of 2 . 8 % .
To c o m e up with a trailing stop or m i n i m u m profit level we will r u n it against
the time-based stop, testing all values between 0.2 and 1% in increments of 0 . 1 % .
This results in m o r e than 125,000 unique yearly trade sequences. Parts of the result
from this r u n can be seen in Figure 11.15, which shows the average profit in relation to different m i n i m u m profit levels and time-based stops. W h e n testing for the

CHAPTER 11

Working with Random Entries

215

m i n i m u m profit level, the criterion is that the market m u s t have closed above a certain open profit and then moved below this level on a later bar for a trade to be
closed out. A m i n i m u m profit of 0 . 6 % produces fairly robust results, no matter
what the m a x i m u m trade length h a p p e n s to be. However, a m a x i m u m trade length
of seven to nine days seems to be the most robust t i m e window. This confirms
what we already suspected when we analyzed Figure 1 1 . 1 0 .
W h e n it c o m e s to the m i n i m u m profit level, you might also say that the market m u s t close with a low above the desired level for all long trades or the high
below the desired level for all short trades. This generally gives the system slightly m o r e slack and results in fewer trades getting closed out at the m i n i m u m profit level, slightly m o r e trades getting closed out at the m a x i m u m profit target, and
a bunch of trades s o m e w h e r e in between because of the time-based stop. There
also will be a b u n c h of trades closed out with a loss, because they never m a n a g e d
to fulfill the m o r e stringent criterion for the profit protection stop to kick in.

FIGURE

11.15

The percentage returns from random entries as a function of the trade length and minimum profit level.

216

PART 3

Getting Out

As you can see from Figure 11.15, it s e e m s that our best alternative is a mini m u m profit level of 0.6% together with a trade length between seven to nine days.
This confirms our previous findings about an eight-day m a x i m u m trading horizon.
T h u s , the final system has the following exit rules:

Exit with a loss if the market moves the w r o n g way by 1.1%

Exit with a profit if the market closes with a 0.6% open profit and then moves
below this level.

Exit with a limit order if the o p e n profit exceeds 2.8%).

Always exit at the close at the eighth day of the trade.

T h e summarized results from adding these exits to the system, together with
the export function for the R A D contract, can be seen in Table 11.15. As you can
see, the average final profit is only one hundredth of a percent. It isn't m u c h , but at
least it's there. Of course, an expected return this small is not tradable, but the point
is that just by using a set of logical exits and stop techniques, we managed to produce a slightly positive average return. We h o p e to boost the results by applying this
set of exits to one or several high-probability entry techniques. Another interesting
observation is that the M F E value is a great deal larger than the absolute value for
the M A E level. All in all, this indicates that the stops a n d exits indeed manage to
create a positive bias out of nothing. (Please r e m e m b e r that we are still placing all
trades randomly and with no concern for the underlying trend.)
Figure 11.16 shows w h a t the average trade could look like, assuming that the
sequence of events is as outlined. T h a t this does not necessarily have to be so can
be seen both from the relatively low standard deviation interval for the close and
the fact that the standard deviation intervals for the M A E a n d M F E seem to be
approximately the same. ( R e m e m b e r that it is reasonable to a s s u m e the standard
deviation should increase the further away we get from the entry.) Whatever the
case, please note h o w narrow the standard deviation interval is; that m e a n s that,
aside from the actual outcome, the exits are doing a very good j o b in streamlining
the trades.
T h e next step in building a system using the Black Jack technique is obviously to attach it to a high-probability entry technique. Let us see h o w these exits
would have fared with the short-term entry techniques already discussed. Table

TABLE

11.15

Trade characteristics for Black Jack system exits.

MAE
Average
St Dev

MFE

Final profit

-1.0841

1.3200

0.0140

0.3192

0.3652

0.1203

CHAPTER 11

Working with Random Entries

F I G U R E

217

11.16

The summarized life for the average random trade, using the Black Jack exits.

11.16 shows the results for the Gold D i g g e r entry technique, traded on the S & P
500, for the period January 1995 to October 1999. T h i s table can be c o m p a r e d
directly to Table 11.3, but before you do that, also r e m e m b e r that w h e n we put
together the Black Jack exits we didn't even look at the S & P 500. If this system
w o r k s on the S & P 500 as well, it m e a n s that we are exploiting a type of market
discrepancy that is the s a m e for all markets, and therefore, is likely to be profitable
on a w i d e variety of m a r k e t s as long as the point values are high enough.
A n d what do you know! This set of exits worked even better than the originals. First, we have a m u c h higher average trade, although it is still slightly on the
low side. This is c o m p e n s a t e d for by a lower standard deviation, which m e a n s that
the risk-adjusted return n o w is m u c h higher. T h e cumulative profit also is m u c h

T A B LE

11.16

Results for Gold Digger/Black Jack, January 1995-October 1999.

Total trades

141

Profit factor

1.26

Lrg winner
Avg winner

Avg profit

0.17%

569

St Dev

1.88%

6,348

Winners

Cum profit

78

55.32%

5.14%

17,348

Lrg loser

1.49%

5,031

Avg loser

23.68%

79,920

Drawdown

Losers

63

44.68%

-8.30%

-28,013

-1.47%

-4,955

-15.66%

-52,853

PART 3

218

Getting Out

higher, as is the percentage of winning trades. On the negative side, we see that the
drawdown is slightly higher as well.
W h e n we built the Gold D i g g e r system, we used the period January 1985 to
D e c e m b e r 1994 for s o m e out-of-sample testing, which naturally put the lid on the
results w h e n c o m p a r e d to the optimization period. T h e out-of-sample period was,
furthermore, m u c h larger in size, w h i c h m e a n t that it should hold m o r e unfavorable m a r k e t conditions. Therefore, it is likely that a comparison with that period
will c o m e closer to the truth regarding what the Gold Digger/Black Jack combination really is worth. C o m p a r e Tables 11.17 a n d 11.5.
The truth is that if you have built a system exactly right, any market condition that does not exactly parallel the conditions under which you built the system
will be unfavorable. Because if you have built a system exactly right, you will have
squeezed out every o u n c e of information the market holds and the system will
therefore have a positive bias, but behave as if the individual results were dropping
in randomly. T h e reason for this is, as long as the system does not behave in this
r a n d o m fashion, the outcome is predictable, and if something is predictable there
is m o r e information to squeeze out. If the conditions have changed from one time
to another that m e a n s that there is m o r e and newer information to squeeze out, and
the system will now perform on a suboptimal leveleven if it temporarily might
produce better-than-expected results.
T h e above reasoning also puts a twist to another interesting question. Is there
such a thing as a system that (seemingly) could be doing too well? For instance,
say that you have put together a system that is expected to p r o d u c e an average
profit of 1%. B u t after you have traded the system for a while you realize that the
average profit, in fact, has b e e n closer to 2 % . W h a t to do? T h e answer is, examine
the system closely a n d try to find out w h a t is causing these excess returns and
adjust the system to take these causes into consideration. Because the truth is that
w h e n the system is not w o r k i n g as intended, you do not know what you are doing;
what might cause excess returns might also be the reason why you have to sell
your h o u s e tomorrow.
Tested over the period January 1985 to December 1994, the Gold Digger/
Black Jack combination has a lower profit factor and average trade than the Gold
Digger system alone. W h e n it comes to the percentage profitable trades and standard
deviation, on the other hand, Gold Digger/Black Jack excels. It also is interesting to
TABLE

11.17

Results for Gold Digger/Black Jack, January 1985-December 1994.

Total trades

292

Winners

Profit factor

1.20

Lrg winner
Avg winner

Avg profit

0.12%

407

St Dev

1.74%

5,864

Cum profit

159

54.45%

7.62%

25,718

1.31%

4,412

35.80%

120,825

Losers
Lrg loser
Avg loser

133

45.55%

-12.61%

-42,559

-1.30%

-4,381

Drawdown - 1 5 . 8 8 %

-53,595

CHAPTER 11

Working with Random Entries

219

note that the results for the out-of-sample period no longer differ that m u c h from
those of the in-sample period, which m e a n s that, although the profit potential is
somewhat low, at least the results are very robust over time.
For the M e a n d e r / B l a c k Jack combination, the results are a little m o r e inconclusive. Table 11.18 shows h o w this system would have performed during the
optimization period, and is directly comparable to Tabic 11.9. As you can see, this
system has a slightly lower profit factor and average profit per trade. T h e percentage of profitable trades also is lower. On the positive side, however, there is a quite
dramatic decrease of m a x i m u m drawdown, from an already low 1 0 . 8 % to a very
low 7 . 8 % .
W h e n it is tested on the out-of-sample period, two sad things happen with the
M e a n d e r / B l a c k Jack combination. T h e m o s t obvious thing, w h e n we c o m p a r e
Tables 11.19 and 1 1 . 1 1 , is that it continues to be outperformed by the original
M e a n d e r system. W h a t is even worse is that the results for the out-of-sample period are no longer similar to those from the in-sample period. As you can see, the
average profit n o w has decreased to 0 . 2 2 % , while the profit factor is down to 1.32.
T h e drawdown also has increased to 15.8%. O n e positive thing is the slightly lower
standard deviation, as c o m p a r e d to the original system, but because the ratio
between the average profit and the standard deviation still c o m e s out m u c h lower,
this does not help m u c h .
T h e B l a c k J a c k exits are not as g o o d together with the M e a n d e r system as
those developed specifically for this system. T h e question we then m u s t ask ourselves is if we should stick to the original exits because we k n o w they work well
with this particular system, or switch to the Black Jack exits, because they should
work well, on average, with any type of entry technique?
To answer this question, we m u s t go back and look at the differences between
the two sets of exits. In doing so, we can see that there are three major differences.
First, the profit target is m u c h further away from the entry in the original set of
exits. Second, the original set of exits also operates without a time-based stop. This
m e a n s that we are letting the profits r u n " m o r e fluently," so to speak, with the original set of exits, m a k i n g the m o s t of w h a t s e e m s to be a very strong entry technique. Finally, the stop loss is also further away from the entry in the original set
of exits, w h i c h m e a n s that we are putting m o r e trust into the entry signal. This confirms the reasoning behind points n u m b e r one and two, because the better the
TABLE

11.18

Results for Meander/Black Jack, January 1995-October 1999.

Total trades

79

Profit factor

1.67

Winners
Lrg winner

Avg profit

0.39%

1,321

Avg winner

St Dev

1.90%

6,379

Cum profit

44

55.70%

35

44.30%

4.88%

16,470

Lrg loser

-5.74%

-19,373

1.75%

5,902

Avg loser

-1.32%

-4,440

34.24%

115,560

Drawdown

-7.76%

-26,190

Losers

PART 3

220

TABLE

Getting Out

11.19

Results for Meander/Black Jack, January 1985-December 1994.

Total trades

136

Profit factor

1.32

Lrg winner
Avg winner

Avg profit

0.22%

731

St Dev

2.76%

9,299

Winners

Cum profit

52.21%

47.79%

21.92%

73,980

Lrg loser

-11.21%

-37.834

1.72%

5,793

Avg loser

-1.42%

-4,799

25.64%

86,535

Drawdown - 1 5 . 7 9 %

-53,291

Losers

entry signal, the m o r e trust we should give it a n d not w o r r y so m u c h if the market

does not happen to take off in the right direction immediately.
T h e answer then, s e e m s to be to stick to the original set of exits. This is also
a matter of personal choice. Personally I really like the reasoning behind the Black
Jack exits. Therefore, that's the set I will stick with, under the assumptions that
these things will even out over time and if there is one thing that the version with
the Black Jack exits is not, that's curve fitted. ( B y the way, there is nothing keeping you from trading each entry technique with several sets of exit techniques,
anyway.)
Finally, let us c o m p a r e this new set of exits with those that we created for the
original Black Jack system in Part 2. Table 11.20 shows what this Black Jack/
Black Jack combination looks like w h e n traded on the S & P 500 over the period
January 1995 to October 1999, while Table 11.21 shows the results for the period
January 1985 to D e c e m b e r 1994. N o t e that neither table is directly comparable to
Table 5.14 w h i c h shows the results for both these periods combined.
However, j u s t by c o m p a r i n g Tables 11.20 and 11.21 we can c o m e to a few
interesting conclusions, the most obvious being that w h a t worked during one time
period did m o s t definitely not work in the other. This is very interesting and
b e c o m e s even m o r e so if we also mix in what we discovered about the original
Gold D i g g e r system in Tables 11.3 and 11.5, w h e r e the scenario w a s exactly the
opposite.
Obviously, h a d we put together an entry/exit combination like Black
Jack/Black Jack, with a profit factor of 1.03 and an average profit per trade of
0 . 0 8 % in early 1995, we would not have traded it. Instead, it is quite reasonable
to assume that we would have started to trade the Gold Digger/Black Jack combi-

TA B L E

11.20

Results for Black Jack/Black Jack, January 1995-October 1999.

Total trades

87

Profit factor

2.16

Avg profit
St Dev

0.50%
1.59%

1,695
5,352

Winners
Lrg winner
Avg winner
C u m profit

59

67.82%

28

32.18%

3.09%

10,429

Lrg loser

-2.90%

-9,788

1.38%

4,654

Avg loser

-1.35%

-4,541

-4.01%

-13,534

53.00%

178,875

Losers

Drawdown

CHAPTER 11

TABLE

Working with Random Entries

221

11.21

Results for Black Jack/Black Jack, January 1985-December 1994.

Total trades

202

48.02%

10

51.98%

Profit factor

0.89

Lrg winner

4.12%

13,905

Lrg loser

-6.29%

-21,229

-0.08%

-257

Avg winner

1.24%

4,199

Avg loser

-1.30%

-4,373

1.54%

5,197

Cum profit - 1 6 . 3 5 %

-55,181

Drawdown - 2 6 . 0 5 %

-87,919

Avg profit
St Dev

Winners

Losers

nation. But with the help of hindsight, we n o w k n o w that we in fact should have
thrown away the Gold Digger/Black Jack combination and started to trade the
Black Jack/Black Jack combination.
So what does this teach u s ? It teaches us that it is p a r a m o u n t to m a k e sure
that the system works, on average, equally as well over several time periods and
markets, and that the input parameters we are using are robust e n o u g h to be trusted. F r o m j u s t looking at the results for the period leading up to D e c e m b e r 1994,
we h a d no way of c o m i n g to that conclusion. It also tells us that even a market/syst e m combination that we trust completely will go through dry spells and that we
never should place all our eggs in o n e basket (more about that in Part 5).
B e c a u s e a s already pointed o u t w h e t h e r a system works better or worse than
expected, in both cases you are taking a w a l k on the wild side, and exactly the
s a m e environment that s e e m s to be so friendly this very m o m e n t will eat you alive
in five m i n u t e s if you haven't m a d e sure that you have other alternative routes of
action w h e n danger strikes.

PART

THREE

A Few Final Thoughts About Part 3

In this part, we discussed the various forms of d r a w d o w n with w h i c h you m u s t be

familiar, or at the very least be aware of, when y o u are examining your trading systems. If you are not aware of where and w h e n the drawdown happens, you do not
k n o w what m e a n s you m u s t use to c o m e to grips with it. T h e n it does not matter
h o w m u c h you might think you k n o w about other, related topics such as trade efficiency, m a x i m u m adverse excursion, and m a x i m u m favorable excursion.
After we explored t h o s e topics as well, we set out to better the s y s t e m s develo p e d in Part 2 by a d d i n g a set of tailor-made stops and exits to e a c h system, b a s ing a lot of our findings on r a n d o m entries, derived with the help of TradeStation's
r a n d o m n u m b e r generator. ( B y the way, if you are using an older version of
TradeStation, you should be able to d o w n l o a d a r a n d o m n u m b e r generator, courtesy of Dave D e L u c a at Trade Works Software, at the following Internet site:
www.mechtrading.com/tradestation/random.html.)
In Part 3, we also introduced a set of new statistical measures to help us evaluate a system's performance quickly and efficiently. In particular, we m a d e use of
kurtosis and skew. T h e s e two m e a s u r e m e n t s will continue to be used in Part 4,
where we take a closer look at different ways of filtering out favorable, long-term
market situations.
I h o p e you also have noticed h o w almost every single system has been applied
to a broad selection of markets, and that I have used almost no market-specific
parameter settings whatsoever. True, m o s t of the short-term systems have only been
applied to the S & P 500, but the reason for that is simply that the S & P 500 is one
of the few markets that is worth trading short-term.
R e m e m b e r a well-working system does not have to be a profitable system,
b u t a profitable system also is a well-working system. T h e difference lies in how we
223

224

PART 3

Getting Out

define well-working a n d profitable. By well-working, we m e a n a system that does

a g o o d j o b catching a specific type of m o v e in several different markets, ending
up with a positive return m e a s u r e d in percentage t e r m s . We use percentage terms,
because that is a universal m e a s u r e that allows us to ignore other market technicalities such as the point value or the current trading level.
By a profitable system, on the other hand, we m e a n a well-working system
that achieves a positive return in dollar terms, as well as in percentage terms.
B e c a u s e all markets are trading at different levels and with different point values,
however, not all well-working s y s t e m s will be profitable on all markets. As far as
the system goes, it all c o m e s down to w h a t type of m o v e we w a n t it to catch. But
p e r h a p s even m o r e important, no system will trade profitably if the necessary prerequisites are not there, namely a high e n o u g h dollar value p e r point in relation to
w h e r e the m a r k e t is trading and low e n o u g h costs in relation to this level.

PART

FOUR

High-probability Filters

To continue to trust your trading strategy after a long d r y spell, especially if at the
same time, the market is breaking out to record highs or lows, is very difficult a n d
probably the first reason why so m a n y of us fail as traders. If only there was a way
to tell w h e n to take the signal, w h e n not to take the signal or, heaven forbid, perhaps
even w h e n to fade the signal. Basically, there are two ways of filtering out high-probability trading opportunities and the question y o u should ask yourself is, "Is it better to trade with the long-term trend a n d can the short-term volatility help to determ i n e the outcome of the t r a d e ? " This is an important question that all traders should
ask themselves, and the only way to find out is to do the necessary research.

225

CHAPTER

12

Filtering

Besides filtering, indicator piling is especially popular a m o n g n e w traders a n d a

horrifying n u m b e r of broker-type salespeople w h o , in one way or another, m a k e
their living explicitly out of you instead of out of the markets. Indicator piling is
executed by piling several similar indicators and "expert c o m m e n t a r i e s " on top of
each other in the high hopes that one day, w h e n they all move in unison, y o u will
have a sure winner. Never m i n d that they all give different answers, a n d with 10
indicators on your screen and only one being the "right one," you actually have a
9 0 % chance of c o m i n g to the w r o n g conclusion if you l o o k at each one at a time;
not to mention w h a t the odds are if you c o m b i n e t h e m all.
Nonetheless, this inefficient a n d risky method is favored by m o s t trading p r o g r a m vendors. A r e n ' t these guys pros, you might ask, shouldn't they k n o w better?
Well, they might or might not k n o w better, but the only way they can m a k e m o n e y
out of the wanna-be trader is to sell him their trading program. A n d the fastest way
they can do that is to show h i m the fancy looking indicator-piling technique. (At
least this is my experience from m o r e than three years of watching these guys and
their marketing tactics from a very close distance.) If you still are fairly n e w to this
game, please do not buy into this method. You do not need all those indicatorseven
though vendors compete with each other to develop the program that has the most.
O n c e you're past the piling stage (which you probably are, since you've read this
far), the most obvious thing to test for first is to see if there is anything to be gained
from only going with the longer-term trend. T h e longer-term trend can be determined
in several different ways, the simplest (but certainly not the easiest) being your own
fundamental and discretionary j u d g m e n t about where the economy in general, and
your markets or stocks in particular, are heading. Then stick to that prognosis over a
prolonged period of time. (Presented later is a m o r e thorough discussion of what
227

PART 4

228

High-probability Filters

makes a trend and how this knowledge might add a m o r e theoretical, or even philosophical, aspect to the way you might perceive your role as a market participant and
what it is you are trying to achieve.)
Alternatively, you could simply only trade in the s a m e direction as a longt e r m moving average. In a series of articles for Futures m a g a z i n e , " S k i m m i n g the
T r e n d " a n d "Fading Away," (February 1999), a n d "In the Pudding," (March 1999),
and in the April 2 0 0 0 issue of Active Trader Magazine, I concluded that trading
only in the direction of a 200-day moving average does, indeed, increase results
significantly. This holds true even if you enter the markets randomly, as long as
you take trades only in the direction of the underlying trend and stick to the same
exit strategy for all trades.
Table 12.1 shows the result of entering 16 different markets randomly in any
direction over the time period January 1980 to October 1999, as compared to
entering randomly but only in the s a m e direction as a 200-day moving average.
Each market w a s traded 12 times, creating a total of close to 3,500 unique yearly
trading sequences per strategy. In both strategies, I stayed in each trade for five
days, c o m e rain or c o m e shine. T h e 16 markets traded were D mark, crude oil,
lumber, copper, gold, dollar index, live cattle, T-bonds, cotton, Japanese yen, natural gas, wheat, Nikkei index, coffee, T-bills, and rough rice.
Without the trend filter, only seven markets had an average profit factor
above one, b u t for none of t h e m can we say with 6 8 % certainty that the true profit factor is above 1. For instance, with an average profit factor of 1 and a standard
deviation of 0.19, we cannot say with 6 8 % certainty that the true profit factor for
rough rice is above 1. For all markets combined, the true profit factor is with 6 8 %
certainty likely to be found somewhere within the interval 0.88 to 1.12.
With the trend filter, however, all markets but one (natural gas) h a d an average profit factor above 1 and for a total of nine we also can say that we can be 6 8 %
sure the true profit factor will be above 1. For all markets combined, the average
profit factor c o m e s out to 1.16, with a standard deviation of 0.13. This m e a n s that
we can be 6 8 % sure that the true profit factor for all market will lie somewhere in
the interval 1.03 to 1.29. (Similar observations that support the findings regarding
the profit factor also can be m a d e for the percentage of profitable trades.) Thus,
with a simple test like this we prove the benefits of always trading with the longt e r m trend, even w h e n we enter randomly and with such a simple exit technique as
always exiting after five days. Imagine what we could do if we were to optimize a
trend-monitoring strategy with a set of well-researched entry and exit techniques.

TREND FILTERS FOR SHORT-TERM SYSTEMS

A basic moving average filter should form the base of any filter testing procedure,
in w h i c h you test which lookback period w o r k s best together with the original
entry technique on a wide variety of markets.

CHAPTER 12

Filtering

TABLE

229

12.1

A simple 200-day moving average trend filter.

Without trend filter

Market

Profit
factor

St
Dev

profitable

With trend filter

St
Dev

Profit
factor

St
Dev

profitable

St
Dev

D-mark

0.98

0.07

52.45

1.90

1.21

0.11

56.08

2.24

Crude oil

0.94

0.16

50.10

2.01

1.15

0.15

56.21

1.83

Lumber

0.98

0.11

50.68

1.79

1.23

0.09

53.56

1.64

Copper

0.97

0.06

50.11

1.59

1.25

0.10

51.43

1.96

Gold

1.04

0.18

50.98

1.82

1.11

0.13

52.23

1.52

Dollar index

1.02

0.09

50.26

2.63

1.18

0.13

52.65

2.23

Live cattle

1.04

0.09

50.43

1.96

1.06

0.11

52.42

1.76

T-bonds

1.04

0.11

50.14

2.21

1.22

0.09

54.81

1.75

Cotton

0.97

0.11

50.09

1.76

1.22

0.11

54.05

1.94

Japanese yen

0.99

0.11

52.24

1.88

1.22

0.08

56.56

1.52

Natural gas

1.03

0.12

49.76

3.46

0.99

0.11

49.26

1.86

Wheat

1.03

0.10

51.15

1.64

1.11

0.11

53.03

1.57

Nikkei index

0.97

0.15

50.60

3.51

1.09

0.11

54.26

1.99

Coffee

1.00

0.13

49.87

2.40

1.06

0.09

51.60

1.65

T-bills

1.01

0.14

51.53

1.95

1.20

0.16

55.32

1.78

Rough rice

1.00

0.19

50.19

2.62

1.26

0.13

55.26

2.25

All markets

1.00

0.12

50.66

2.32

1.16

0.13

53.67

2.65

Alternatively, you could incorporate the on-balance-volume ( O B V ) indicator

that weights each day's price action with its volume, creating a n e w time series that
holds b o t h price and v o l u m e information. In this way, the time series increases/
decreases m o r e in value the m o r e volume there is behind the m o v e of the price, the
interpretation being the s a m e as for the moving average of price. That is, w h e n the
O B V indicator is above its long-term moving average the trend is up and w h e n it
is below, the trend is down. Personally, I think this is one of the best and m o s t versatile indicators there is, because of its ability to in one single and simple calculation, cover everything there is to k n o w and monitor in the field of technical analysis, namely, the psychology of the m a s s e s . Figure 12.1 shows w h a t this can look
like for the S & P 500, w h e n charted together with the O B V indicator and its 2 0 0 day m o v i n g average.
O n e important thing to r e m e m b e r w h e n looking at Figure 12.1 is that the
trend, as you decide to define it, does not necessarily have to coincide with the actual trend of the price chart. For instance, according to this indicator, which also takes
the v o l u m e behind the m o v e into consideration, the trend for the stock market has
been down since July 1998, although we all k n o w that the prices have not fallen
m u c h (if any). In this case, however, the O B V indicator tells us that the volume on

230

PART 4

FIGURE

12.1

The S&P 500 futures contract charted together with a 200-day moving average and the OBV indicator.

High-probability Filters

CHAPTER 12

Filtering

231

the down side has increased, while there has been a diminishing volume on the up
side, m a k i n g it increasingly harder to trade any rallies successfully. Also, there is
nothing keeping you from having several different strategies, based on several different trend filters. It is important to stay consistent with your beliefs and trust w h a t
your research has shown you for that particular system.
If you like to take a m o r e fundamental approach, you instead could calculate
a ratio between the market of your interest and the short-term interest rate, such as
that of the T-bill. T h e reasoning behind this indicator is that w h e n the price for Tbills is increasing faster than the price for " y o u r market," m a n y m a r k e t participants
prefer the safe haven of interest rates instead of taking the c h a n c e of investing in
a market that they d e e m to have a very low profit potential. This could present a
good opportunity to go short, especially if " y o u r m a r k e t " starts to decline as well.
If the opposite holds true, however, and your m a r k e t is increasing faster in price
than the T-bills, m o r e a n d m o r e investors n o w prefer to b u y the market instead of
staying in interest rates. This could present a good opportunity to go long. Again,
the m e t h o d is the s a m e as for the basic moving average filter. With the help of a
long-term moving average we can decide whether or not the m o n e y is flowing into
or out of the market in w h i c h we are interested.
Finally, I use another method that sort of turns the entire reasoning around.
With the three m e t h o d s above, we strive to trade in the direction of the underlying
trend, but because m a n y short-term systems try to enter with limit orders, there
might be plenty of contradicting signalsshort-term entry signals saying " b u y "
and long-term trend filters saying "sell." To work around this, you might formulate
a rule that says, "as long as the trend is not down, I can go long, and as long as the
trend is not up, I can go short." O n e way of doing this is to use a basic highest
high/lowest low breakout indicator, which says that as soon as we m a k e a new high
(low) the long-term trend is up (down) and, as long as the short-term limit order,
sell (buy) signal does not collide with a new high (low), we can take the trade.
In this chapter we test a few versions and different lengths of these trend filters,
together with all our short-term systems, on a portfolio of different markets to see if
there is anything to be gained from adding filters to the systems as they function so far.
For the basic price filter, the O B V filter, and the T-bill ratio filter we test if
it is better to go with the slope or the crossover of their respective moving averages. This counts for a total of six different filter techniques. For each m o v i n g
average filter, 10 different lengths of the moving average will be tested, ranging
from 50 to 250 days, in steps of 20 days. For the breakout filter, the lookback period will vary from 20 to 120 days, in steps of 10 days. E a c h trade is entered randomly, but only in the s a m e direction as the underlying trend, as indicated by the
filter, and each market is traded 10 times for G o d knows h o w m a n y trades and
unique yearly trading sequencescertainly enough to let us c o m e up with some
meaningful conclusions anyway. T h e exit strategy for all m a r k e t s and filters is the
generic exits a n d stops combination we derived in Part 3. T h e 16 markets tested

PART 4

232

High-probability Filters

over the period January 1980 to October 1992 are D mark, crude oil, lumber, c o p per, gold, dollar index, live cattle, T-bonds, cotton, Japanese yen, natural gas,
wheat, Nikkei index, coffee, S & P 500, and r o u g h rice.
In TradeStation, the code for taking a short position with any of these filters,
together with a simple export function that exports the value of the average trade
for each t i m e the system runs through the market, looks something like this:
Inputs: Counter(O), SystemSwitch(O), OBVLen(O), MALen(O), RaLen(O),
BOLen(O);
Vars: TotTr(O), MP(0), FTE(O), Firstlnput(O), TradeStrl("");
If (SystemSwitch =3 and OBV < Average(OBV, OBVLen)) Or
(SystemSwitch =4 and Average(OBV, OBVLen) < Average(OBV, OBVLen)[l]) Or
(SystemSwitch =5 and Close < Average(Close, MALen)) Or
(SystemSwitch =6 and Average(Close, MALen) < Average(Close, MALen)[l]) Or
(SystemSwitch =7 and Ratio < Average(Ratio, RaLen)) Or
(SystemSwitch =8 and Average(Ratio, RaLen) < Average(Ratio, RaLen)[l]) Or
(SystemSwitch =9 and Close < Highest(Close, BOLen)) Then
Sell at Close;
If TotTr > TotTrfl] Then Begin
If MP[1] =1 Then
FTE =(ExitPrice(l) - EntryPrice(l)) / EntryPrice(l);
I f M P [ l ] = - 1 Then
FTE =(EntryPrice(l) - ExitPrice(l)) / EntryPrice(l);
SumFTE = SumFTE + FTE;
End;
If LastBarOnChart Then Begin
If SystemSwitch =3 or SystemSwitch =4 Then
Firstlnput = OBVLen;
If SystemSwitch =5 or SystemSwitch =6 Then
Firstlnput = MALen;
If SystemSwitch =7 or SystemSwitch =8 Then
Firstlnput = RaLen;
If SystemSwitch =9 Then
Firstlnput = BOLen;
FTE = SumFTE / TotalTrades;
TradeStrl =LeftStr(GetSymbolName, 2) + "," + NumToStr(FTE, 4) +
"," + NumToStr(FirstInput, 2) + NewLine;
FileAppend("D:\Temp\LT-Filter.csv", TradeStrl);
End;

CHAPTER 12

Filtering

233

With this code you can set up a w o r k g r o u p containing all the markets you
would like to test, and run the system optimizer on all markets at once. While the
c o m p u t e r is chewing away, go do something else for a change, because this can
take a while. By the way, be careful w h e n you are exporting data from
TradeStation while using the optimizer. For some reason, w h e n the optimization is
done, the p r o g r a m once again runs through and exports the results for the "best"
input variable combination, not o n c e , but twice, and sometimes even three times!
Weed out these extra runs from your spreadsheets manually before you do any further analysis.
W h e n it c o m e s to volatility, in any market, higher volatility also m e a n s higher risk because of the greater likelihood that the market can move against you.
Obviously then, from a safety point of view, the lower the volatility the better. T h e
problem is that especially for shorter-term trading, high volatility also m e a n s a
higher profit potential. T h e trick is to find those short-term situations w h e n the
volatility is high enough to m a k e it worthwhile to trade but not so high that only a
fool would enter the market. But because there is m o r e than one way to measure
volatility, how do you k n o w w h i c h m e t h o d will give the most accurate reading?
T h e truth is, nobody knows, and your guess is as g o o d as m i n e . Probably the
m o s t c o m m o n way is to calculate the standard deviation of the closes, w h i c h is
basically what you do if you are an options trader and, therefore, work with different option pricing formulas, such as Black and Scholes or Black-76. However,
over the last couple of years there s e e m s to be at least one m e t h o d that has
emerged as the m o s t popular a m o n g technical analysts. This is the true range or
average true range method that states that w h e n the true range of today's price
action is below the average true range for a specified lookback period (sometimes
multiplied by a chosen factor) the most recent volatility is lower than expected, and
hence, the market is about to calm down a n d b e c o m e less risky to trade, no matter the direction of the potential trade. T h e true range is defined as the distance
between today's high (low), and today's low (high) or yesterday's close, whichever p r o d u c e s the largest distance.
To test if there is any validity to this c o m m o n l y used technique, I set out to
test this technique over several different lookback periods, ranging from 10 to 40
days, in steps of three days, using the same markets a n d set up as for the long-term
filters. T h e rule w a s not to take a trade in any direction if the most recent volatility (true range) was higher than the average volatility (average true range) over the
lookback period.
T h e m a i n disadvantage, however, with both the standard deviation method
and the average true range method is that none takes the direction of the volatility
into consideration. For instance, in a strongly up-trending market that only climbs
higher and higher, it is obvious that most, if not all, of the volatility is to the up
side. This, in turn, is of course very good if you are long in the market but not so
good if you are short. Therefore, it also could be a good idea to try to use some

234

PART 4

High-probability Filters

sort of volatility m e a s u r e that can tell the difference between up volatility a n d

down volatility, or better yet, w h e n the market is about to shift from a higher down
volatility to a higher up volatility, and vice versa.
Fortunately, there are a few m e t h o d s that readily lend themselves to this purpose. If you have studied the math behind m a n y of the m o s t c o m m o n l y used oscillators that you can find in m o s t technical analysis p r o g r a m s , you will realize that
two of these indicators are your regular everyday technical analysis tools, the RSI
indicator and the ADX indicator.
In the R S I indicator, the reasoning is simple enough. W h e n the indicator
reading is above 50, the upward volatility is stronger than the downward volatility, and vice versa. For the A D X indicator, however, it is slightly m o r e complicated, b e c a u s e a high A D X reading could m e a n either a strong upside volatility or a
strong downside volatility. Therefore, instead of using the A D X indicator, it is better to use two of the c o m p o n e n t s u p o n which it is built, the DMI-plus and DMIminus lines. With the D M I - p l u s line above the D M I - m i n u s line, the upward
volatility is greater than the downward volatility, a n d vice versa. For the R S I indicator you also can experiment with different trigger levels and, for example, only
allow long trades if the R S I reading is above 60 a n d short trades if the RSI reading is below 4 0 , leaving the area in between as a neutral zone. You also could add
to that the usual overbought/oversold technique and not go long (short) if you consider the R S I to be too high (low).
To test if these two indicators could serve as filters for our short-term trading m o d e l s , I set out to test t h e m with lookback periods ranging from 10 to 40 days
in steps of three days. T h e rule was only to take a trade in a specific direction if
the volatility reading suggested that the volatility w a s stronger in that direction as
well. For this I used the s a m e markets a n d general setup as for the long-term filters a n d for the average true range filters. T h e TradeStation code for all the volatility filters basically looked the s a m e as for the trend filters.
T h e results for all filters were then s u m m a r i z e d in separate tables, as shown
in Tables 12.2 and 12.3, w h i c h show the results for the basic moving average
crossover filter a n d the breakout filter, respectively. For instance, for the 50-day
MA cross filter the average profit per trade w a s 0.12%, from trading all 16 markets 10 times, using random entries and the Black J a c k exits were derived in Part
3. T h e m o s t robust results seem to be around the 110- to 170-day area, with the
absolute best risk-reward alternative being the 110-day average. Because the true
best average might m o v e around, I decided to go with the 130-day alternative. This
is how you m u s t reason in any type of optimization procedure.
If the results look p r o m i s i n g for the regular moving average filter, the opposite holds true for the breakout filter. For instance, with a 60-day lookback period
and the exact s a m e set up as for the moving average filter, the breakout filter only
m a n a g e d to p r o d u c e an average profit per trade of 0 . 0 5 % , which also s e e m s to be
the best alternative after we have m e a s u r e d in all the other variables, such as the

CHAPTER 12

Filtering

235

TABLE

12.2

MA cross filter, January 1980-October 1992.

Length

Average

St Dev

Ratio

Kurtosis

Skew

50

0.1201

0.2071

0.5799

6.2312

2.0777

70

0.1186

0.2116

0.5606

6.1378

1.3598

90

0.0821

0.1400

0.5863

4.5955

0.2875

110

0.1252

0.1766

0.7088

8.2450

1.3125

130

0.1131

0.1722

0.6568

4.6584

0.8590

150

0.1058

0.1718

0.6157

4.5265

0.1703

170

0.1139

0.1715

0.6641

10.4910

-0.9180

190

0.0818

0.1578

0.5181

4.6071

-0.5006

210

0.1001

0.1525

0.6563

18.6751

-2.9483

230

0.1023

0.1538

0.6654

8.3662

0.6040

250

0.1202

0.2020

0.5950

13.4122

0.2106

standard deviation, the kurtosis, a n d the skew. As we will see, however, j u s t

b e c a u s e one filter s e e m s to be inferior to another at this stage of the testing p r o cedure it does n o t m e a n that it will not work in the end.
For all different lookback periods for all different filters I p i c k e d out one
lookback period from e a c h filter, which I set out to test together with the complete
system. T h e different filter/lookback periods I decided to test further can be seen
in Table 12.4. For instance, the D M I indicator proved m o s t useful with a lookback
period of 34 days, w h i c h generated an average profit of a l m o s t 0 . 1 3 % p e r trade.

TABLE

12.3

Breakout filter, January 1980-October 1992.

Length

Average

St Dev

Ratio

Kurtosis

Skew

20

0.0460

0.1478

0.3112

5.3691

0.3835

30

0.0408

0.1385

0.2946

3.7634

1.2145

40

0.0309

0.1267

0.2437

6.0328

-0.7399

50

0.0456

0.1475

0.3088

4.5719

0.7597

60

0.0511

0.1461

0.3499

7.2699

1.7590

70

0.0401

0.1578

0.2543

5.0591

-0.1657

80

0.0317

0.1196

0.2649

7.2570

1.5680

90

0.0318

0.1486

0.2141

5.0252

-1.1230

100

0.0293

0.1438

0.2034

10.8615

-1.7311

110

0.0474

0.1597

0.2966

5.2053

-0.4098

120

0.0419

0.1318

0.3182

4.0369

0.5033

PART 4

236

TABLE

High-probability Filters

12.4

Chosen input parameters for different filters.

Filter

Length

Average

St Dev

Ratio

Kurtosis

Skew

DMI

34

0.1287

0.1860

0.6920

4.9766

1.8047

True range

37

0.0467

0.1807

0.2583

6.1131

-0.1240

RSI

28

0.1128

0.2165

0.5210

4.0422

0.1418

OBV-cross

70

0.1019

0.1686

0.6045

3.6123

1.0086

OBV-slope

70

0.1026

0.1767

0.5805

5.9592

1.4997

MA-cross

130

0.1131

0.1722

0.6568

4.6584

0.8590

MA-slope

70

0.1261

0.1679

0.7510

6.9234

1.7958

Ratio cross

70

0.1117

0.1984

0.5629

4.3042

1.6645

Ratio slope

90

0.0776

0.2137

0.3633

9.8603

-0.9665

Breakout

60

0.0511

0.1461

0.3499

7.2699

1.7590

T h o s e that did not work at all w e r e the true range method, the T-bill ratio slope
indicator, and the breakout method.
A few things worth mentioning from this test can c o m e in h a n d y for your
own testing: the O B V and ratio indicators probably would have d o n e better had
they been tested with shorter lookback periods. T h e not-so-tantalizing results for
the breakout filter probably are because of the b a c k w a r d s reasoning you m u s t use
for a filter like this, in combination w i t h the r a n d o m entries, which, as things are
right now, allow for a lot of trades with the market in no trend at all, or in consolidation m o d e . T h e s a m e holds true for the true range method, when c o m p a r e d to
the results for the D M I and R S I indicators. Finally, it is interesting to note that the
m o v i n g average crossover m e t h o d does not s e e m to work quite as well as the slope
m e t h o d of half the length. This confirms our findings from Part 2, w h e r e we put
together our weekly directional slope system.

THE GOLD DIGGER SYSTEM

W h e n e a c h filter w a s tested o n t h e S & P 5 0 0 t o g e t h e r w i t h t h e G o l d D i g g e r syst e m , over the two t i m e p e r i o d s J a n u a r y 1985 to D e c e m b e r 1994 a n d January
1995 to O c t o b e r 1999, t h e ratio cross filter p r o d u c e d the b e s t results. T h e s e can
be seen in Tables 12.5 a n d 12.6 a n d directly c o m p a r e d to Tables 11.16 and
11.17.
Perhaps it worked a little t o o well: high profit factors and low drawdowns
count for very little if the trades are too few to m a k e it possible to c o m e to any
meaningful conclusions. T h e reason for the very few trades is to be found in the
original composition of the entry technique, which needs to be modified to make
it worthwhile trading it together with a filter.

CHAPTER 12

TABLE

Filtering

237

12.5

Gold Digger (original)/ratio cross filter, January 1985-December 1994.

v
Total trades

30

16

55.33%

14

46.67%

Profit factor

1.32

Lrg winner

2.80%

9,450

Lrg loser

-2.07%

-6,986

Avg winner

1.37%

4,613

Avg loser

-1.19%

-4,002

C u m profit

5.06%

17,078

Drawdown

-4.16%

-14,040

Avg profit

0.18%

593

St Dev

1.51%

5,093

Winners

Losers

This is largely b e c a u s e the original entry trigger d e m a n d s two w e e k s in a row

in the s a m e direction, which c o m e s awfully close to a complete change of trend,
before we are allowed to enter with a limit order in the opposite direction. To m a k e
it a little easier to enter a trade we can loosen the original entry somewhat and only
look at the latest week's market action. If we do that, it turns out that the breakout
filter produces the best results for the period covering January 1985 to D e c e m b e r
1994, which is shown in Table 12.7.
After we modified the original entry technique, the breakout filter was the
m o s t profitable for the period January 1985 to D e c e m b e r 1994, with a profit factor of 1.30 and an average profit p e r trade of 0 . 1 7 % . T h e d r a w d o w n is a tolerable
7.94%. For the period January 1995 to October 1999, the profit factor also came
out to 1.30, while the average profit increased somewhat to 0 . 2 0 % p e r trade. This
is shown in Table 12.8. However, the increase for the average profit also was
a c c o m p a n i e d by a slightly higher increase of the standard deviation, which m e a n t
that the risk-adjusted return also decreased somewhat. T h e drawdown also
increased to approximately 12%. If you c o m p a r e this version of the Gold D i g g e r
with its slightly loosened entry trigger and the attached filterwith the version
depicted by Tables 11.16 and 11.17, you see that this latest version beats its predecessor in almost every single category, not the least of w h i c h is the drawdown,
which is an indication of h o w well the system will let you sleep at night.
N o w all that needs to be d o n e is to m a k e sure that this new filter/entry c o m bination indeed betters the results as c o m p a r e d to the new entry technique alone.
If it does not, there is little use adding the filter. T h e results for trading the new
entry technique only, without the filter, can be seen in Tables 12.9 and 12.10, and
as you can see these results are not as good as those with the filter, especially when

TABLE

12.6

Gold Digger (original)/ratio cross filter, January 1995-October 1999.

Total trades

Profit factor

5.91

Winners

77.78%

22.22%

Lrg winner

2.80%

9,450

Lrg loser

-1.10%

-3,713

1.86%

6,268

Avg loser

-1.10%

-3,713

11.21%

37,834

Drawdown

-1.10%

-3,713

Avg profit

1.20%

4,050

Avg winner

St Dev

1.65%

5,584

C u m profit

Losers

238

PART 4

TABLE

High-probability Filters

12.7

Gold Digger (modlfied)/break filter, January 1985-December 1994.

Total trades

367

Winners

Profit factor

1.30

Lrg winner
Avg winner

Avg profit

0.17%

569

St Dev

1.52%

5,135

TABLE

Cum profit

193

52.59%

174

47.41%

5.03%

16,976

Lrg loser

-4.48%

-15,120

1.38%

4,642

Avg loser

-1.17%

-3,949

77.58%

261,833

Drawdown

-7.94%

-26,798

101

53.72%

Losers

87

46.28%

5.14%

17,348

Lrg loser

-8.30%

-28,013

1.62%

5,464

Avg loser

-1.44%

-4,862

41.68%

140,670

Drawdown - 1 2 . 1 1 %

-40,871

Losers

12.8

Gold Digger (modified)/break filter, January 1995-October 1999.

Total trades

188

Profit factor

1.30

Lrg winner
Avg winner

Avg profit

0.20%

685

St Dev

1.88%

6,329

Winners

Cum profit

it c o m e s to the drawdown and percentage of profitable trades. T h e profit factors

and average trades, too, are not as g o o d without the filter. T h e lower cumulative
profit, with the filter attached, does not matter at this point, because this will be
dealt with later by attaching the proper m o n e y m a n a g e m e n t during the "optisizi n g " process.
T h u s , by following the system building sequence outlined in Part 3, we have
not only m a n a g e d to consistently improve the original system from Part 2, which
really was nothing m o r e than a very simple entry technique, but we also have m a n aged to do so without stacking indicators on top of each other and curve fitting the
final version to a specific market. In fact, even though this is a system specifically designed for the S & P 500, m o s t of the time we haven't even used the S & P 500
index in our testing and w h e n we have, we have traded it randomly, summarizing
the results together with 15 other markets. F u r t h e r m o r e , as a final step we even
took away some of the original c o m p o n e n t s , ending up with a m o d e l that trades far
m o r e often than its predecessor.

TABLE

12.9

Gold Digger (modified)/no filter, January 1985-December 1994.

Total trades

454

Winners

Profit factor

1.26

Lrg winner
Avg winner

Avg profit

0.15%

513

St Dev

1.68%

5,654

Cum profit

242

53.30%

7.62%

25,718

Losers
Lrg loser

1.36%

4,595

Avg loser

86.55%

292,106

212

46.70%

-12.61%

-42,559

-1.23%

-4,147

Drawdown - 1 8 . 2 9 %

-61,729

CHAPTER 12

239

Filtering

TABLE

12.10

Gold Digger (modified)/no filter, January 1995-October 1999.

Total trades

236

Winners

Profit factor

1.23

Lrg winner
Avg winner

Avg profit

0.15%

520

St Dev

1.81%

6,095

C u m profit

120

50.85%

5.14%

17,348

Losers
Lrg loser

116

49.15%

-8.30%

-28,013

1.60%

5,401

Avg loser

-1.34%

-4,529

38.37%

129,499

Drawdown

-13.48%

-45,495

THE MEANDER SYSTEM

For the Meander, it turned out to be very difficult to find a filter that could p r o duce consistent results. In fact, the best filter w a s the moving average slope filter,
which p r o d u c e d a very high profit factor for the latter of our two testing periods,
but still failed miserably during the period January 1985 to D e c e m b e r 1994. Tables
12.11 and 12.12 show the results, which are comparable to Tables 11.18 and 11.19.
T h e reason for this can probably be found in the composition of the original
entry technique, which is already taking the trend into consideration by looking at
the last five weekly bars to calculate the entry levels. This w a s very difficult to
work around by, for instance, altering the entry level by changing the n u m b e r of
standard deviations away from the m e a n needed to signal an entry. N o r did it help
to switch away from the Black Jack exits, back to the original exits, also developed
in Part 3. Therefore, we will keep the M e a n d e r system, together with the Black
Jack exits, as it is.
At this point, it also could be worthwhile pointing out the importance of trying to c o m e up with a logical reasoning and a way to understand why a specific

TABLE

12.11

The Meander/MA slope filter, January 1985-December 1994.

Total trades

60

Profit factor

0.89

Lrg winner

--0.07%

-250

Avg winner

1.68%

5,685

Cum profit

Avg profit
St Dev

TABLE

Winners

32

55.33%

2.80%

9,450

Losers
Lrg loser

1.09%

3,672

Avg loser

-4.13%

-13,939

Drawdown

28

46.67%

-8.05%

-27,169

-1.40%

-4,732

-13.19%

-44,516

12.12

The Meander/NA slope filter, January 1995-October 1999.

Total trades

29

Profit factor

3.36

21

72.41%

27.59%

Lrg winner

4.21%

14,209

Lrg loser

-1.64%

-5,535

1.52%

5,127

Avg loser

-1.19%

-3,999

24.56%

82,890

Drawdown

-3.41%

-11,509

Winners

Avg profit

0.77%

2,609

Avg winner

St Dev

1.63%

5,498

C u m profit

Losers

PART 4

240

High-probability Filters

filter/system combination does not work. R e m e m b e r that there is no such thing as

a failed experiment, as long as we can learn something and c o m e to s o m e conclusion about how or what to do to improve results in future experiments.

THE BLACK JACK SYSTEM

W h e n we built the Black Jack entries in Part 2, we h a d already added a 200-day
moving average filter. With this filter, the system worked very well during the latter of our two testing periods but not at all during the former. To see if any of our
other filter techniques would work any better and be m o r e consistent, I substituted the original filter for each o n e derived in this section. As a result, the system
started to produce fairly consistent results together with the O B V slope filter.
T h e results for the time period January 1985 to D e c e m b e r 1999 can be seen
in Table 12.13 and directly c o m p a r e d to those in Table 11.21. As you can see, we
have taken a negative average profit p e r trade and turned it into a positive one of
0 . 1 9 % . Other positive signs are a low standard deviation and a tolerable drawdown. T h e percentage of w i n n i n g trades also is very high.
For the period January 1995 to D e c e m b e r 1999, the results are not quite as
good as with the original system, as can be seen if you c o m p a r e Tables 12.14 and
11.20. However, in the larger scheme of things, this matters very little, because the
system now is m u c h m o r e robust a n d reliable, and h e n c e m u c h m o r e likely to hold
up in the future as well.

TABLE

12.13

Black Jack (modified)/OBV slope filter, January 1985-December 1994.

Total trades

202

Winners

117

57.92%

Losers

85

42.08%
-10,429

1.38

Lrg winner

4.12%

13,905

Lrg loser

-3.09%

Avg profit

0.19%

641

Avg winner

1.20%

4,038

Avg loser

-1.20%

-4,036

St Dev

1.43%

4,838

43.35%

146,981

-10.88%

-36,720

Profit factor

TABLE

C u m profit

Drawdown

12.14

Black Jack (modified)/OBV slope filter, January 1995-October 1999.

Total trades

92

60

65.22%

32

34.78%

Profit factor

2.00

Lrg winner

4.04%

13,635

Lrg loser

-2.90%

-9.788

1.41%

4,775

Avg loser

-1.33%

-4,478

50.02%

168,818

Drawdown

-5.03%

-16,976

Winners

Avg profit

0.46%

1,556

Avg winner

St Dev

1.63%

5,485

Cum profit

Losers

CHAPTER

13

Long-term Volatility Filters

For long-term systems, we m u s t think a little differently. For short-term systems,

the p u r p o s e of the filter w a s to identify the type of trend that holds as m a n y and as
favorable short-term moves as possible. For the longer-term system the trend is, p e r
definition, already defined. Instead, we m u s t find the most opportune, short-term
entry point as early as possible within this trend. For one thing, this m e a n s that we
no longer can trade the system randomly. In theory, the best way to achieve this is
to look for either low-volatility situations in general, or situations w h e r e the direction of the volatility coincides with the anticipated m o v e of the market. H e n c e , the
p a r a d o x is that for the long-term system filter we m u s t work with shorter-term data,
a n d vice versa.
A g o o d indicator to use for the former m e t h o d is the average true range indicator, as we did for the short-term systems. This time, however, we will do it a little differently a n d change the range multiplier as well as the lookback period. A
two-variable test also m e a n s that we once again can m a k e use of the surface chart
m e t h o d that we developed for the directional slope system in Part 2. For the lookback period we varied the length from 5 to 25 days, in steps of 2 days. T h e range
multiplier we varied from 0.5 to 2.5, in steps of 0.2. T h e added trading rule w a s not
to take the trade if the true r a n g e for the day of the breakout surpassed the average
true range t i m e s its multiplier.
For long-term systems, with the direction of the trend already defined, we also
can use the A D X indicator, which could not be used in a short-term system, because,
as a standalone indicator, it says nothing about the direction of the volatility. But if
we use it together with any other long-term indicators, we can assume that a high
A D X reading m e a n s a high volatility in the direction of the trend, as indicated by the
241

242

PART 4

High-probability Filters

second indicator, which for our purposes is also the original entry strategyor this
is the theory m a n y system vendors seem to adhere to, because as the A D X indicator is their first choice when it c o m e s to measuring the strength of the trend. The
A D X test also was d o n e as a two-variable test, w h e r e we tested different lengths for
the lookback period together with several different trigger levels. For the lookback
period we varied the length from 10 to 20 days, in steps of 1 day. T h e trigger level
we varied from 15 to 35, in steps of 2. T h e added trading rule was not to take the
trade if the A D X reading for the day if the breakout was below the required trigger
level. Both system/filter combinations also m a k e use of the stops a n d exits for each
respective system we developed in Part 3.
A n o t h e r important difference between the long-term a n d short-term testing
procedures is that for the long-term systems we cannot look at the final profit for
the trade, because this would w e e d out far too m a n y trades and m a k e the results
less meaningful. To do that would also contradict the reason we added the filter to
the system in the first place, which was to find the m o s t opportune m o m e n t to
enter in a trend that w a s already defined. Therefore, the best way is to try to minimize the STD, described in detail in Part 3, in relation to the final profit of the
trade. However, j u s t e x a m i n e the S T D w o u l d n ' t m a k e m u c h sense either, because
we still would like the trade to go on for as long as possible within the confines of
the original system. In essence, this m e a n s that we m u s t examine the entry efficiency of the trade.
In Part 3 we concluded that, for a long-term system, it is reasonable to
a s s u m e that the S T D will be the s a m e as the m a x i m u m adverse excursion (MAE)
and therefore can be derived the s a m e way, although the interpretations and treatm e n t s of the two are not the s a m e . T h e M A E should primarily be dealt with by
adjusting the stops, w h i c h we did in Part 3, w h e r e we also took care of all the other
types of drawdowns and excursions. T h e STD, on the other hand, should primarily be dealt with by filtering out the m o s t opportune trading situations, which is
what we are about to do now.
A l t h o u g h we might be using the s a m e techniques to derive the necessary data
for the S T D as for the M A E , it is important to understand that the two are not the
s a m e . N o r does either o n e of t h e m equal R I N A S y s t e m s ' entry efficiency. The difference is that the S T D should primarily be adjusted with the entry technique,
although the M A E should primarily be adjusted with a specific exit technique,
such as a stop loss. Together, the M A E and S T D m a k e up the entry efficiency.
T h e m a r k e t s used for testing were D mark, crude oil, lumber, copper, gold,
dollar index, live cattle, T-bonds, cotton, Japanese yen, natural gas, wheat, Nikkei
index, coffee, T-bills, and r o u g h rice. T h e time period covered was January 1980,
to October 1992.
Figure 13.1 shows that the D y n a m i c Breakout System ( D B S system) produced the highest entry efficiency in combination with the average true range filter with a seven-day lookback period and a multiplier of 0.5. B e c a u s e this is not a

CHAPTER 13

Long-term Volatility Filters

FIGURE

243

13.1

The entry efficiency for the DBS system in relation to the true range multiplier and its lookback period.

particularly robust solution, it cannot be our preferred choice. Instead, we should

strive to place ourselves s o m e w h e r e in the area around a lookback period of nine
days and a multiplier around 1.5. In this way, we avoid any of the areas of lower
efficiency around, for instance, the 1.1 multiplier, if it j u s t so h a p p e n s that the true
best combination might shift around a little in the future.
Support for this also can be found in Figure 13.2, which depicts the standard
deviation of the efficiency. In Figure 13.2 it seems that, as far as the standard deviation goes, we should be able to go with a multiplier of 1.5, without getting too close
to the higher standard deviation areas. Which means we will settle for a 9/1.5 lookback/multiplier combination.
Figure 13.3 shows that the D B S system produced the highest entry efficiency
in combination with the A D X filter with a 20-day lookback period and a trigger
level of 3 3 . Because this too is not a particularly robust solution, however, it cannot
be our preferred choice. Instead, we should strive to place ourselves somewhere in
the area around a lookback period of 12 days and a trigger level at or above 2 5 . In

244

PART 4

FIGURE

High-probability Filters

13.2

The standard deviation of the entry efficiency for the DBS system and the true range filter.

this way, we avoid any of the areas of lower efficiency around, for instance, a trigger level of 21 and a lookback period of 11 days, if it happens that the true best
combination might shift slightly in the future
This is confirmed by Figure 13.4, w h i c h depicts the standard deviation of the
efficiency. In Figure 13.4, it seems that, as far as the standard deviation goes, we
should go with a trigger level of 25 and a lookback period of 11 days, without
increasing the standard deviation too m u c h ; because this seems to be possible
without running the risk of lowering the efficiency, this is our preferred choice.
Table 13.1 summarizes all the preferred choices for our three long-term syst e m s and filters. T h e time period covered is from January 1980, for all systems, to
October 1999 for the D B S and directional slope systems, and to October 1992 for
the standard deviation breakout system. T h e trade-by-trade export function from
Part 1 is used to export the results into an Excel spreadsheet for further evaluation
and comparison with earlier, pre-filter results.

CHAPTER 13

245

Long-term Volatility FiIters

FIGURE

13.3

The entry efficiency for the DBS system in relation to the ADX trigger level and its lookback period.

TABLE

13.1

Preferred system/filter combinations.

ADX filter
System

True range filter

LB period

Trigger

LB period

Multiplier

Dynamic Breakout

11

25

1.5

Standard Deviation Breakout

14

21

17

0.9

Directional Slope

11

25

11

1.7

PART 4

246

F I G U R E

High-probability Filters

1 3 . 4

The standard deviation of the entry efficiency for the DBS system and the ADX filter.

THE DBS SYSTEM

Tables 13.2 a n d 13.3 show the result for the DBS/true range combination, as outlined in Table 13.1. T h e s e tables can be directly c o m p a r e d to Tables 10.9 and
10.10. Table 13.2 shows that of all 16 markets, only five are worth trading (if we
use our m l e of t h u m b that says that the gross average profit should be at least three
times the expected slippage and commission, which we estimated at $75). This is
no different from Table 10.9, so the filter did not help any in this regard, even
though it did increase the value of the average trade in nine markets.
W h e n it c o m e s to t h e profit factor, the filter left us with a profit factor
above 1 in twelve m a r k e t s . T h i s , t o o , is the s a m e as w i t h o u t the filter, but
b e c a u s e the filter only m a n a g e d to increase the profit factor in seven markets
and, h e n c e , d e c r e a s e it in n i n e , it is o b v i o u s that this s y s t e m does not benefit
from this filter. It d o e s not m a t t e r that the filter m a n a g e d to lower the drawdown
for nine m a r k e t s .

CHAPTER 13

247

Long-term Volatility Filters

TABLE

13.2

Performance measurements for all markets for the DBS/true range combination.
After filter (modified system with exits and filter)
Market

P. factor

Avg. trade

2 St. dev.

Drawdown

Corn

1.31

69.73

1,425.07

-3,556.69

S&P 500

0.94

-289.83

27,234.52

-144,606.61

Orange juice

1.35

104.50

2,407.67

-9,192.10

Live cattle

0.98

-6.81

2,293.99

-12,059.97

Lumber

1.20

114.27

3,630.98

-11,792.83

Coffee

1.19

153.91

5,558.24

-19,517.05

Japanese yen

2.13

1,207.82

9,862.45

-13,700.98

Copper

0.87

-52.83

2,224.48

-11,294.01

Gold

1.23

107.37

2,799.42

-9,348.00

Eurodollar

1.77

263.05

3,001.01

-7,073.90

Dollar index

2.23

776.99

6,054.09

-7,816.56

Cotton

1.32

165.24

3,923.99

-10,654.24

CRB index

0.59

-505.79

4,982.93

-54,603.45

Crude oil

2.22

555.04

4,772.65

-7,144.46

Canada dollar

1.20

62.68

2,055.76

-7,899.86

T-bonds

1.18

258.24

8,695.56

-24,953.48

Unfortunately, the depressing results continue when we look at the D B S / A D X

combination, as illustrated by Tables 13.4 and 13.5. The only thing positive thai can
be said about this combination is that it did m a n a g e to make one m o r e market (gold)
worth trading, but other than that it only managed to increase the value of the average trade and the profit factor for five markets (lowering them for 11) and decrease
the drawdown and standard deviation for five and eight markets, respectively.
T h e s e are not very impressive n u m b e r s at all and certainly not impressive
enough to warrant the extra set of rules that we would have to add to the original
system. T h e conclusion that can be m a d e from these tests is, therefore, that this
version of the D B S system, w h e n it c o m e s down to measuring p u r e profits or
robustness of the results, does not benefit from being traded together with either
of the two filter techniques discussed.

THE SDB SYSTEM

For the Standard Deviation Breakout ( S D B ) system c o m b i n e d with the true range
filter, the results are, however, m o r e encouraging. Of all 16 markets tested, a total
of 12 and 11 resulted in higher profit factors and average profits per trade. Tables
13.6 and 13.7 show the results for this system/filter combination, which also are

PART 4

248

TABLE

High-probability Filters

13.3

Summarized differences, before/after filter for the DBS/true range combination.

Differences
Better

Market

P. factor

Avg. trade

2 St. dev.

Drawdown

Corn

-5.02%

-15.74%

12.99%

-67.29%

-0.46%
1.44%

11.03%

S&P 500

-13.86%

-10.02%

-26.22%

-1.44%

5.85%

-3.47%

-190.30%

0.20%

12.31%

0
3

Orange juice
Live cattle
Lumber

6.61%

53.80%

1.39%

-2.02%

Coffee

17.98%

1344.93%

19.90%

-15.69%

Japanese yen

-3.59%

-2.29%

1.71%

0.00%

0
2

-6.72%

106.33%

-1.11%

11.48%

-15.00%

-44.85%

-5.57%

9.25%

Eurodollar

4.77%

5.67%

-0.58%

-17.53%

Dollar index

6.45%

10.65%

5.60%

-0.58%

Cotton

5.82%

26.01%

2.09%

-1.73%

CRB index

-1.00%

0.56%

-2.75%

-3.90%

Crude oil

-0.47%

1.90%

2.37%

-4.80%

6.67%

56.39%

6.07%

9.69%

-4.43%
7

-22.43%

0.09%

-0.53%

Copper
Gold

Canada dollar
T-bonds
Better

TABLE

13.4

Performance measurements for all markets for the DBS/ADX combination.

After filter (modified system with exits and filter)
Market

P. factor

Avg. trade

2 St. dev.

Drawdown

Corn

1.26

58.29

1,396.50

-3,355.61

S&P 500

0.86

-732.10

27,240.46

-170,608.33

Orange juice

1.42

127.51

2,562.59

-10,495.37

Live cattle

0.99

-4.31

2,293.65

-11,805.89

Lumber

1.12

68.83

3,572.99

-13,732.89

Coffee

1.10

79.02

4,792.56

-19,299.50

Japanese yen

2.14

1,205.34

9,841.21

-16,746.74

Copper

0.87

-54.77

2,166.07

-11,078.56

Gold

1.58

237.19

2,916.27

-6,728.37

Eurodollar

1.68

238.06

2,992.40

-8,849.74

Dollar index

1.71

525.55

5,854.09

-10,174.83

Cotton

1.25

131.08

3,910.42

-11,709.41

CRB index

0.50

4,884.03

-55,666.35

Crude oil

2.36

-647.78
600.54

4,868.78

-7,508.33

Canada dollar

1.11

37.30

1,929.15

-6,386.36

T-bonds

1.19

264.28

8,514.21

-25,036.36

CHAPTER 13

249

Long-term Volatility Filters

TABLE

13.5

Summarized differences, before/after filter for the DBS/ADX combination.

Differences
P. factor

Avg. trade

2 St. dev.

Corn

-8.80%

-29.57%

-2.46%

4.75%

3.28%

-17.38%

1.46%

1.63%

Orange juice

-5.06%

-9.97%

4.90%

20.85%

Live cattle

-2.88%

-157.13%

0.19%

9.94%

Lumber

-0.72%

-7.36%

-0.23%

14.10%

Coffee

8.14%

641.83%

3.38%

-16.63%

S&P 500

Drawdown

Better

Market

Japanese yen

-3.04%

-2.49%

1.49%

22.23%

Copper

-7.71%

113.89%

-3.71%

9.35%

9.14%

21.81%

-1.63%

-21.37%

-0.97%

-4.37%

-0.87%

3.17%

-18,38%

-25.16%

2.11%

29.41%

0.07%

-0.04%

1.74%

8.01%

-16.57%

28.79%

-4.68%

-2.03%

Gold
Eurodollar
Dollar index
Cotton
CRB index

5.58%

10.25%

4.43%

0.05%

Canada dollar

-0.74%

-6.92%

-0.47%

-11.33%

T-bonds

-4.06%

-20.62%

-2.00%

-0.20%

Crude oil

Better

TABLE

13.6

Performance measurements for all markets for the SDB/true range combination.
After filter (modified system with exits and filter)
Market

P. factor

Avg. trade

2 St. dev.

Drawdown

Crude oil

4.22

808.63

4,276.38

-2,192.25

T-bonds

1.34

517.60

11,124.55

-19,518.81

T-bills

2.54

865.65

6,978.05

-5,794.19

Rough rice

3.57

1,266.12

9,152.78

-3,587.94

Nikkei index

1.21

770.16

22,715.90

-9,832.09

8,817.78

21,619.06

0.00

3,353.59

-9,101.18

Natural gas
Live cattle

1.28

160.80

Lumber

1.50

376.28

5,895.38

Coffee

1.73

274.34

4,278.94

-4,591.37
-4,998.12

Japanese yen

1.64

966.81

12,290.12

-13,073.75

Copper

2.29

765.87

7,659.45

-5,704.59

Gold

1.92

693.90

5,423.84

-5,592.07

Dollar index

2.89

1,418.88

7,538.99

D mark (Euro)

2.93

1,693.64

9,198.07

-4,611.68
-11,567.57

Cotton

4.03

948.52

4,829.12

-2,639.46

Wheat

1.42

202.91

3,382.67

-6,670.94

250

PART 4

TABLE

High-probability Filters

13.7

Summarized differences, before/after filter for the SDB/true range combination.

Differences
Market

P. factor

Crude oil

22.73%

8.25%

3.77%

17.41%

18.00%

Rough rice
Nikkei index

T-bonds
T-bills

Drawdown

Better

2.23%

21.56%

1.82%

-1.85%

27.93%

8.17%

2.98%

182.47%

906.66%

200.80%

-30.69%

-70.47%

-87.41%

-5.39%

100.47%

466.14%

235.64%

-100.00%

3
2

Natural gas
Live cattle

Avg. trade

2 St. dev.

17.43%

232.39%

11.10%

5.69%

Lumber

5.31%

40.72%

23.63%

-17.61%

Coffee

38.21%

212.89%

74.05%

20.22%

Japanese yen

-31.83%

-43.52%

-4.60%

23.62%

Copper

99.84%

697.70%

89.14%

-17.55%

Gold

50.94%

179.49%

0.87%

-34.66%

Dollar index

18.80%

-2.01%

-5.99%

5.83%

D mark (Euro)

-1.15%

-7.41%

-2.89%

-4.20%

Cotton

48.15%

41.47%

6.60%

-32.16%

Wheat

-23.62%

-33.99%'

8.45%

39.44%

Better

12

11

comparable to Tables 10.19 and 10.20. A n o t h e r positive factor is that the number
of tradable markets, as measured by the value of the average trade, also has
increased from 12 to 14.
Less optimistic is the fact that the risk, as measured by the standard deviation, has increased for a total of 12 markets. But as long as the risk does not
increase at a faster rate than the value of the average trade, the risk-adjusted return
still increases, in 11 out of 12 instances. T h e only market that really does not like
this filter strategy is the wheat market, w h i c h does not m a n a g e to improve by one
single measure.
Given the good performance of the S D B system together with the true range
filter a n d the popularity of the A D X filter a m o n g the elite system builders, one
would think that the S D B s y s t e m / A D X filter combination is b o u n d to be a success. W r o n g ! As Tables 13.8 a n d 13.9 show, the A D X filter did not work this time
either, improving the profit factor for only seven and the value of the average trade
for only eight, out of 16 markets. C o m p a r e d to Tables 10.19 and 10.20, this system/filter combination also m a d e a total of four markets untradable, as measured
by the profit factor, by lowering it to 1.0 or lower.
A m o n g the markets that really were hit by this combination are rough rice
and Nikkei index. For the Nikkei index, for instance, the profit factor decreased

CHAPTER 13

Long-term Volatility Filters

TABLE

251

13.8

Performance measurements for all markets for the SDB/ADX combination.

After filter (modified system with exits and filter)
Market

P. factor

Avg. trade

2 St. dev.

Drawdown

Crude oil

2.91

685.61

4,302.23

-1,738.02

T-bonds

2.36

1,633.85

13,332.82

-8,787.06

T-bills

2.09

692.96

7,150.36

-7,690.38

Rough rice

0.90

-64.51

3,130.25

-6,798.88

Nikkei index

0.11

-3,477.53

6,978.33

-13,336.11

Natural gas

25.96

6,892.23

20,012.22

-1,104.33

Live cattle

1.00

0.27

3,378.56

-8,096.56

Lumber

1.27

191.77

5,353.82

-6,417.54

Coffee

0.90

-34.70

2,004.95

-4,677.28

Japanese yen

2.47

2,000.98

13,937.97

-13,918.76

Copper

2.59

920.57

8,005.03

-5,594.21

Gold

1.52

416.74

5,333.78

-7,406.50

Dollar index

2.47

1,427.69

7,923.35

-4,357.46

D mark (Euro)

2.51

1,502.94

8,893.76

-8,294.33

Cotton

3.38

877.76

5,145.42

-3,821.89

Wheat

1.84

332.38

3,119.55

-5,114.78

TABLE

13.9

Summarized differences, before/after filter for the SDB/ADX combination.

Differences
Market
Crude oil
T-bonds
T-bills

P. factor

Avg. trade

2 St. dev.

Drawdown

Better

-15.41%

-8.22%

2.85%

-3.62%

83.31%

270.60%

22.03%

-55.81%

-2.59%

2.41%

10.84%

36.69%

Rough rice

-28.76%

-151.29%

2.88%

31.34%

Nikkei index

-97.38%

-156.83%

-70.94%

171.92%

Natural gas

271.87%

342.51%

210.70%

6.04%

-8.07%

-99.44%

11.93%

-5.97%

Lumber

-10.39%

-28.29%

12.27%

15.15%

Coffee

-27.89%

-139.58%

-18.45%

12.50%

0
1

2.70%

16.89%

8.19%

31.61%

125.63%

858.83%

97.67%

-19.15%

19.66%

67.85%

-0.80%

-13.46%

1.38%

-1.40%

-1.20%

0.00%

-15.17%

-17.84%

-6.11%

-31.31%

24.14%

30.91%

13.59%

-1.77%

Wheat

-0.83%

8.13%

0.02%

6.91%

Better

Live cattle

Japanese yen
Copper
Gold
Dollar index
D mark (Euro)
Cotton

252

PART 4

High-probability Filters

from 4.09 without the filter, to 0.11 with the filter. W i t h too few markets benefiting in the form of lower standard deviation of the o u t c o m e of the returns and the
m a x i m u m drawdown, it does not matter that other markets, like natural gas, really excel. T h e conclusion m u s t be that the A D X filter does not do its j o b together
with this system and, therefore, must be put aside in favor of the true range filter.

THE DIRECTIONAL SLOPE SYSTEM

For the directional slope s y s t e m I d e c i d e d to s h o w only those tables depicting the
s u m m a r i z e d differences, simply b e c a u s e neither of the two filters m a n a g e d to
a d d to the p e r f o r m a n c e . Table 13.10 shows the differences b e t w e e n trading with
a n d w i t h o u t the true r a n g e filter. Table 13.11 shows the differences b e t w e e n
trading with a n d w i t h o u t the A D X filter. T h e s e tables are directly c o m p a r a b l e to
Table 9.10.
T h e use of the A D X indicator as a filter in systems like these always is questionable, at best. However, the above results do not rule out that the A D X indicator m i g h t work very well together with other systems, s o m e of the t i m e and/or
traded on s o m e other markets. But to m e , a test like this shows that this indicator

TABLE

13.10

Summarized differences, before/after filter.

Differences
Market

Avg. trade

2 St. dev.

Drawdown

-13.45%

-23.14%

-7.08%

36.24%

-3.29%

2.66%

-6.85%

4.38%

Orange juice

-10.60%

-19.86%

-8.11%

10.49%

Live cattle

-10.81%

-58.90%

-7.21%

17.61%

Lumber

-23.44%

-35.52%

-6.53%

24.08%

Coffee

-10.44%

-16.48%

-2.38%

-9.31%

Corn
S&P 500

Japanese yen

P. factor

Better

-3.00%

-8.86%

-4.06%

6.74%

Copper

-27.55%

-48.55%

-1.50%

2.39%

Gold

-28.93%

-81.58%

-4.25%

59.60%

14.57%

22.10%

0.47%

-9.70%

Dollar index

9.29%

18.43%

3.66%

-8.42%

Cotton

4.16%

2.51%

-1.36%

-23.55%

Eurodollar

CRB index

-35.98%

154.00%

-17.92%

26.24%

Crude oil

-21.23%

-19.88%

-6.81%

-10.99%

Canada dollar

-6.42%

-25.23%

-1.04%

9.01%

T-bonds
Better

-5.02%

-19.99%

0.40%

12.35%

13

CHAPTER 13

253

Long-term Volatility Filters

TABLE

13.11

Summarized differences, before/after filter.

Differences
Market

P. factor

Avg. trade

2 St. dev.

Drawdown

Better

Corn

-38.94%

-37.13%

4.30%

155.44%

S&P 500

-10.60%

46.20%

4.39%

-15.70%

Orange juice

-20.65%

-19.25%

16.06%

86.99%

Live cattle

-11.90%

-62.62%

-11.03%

28.56%

Lumber

^13.84%

-67.32%

-2.69%

17.19%

Coffee

-12.38%

-14.32%

3.05%

8.34%

Japanese yen

-18.72%

-30.38%

-15.31%

67.02%

Copper

-18.09%

-18.57%

21.57%.

-11.58%

Gold

-8.18%

-12.10%

13.98%

-23.63%

Eurodollar

-2.30%

19.91%

12.07%

-18.51%

-36.89%

-65.45%

-13.76%

28.87%

34.31%

60.46%

7.93%

^5.17%

CRB index

-58.62%

295.44%

-25.85%

-9.55%

Crude oil

-25.59%

-22.13%

-1.74%

47.38%

Canada dollar

-17.99%

-60.76%

12.89%

48.30%

6.22%

42.15%

8.42%

12.07%

Dollar index
Cotton

T-bonds
Better

is not robust and reliable enough for me to allow it to m a k e its way into any of my
systems. In fact, as soon as I see a system developer speaking at a seminar about
filtering w i t h the A D X indicator, I start to wonder if this g u y really knows what he
is doing.
B e c a u s e I h a d a pretty good idea beforehand about the results of the longt e r m filters, I debated even bringing up the topic. W h a t finally m a d e me decide to
do it was my belief that there is no such thing as a "failed" experiment, and that
even when the hypothesis leading up to the test cannot be proved, the results are
still worth m e n t i o n i n g so that others w o n ' t spend their time on unnecessary work.

CHAPTER

14

What Makes a Trend

This chapter was cowritten with Max von Liechtenstein, graduate student majoring in economics at Uppsala University, Sweden, and influenced by the ideas of
Richard Werner, chief economist, Profit Research Center, Tokyo, Japan; Edgar
Peters, chief investment strategist, Panagora Asset Management, Boston; Dr. Henry
Pruden, Professor, Golden Gate University, San Francisco; and Dr. Bruno Latour,
Professor, Ecole Nationale Superieure des Mines, Paris, France and Dr. Knut
Wicksell (d. 1926), Professor, Lund University, Sweden.
Popular histories from LaSalle and Wall Street tell the story of tycoons and
market operators like N a t h a n Rotschild, Jesse Livermore, and George Soros. But
underneath these colorful stories there also runs a m o r e serious history of the development of the financial system as the heart of capitalist society. T h e s e stories also
provide insights into why technical analysis and rule-based trading in general are
profitable if understood and treated correctly.
O n e lesson we learn from the tycoons and market operators is that failure in
the financial markets is based on lack of discipline and knowledge, and that trading is a business and a craft carried out in a work process rather than as a set of individual and isolated decisions. For those w h o do not want to understand the markets,
calling t h e m " r a n d o m " is an easy way out, but without knowledge, the individual's
chances of meaningful profits are greatly reduced. Granted, there always will be a
few uninformed traders w h o will m a k e incredible profits. However, these results
are usually not based on knowledge, but on luck. That is, trading limited to the act
of isolated decisions is plain gambling with or without a favorable outcome, while
truly skillful trading is a work process, which interweaves tasks like analysis and
m o n e y and portfolio m a n a g e m e n t .
255

256

PART 4

High-probability Filters

To trade mechanical trading systems, we m u s t understand that our w o r k

process contains m u c h m o r e than the actual decision to b u y or sell something as
signaled by the system. A general p r o b l e m with today's finance is the preoccupation with a s s u m i n g and studying the separate decisions of individuals m a d e at a
fixed point in time. A better perspective is to assume and study individual work
processes, w h e r e each decision is not only a decision by itself, but also part of a
larger plan that might take a certain amount of time to w o r k through.
We deliberately u s e the word "work," because w h e n today's general k n o w l edge of finance only focuses on the single decision and the allocation of wealth,
the part missing in the equation is h o w this wealth is created in the first place. O n e
c o n s e q u e n c e of this is the widespread belief that market fluctuations are a source
of wealth a n d that fortunes are m a d e or destroyed by prices going up and down.
T h a t is not so. Prices are a product and not a cause of wealth. Prices and v o l u m e s
are therefore just materialized wealth, which is caused and created by the aggregated work processes of individuals in the economy. Therefore, you m u s t put in
effort, and work and c o m p e t e to get wealthy from price changes, or else j u s t g a m ble and lose.
We believe the concept of w o r k to be a better starting point for discussing
profitable rule-based trading. T h e advantage with the concept of work, in c o m parison with decision, is that trading can be understood as a form of problem solving. T h u s , work is w h a t we do by carrying out different tasks in a working role. If
an individual wants to learn the work of rule-based trading, he m u s t develop an
ability to solve the p r o b l e m of m a k i n g m o n e y in financial markets by using the
four P's of speculation: philosophy, principles, procedures, and performance.
T h e emphasis in this chapter is on philosophy, whereas the rest of the book
is about principles, procedures, a n d performance. Philosophy answers the question
why, and explains the value of the suggested principles and procedures in the
book. However, it is hard to discuss a philosophy without points of reference. It is,
therefore, essential to understand the development of general knowledge in the
field of finance and w h e r e we stand today. We try to do this as it relates to how to
isolate the underlying long- and short-term trends.
T h e development of general knowledge of the financial markets has taken a
curious route since the 1950s. While the a c a d e m i c world has been searching for
general a n d abstract theories that could be derived and proven mathematically,
practitioners have concentrated on relevant principles that have proven practical to
use in their own working environments. Before the 1950s, both practitioners and
academicians participated in the creation of general knowledge. T h i s meant that
practitioners, especially within the subgroup of financial analysts, influenced what
was taught in higher learning institutions. After the 1950s, however, the academic
world alone to a large extent created the general knowledge of the financial markets; important k n o w l e d g e regarding the financial market from traders and technical analyst simply was overlooked. T h e p r o b l e m with the d o m i n a n c e of the

CHAPTER 14

What Makes a Trend

257

scholars is the lack of relevant practical principles adapted to the "real-world"

restrictions in the practitioners' everyday w o r k processes.
T h e scholars' d o m i n a n c e in defining the general knowledge of the financial
markets started in the area of portfolio theory. In the early 1950s, H a r r y
Markowitz showed how a stock portfolio, with a certain objective of expected
return, could be optimized to have the lowest possible variability in periodic
return. In the early 1960s, the academic society strengthened its d o m i n a n c e over
financial theory through William Sharpe's Capital Asset Pricing Model (CAPM)
and E u g e n e Fama's Efficient Market Hypothesis (EMH). C A P M states that an
asset price, with a certain expected return, depends on its risk relative to the market. E M H states that, on average, competition will cause n e w information regarding intrinsic values to be reflected instantaneously in actual prices. T h e scholars'
d o m i n a n c e of general knowledge consolidated in the 1970s with a set of option
pricing m o d e l s by Fisher Black, Robert Merton, and M y r o n Scholes.
T h e cornerstones of m o d e r n finance are the arbitrage principle, formed by
M e r t o n Miller and Franco Modigliani, and the random walk hypothesis (RWH). In
m o d e r n finance, arbitrage is always possible, because investors can borrow freely.
Furthermore, in an informationally efficient market, price changes m u s t be u n p r e dictable a n d r a n d o m . T h e randomness is achieved through competition between
active rational investors seeking greater wealth. Arbitrage between competitors
eliminates all profit opportunities by incorporating available information instantaneously into m a r k e t prices. Therefore, according to the scholars, a truly efficient
market is one where price changes are completely r a n d o m and unpredictable. T h e
implication is that investors should use probability theory to m a k e rational investment decisions in r a n d o m real and financial markets.
In the past few years s o m e scholars, within the field of behavioral finance,
have attacked m o d e r n finance and received strong support from traders and technical analysts in the Wall Street community. T h e attackers believe that financial
asset prices change in a predictable, biased, r a n d o m process rather than in an
unpredictable r a n d o m w a l k and that the persistence in behavior of prices and the
way they change are explained by social factors like ideas and h u m a n bias.
Because of this, price changes in financial markets are likely to and will trend
from time to time. B u t even if behavioral finance is a good step forward, it is limited to the concept of decision m a k i n g by only looking at social factors, such as
individual wants or needs at specific points in time.
If changes in asset prices follow a biased r a n d o m walk, then they are to some
degree predictable. T h u s , fundamental analysis can be used to beat the market.
Small firms a n d value strategies have shown this. A n o t h e r insight from practitioners is the value of rule-based trading. It has been s h o w n that asset prices tend to
overreact and that, for instance, relative strength strategies, therefore, can be profitable. This, too, contradicts m o d e r n finance, which considers price changes to be
essentially unpredictable.

258

PART 4

High-probability Filters

However, social factors only explain half the story and m u s t be complem e n t e d by m o r e material factors, forming a global structure that explains the creation and flow of the necessary funds for materializing these " w a n t s " and "needs."
Together, the social factors (the wants) and the global structure (the funds, or flow
of funds) create the necessary d e m a n d to create different markets for all types of
g o o d s and services, other factors of production, and financial assets. But that is
not all. Together they also form a complex process of at least partially predictable
trends in several different markets intervening with each other. This stresses the
need to look at any type of trading as a rational work process rather than a quest
for a set of optimal decisions at different fixed points in time. This way of looking
at trading also emphasizes the need for m o n e y m a n a g e m e n t and portfolio management, which, outside of the world of rational investors, b e c o m e s something
completely different from just buying a n d holding the m o s t efficient portfolio in
accordance with C A P M a n d E M H .
T h e rule-based approach to trading is essentially a reflection of the idea that
prices move in at least partially predictable trends and by following the s a m e strategy, or work process, you end up being right m o r e often than not and/or make
m o r e m o n e y than you lose. T h e following e x a m p l e illustrates what we mean by a
trend b e i n g partially predictable.
In his book, Patterns in the Dark (Wiley, 1999), E d g a r Peters describes a
j u d g e w h o goes m a d . T h e j u d g e ' s madness stems from a simple observation: every
day, b a d people cause b a d things to happen. Could so m a n y tragedies simply be a
string of r a n d o m events? N o , according to the j u d g e , they m u s t be the result of a
conspiracy. This story captures our ability and desire to i m p o s e a stable order and
on specific explanations for what is happening around us in our everyday life, even
w h e n there are n o n e . T h e observation that bad people cause bad things to happen,
however, implies that there exists a stable and certain structure. What the judge
missed is that most victims are still singled out by m o r e or less purely random
events. T h u s , good people b e c o m e victims due to local randomness in joint action
with a global structure.
In nature, there are m a n y complex processes (for example the weather) that
consist of several elements acting as an entity, forming local randomness in joint
action with global structure. This is also true in technological and financial (technofinancial) systems. On the one hand, individuals act together, forming a global structure based on similar material prerequisites and conditions. But because of different
social factors, such as individual preferences and willingness to change, they also
create local randomness. The complexity of it all causes the evolving entity to look
like a conspiracy, although no mastermind is behind it.
In his book, Aramis or the Love of Technology (Harvard University Press,
1996) B r u n o Latour points out that if we want to understand material changes like
price and v o l u m e changes in traded assests, we need to define a n d integrate social
and material factors in the s a m e model. In other words, in the economy, the joint

CHAPTER 14

What Makes a Trend

259

action between local randomness a n d global structure forms an at least partially

certain and stable change of events within a complex process. This joint action
forms a trend of some sort in something we try to predict or explain. This implies
that e c o n o m i c trends are partly created by us as we interact and take part in forming these changes with our predictions and explanations.
Up until recently, standard neoclassic e c o n o m i c s and m o d e r n finance have
b e e n occupied with explaining e c o n o m i c fluctuations and g r o w t h in the gross
domestic product ( G D P ) only as a result of social factors, such as investors' allocation of purchasing power, neglecting h o w this purchasing power is created in the
first place. This logic depends on the assumption that market prices and quantities
always equal the value they offer the individual. T h i s assumption also applies to
the markets, where the net present value ( N P V ) equals the intrinsic value of an
asset, because arbitrage otherwise would be possible. In both cases, the reasoning
is justified by an assumption that individuals always can borrow freely. As a consequence, change in G D P is explained only by social factors such as individual
preferences, b e c a u s e this is the only variable left that can explain the portfolio
allocation decisions in the economy. This implies that intrinsic values c o m m a n d or
determine prices and quantities in the markets.
But, as we all know, we cannot b o r r o w freely, and this m a k e s the inclusion
of individuals' budget or i n c o m e constraints in absolute terms a great fallacy, in
neoclassic economics and m o d e r n finance because even Mr. Joe Sixpack acknowledges the p r o b l e m with allocating a nonexistent purchasing power. This problem
stems from the neoclassic e c o n o m i c s ' definition of purchasing power and m o n e y
as savings a n d deposit aggregates like M l , M 2 , etc., which, in turn, has its origin
in a too-simple m o d e l b a s e d on a barter e c o n o m y instead of a credit-driven capitalist economy. Because the savings themselves are a result of the creation and
allocation of purchasing power, this has led to difficulties in explaining financial
crisis, growth, and fluctuations in both G D P and asset prices and why these
processes to s o m e degree are predictable.
To solve this p r o b l e m s o m e scholars inspired by K n u t Wicksell's book,
Interest and Prices ( A . M . Kelley, 1965) have instead started to explain the creation
of p u r c h a s i n g power as a creation of credit a n d b a n k s ' willingness to lend. With an
emphasis on the material constraints in society, changes in both transaction flows
and G D P can n o w not only be explained by individuals' portfolio allocation alone,
but also by the b a n k s ' credit creation process as an underlying force inducing the
wants a n d needs of individuals. T h e distinction between social and material factors also is important, b e c a u s e financial crisis a n d bubbles otherwise tend to be
explained only by irrational decision-making by individuals rather than by a market failure within the system. Two such important material factors that affect the
allocation of purchasing power are demographics a n d fiscal policies. These factors
also create trends in the markets, because they determine the supply (allocation)
of funds.

260

PART 4

High-probability Filters

By m e a n s of credit, m a c h i n e s are created, managed, and fed with raw materials. By m e a n s of credit, products are c o n s u m e d h o u s e and cars are bought, and
m o n e y is spent abroad by foreign customers. T h a n k s to the credit creation process,
single individuals a n d the economy as a whole can obtain the necessary funds
needed to materialize all our "wants." All our "wants," as explained by behavioral
finance, will not cause the market to trend by itself. Instead, the ability of a market to trend is dependent on how m u c h a n d what type of credit it can attract. That
is, short- and long-term trends are separated by whether the purchasing power is
created inside the market, as investors speculate on margin, or c o m e from outside
flows.
T h e importance of credit in understanding markets can be illustrated by comparing our capitalist society with the barter economy of ancient Egypt. T h e econ o m i c world of Pharaoh w a s very simple. He j u s t needed to use his whip and
power to c o m m a n d labor and material to build pyramids. T h e pyramid of Cheops
c o n s u m e d 600,000,000 days of h u m a n labor and all that was created after a period of 20 years was a frozen asset.
In our economy, there is no solitary pharaoh with the power to c o m m a n d labor
and material by word alone. In our economy, labor and material is c o m m a n d e d by
money. In a barter economy, m o n e y is viewed as a tangible, " i m m o r t a l " asset, the
quantity of which will only change as m o r e gold or other precious material can be
d u g up from the ground. In a m o d e r n , capitalist society, however, the life cycle of
m o n e y and credit is such that it is born when a b a n k issues a loan and is put to death
as the person w h o borrows it pays it back. There is neither value nor power in
m o n e y itself, only in what it r e p r e s e n t s c o m m a n d of labor and material.
In place of Egypt's single pharaoh, we have several pharaohs seeking comm a n d of labor and material by m e a n s of credit. At first, it does not really matter
h o w this money, or credit, is used, because it creates d e m a n d a n d transactions in
the economy. T h u s , at first, credit, or money, is a fabulous agent both for borrowers and others in the society. However, if credit c o m m a n d s labor and material into
something unproductivefor e x a m p l e , to build a p y r a m i d t h e n this capital loss
is really lost labor. Further, credit that cannot be repaid hurts society, because it
discourages b a n k s ' creation of n e w creditthe c o m m a n d of labor and material for
u p c o m i n g periods.
T h e price borrowers have to pay to c o m m a n d labor and material is annual
interest. To increase the likelihood for the debt to be amortized (or put to death),
banks often require collateral for their lending. T h e s e restrictions have a functional significance because they force borrowers to m a k e sure that their c o m m a n d of
labor and material will be productive e n o u g h to generate income that pays for
interest, repayment, and s o m e profit. T h u s , credit constraints restrain pyramid
building. Nevertheless, our m o d e r n world is filled with thousands of "pyramids,"
and the invisible ones are the m o s t dangerous, because they can survive only by
new loans.

CHAPTER 14

What Makes a Trend

261

In his article, "Towards a N e w M o n e t a r y P a r a d i g m " (Kredit und Kapital,

1997) Richard Werner points out that there is a fundamental difference in outcome
b e t w e e n transactions in the real and financial markets. Transactions in real markets for g o o d s , services, and factors of production, involve income or production
flows. Financial market transactions involve asset stocks. Transactions in real markets are an e c o n o m i c activity measured as GDP. Transactions in financial markets
are a noncreative activity that is not a part of GDP. Thus, when changes in asset
prices or G D P are explained or forecast, total credit creation m u s t be subdivided
into a real and a financial part. Quantities of stocks are m o r e fixed in c o m p a r i s o n
w i t h production flows. Therefore, in the real market, an increase in d e m a n d is
mainly m e t by an enlarged quantity supplied, whereas in the financial market,
w h e r e the supply is m o r e limited, an increase in d e m a n d will, by necessity, result
in an increase of price.
A stock-exchange speculation with ever-increasing prices, such as that
before the crash of 1929, is an invisible pyramid. Its stones are the supply of funds
and h u m a n sentiment. T h e " 1 9 2 9 p y r a m i d " c o n s u m e d credit in an uncontrollable
manner. In two years, b r o k e r s ' loans on the N e w York Stock E x c h a n g e alone
increased by $5 billion, which only served to inflate stock prices. T h e pyramid
caused a great n u m b e r of problems. For one thing, credit could have been used for
m o r e productive p u r p o s e s , b u t w a s not. But even m o r e seriously, it caused the
lenders substantial losses, w h i c h decreased production and e m p l o y m e n t (i.e. c o m m a n d for labor and material) during the period that followed.
Today, we o n c e a g a i n s p e a k of the " n e w e c o n o m y " as we e x p e r i e n c e a
p e r i o d of h i g h g r o w t h in GDP, a b o o m i n g stock m a r k e t , and a low g r o w t h in
consumer prices. This is because new and improved technology has made it
p o s s i b l e t o m e e t t h e i n c r e a s e d d e m a n d for g o o d s and s e r v i c e s w i t h i n c r e a s e s i n
productivity, t h u s i n c r e a s i n g G D P w i t h o u t i n c r e a s i n g inflation. A t the s a m e
t i m e , t h e r e has been a s h a r p shift in d e m o g r a p h i c s , r e s u l t i n g in a d r a m a t i c
increase in the ratio of savers to s p e n d e r s . A lot of these savers have invested
their s a v i n g s in m u t u a l funds, w h i c h m e a n s a l o n g trend of m o n e y flowing out
of t h e real m a r k e t s a n d into (allocated to) the financial m a r k e t s .
A d d to this that several governments throughout the world have both lowered
their taxes and started to b u y b a c k h u g e a m o u n t s of b o n d s issued in the 1980s (the
United States is supposed to have bought b a c k all its debt by 2 0 1 3 ) . This also
transfers m o n e y that would have been spent in the real markets to the financial
m a r k e t s . All in all, this has created a situation of high e c o n o m i c growth, coupled
with a very low inflation in the real markets, but with highly inflated financial
markets. Once again, as during the margin lending of the 1920s, the b o o m i n g markets have m a d e banks lend heavily for speculative p u r p o s e s . Normally, this would
result in higher interest rates, as central banks fight expected inflation and private
banks d e m a n d a higher risk p r e m i u m for m o n e y lent for speculative purposes.
However, the interest rates are b e i n g h a m p e r e d by competition a m o n g the banks,

262

PART 4

High-probability Filters

governments decreasing outstanding stock of b o n d s , and the fact that there is no

inflation to fight in the real economy.
In times w h e n banks lend heavily for speculative p u r p o s e s , such as the margin lending of the 1920s or as is seen today, asset price inflation occurs, although
c o n s u m e r prices m a y hardly rise at all. Banks tend to extend loans with real or
financial assets as collateral, w h e r e an increase in collateral values alleviates credit constraints. If banks extend loans with real or financial assets as collateral, then
an increase in collateral values alleviates credit constraints. T h u s , invisible pyram i d s of extreme changes in asset prices always d e p e n d on increasing asset-related
lending from the banks. T h e paradox is, however, that while individual banks
assume asset prices to be independent of their actions, rising asset prices cause
extended asset-related loans, which lead to a continued increased d e m a n d for
financial assets.
Technical analysis is increasingly recognized for explaining price predictability based on a study of investors' actions by charting the development of traded volu m e and price of a financial asset. This school of thought in general, and rule-based
trading in particular, perceive market behavior to be both deterministic and random.
Technical analysis can be broken down into four essential areas: creation-of-funds,
flow-of-funds, sentiment, and market microstructure indicators. T h e creation-offunds and flow-of-funds analyses represent the m o r e stable trends that are reflected in the sentiment and market microstructure analysis. The creation-of-funds indicators include central b a n k s ' open market operations and private b a n k s ' loans to the
total economy, subdivided into financial and real sectors. Flow-of-funds indicators
analyze the financial position of various investor groups in an attempt to measure
their financial capacity for buying or selling stocks. Flow-of-funds analysis is concerned with trends in mutual fund cash positions and other major institutions such
as pension funds and insurance companies. Other flow-of-funds indicators are new
equity offerings, secondary offerings, and c u s t o m e r s ' free balances, which are normally a source of cash on the b u y side. T h e use of margin (or leverage) borders is
both a long-term creation-of-funds indicator, as the use of margin is a form of borrowing, and a shorter-term sentiment indicator, as the current use of margin by larger market participants oftentimes can reflect a m o r e short-term use of what normally is d e e m e d stable or m o r e long-term money.
The sentiment indicators monitor the different market participants such as
mutual funds and floor specialists. They monitor the emotions or expectations of
investors moving from one extreme at a market bottom to another at a market top. It
is assumed that different groups of investors are consistent in their actions at major
market turning points. Advisory services and newspapers, for example, are two
groups that are considered to be wrong more often than not at or just prior to major
turning points. A very good sentiment indicator in the commodity futures markets is
Commodity Futures Trading Commission's (CFTC) C o m m i t m e n t of Traders Report
that monitors the positions of commercial and noncommercial traders.

CHAPTER 14

What Makes a Trend

263

We perceive the material factors to be captured by the creation-of-funds and

flow-of-funds indicators, while the social factors are captured by the sentiment
indicators. However, because of the difficulty to m e a s u r e , quantify, and rank these
indicators' relative importance and h o w they affect different markets at different
points in time, we often m u s t use what we call market microstructure indicators.
In essence, all technical analysis indicators that are based on price, volume, time,
breadth, and volatility are market microstructure indicators.
In his article, "Life Cycle Model of C r o w d Behavior" {Technical Analysis of
Stocks and Commodities, January 1999), Henry Pruden describes how an a d o p tion-diffusion model, based on behavioral finance, can be used to explain the rise
and fall of financial asset prices. Dr. Pruden argues that the adoption-diffusion
m o d e l captures how innovations over t i m e are adopted in a techno-financial system, as illustrated by the familiar bell-shaped normal distribution curve. We have
expanded Dr. Pruden's thoughts to e n c o m p a s s m o r e material factors when we
explain the existence of long- and short-term trends of changing supply/demand
relationship in the specific market.
An idealized market cycle begins with a consolidating phase, known as
either a distribution or accumulation area. T h e S-shaped cumulative normal distribution curve in Figure 14.1 illustrates what this m a y look like. In a consolidating phase, the d e m a n d and supply of assets are in a relative equilibrium. In an
accumulation (distribution) phase, assets pass from weaker (stronger) to stronger
(weaker) h a n d s . But asset prices can only advance if investors have the purchasing
power and courage to buy. That is, the credit creation process together with other
material factors, such as demographics and fiscal policies, induce the necessary
funds, while at the s a m e time the social factors, such as all our wants and needs,
are strong enough to entice buying.
S h o r t - t e r m t r e n d s are c a u s e d b y i n v e s t o r s ' s p e c u l a t i o n o n m a r g i n , w h i c h
forces the m a r k e t prices b e y o n d w h a t the e c o n o m y can afford in the long run.
T h e r e f o r e , a highly leveraged trader has a w e a k h a n d . W h e n e v e r the value of
t h e leveraged trader's h o l d i n g s s h r i n k s , he m u s t sell a p o r t i o n of his h o l d i n g s .
W h e n the p r o p o r t i o n of stocks in strong h a n d s is a b n o r m a l l y high, the m a r k e t
is in a strong p o s i t i o n . T h e r e f o r e , t h e longer the p e r i o d of a c c u m u l a t i o n , the
g r e a t e r t h e a m o u n t of strong h a n d s a n d the larger the b a s e from w h i c h p r i c e s
can rise. In real life this can be a p p r o x i m a t e d with the help of a l o n g - t e r m m o v ing a v e r a g e , a p p l i e d either directly on price or to the n o w familiar O B V indicator, as in F i g u r e 1 2 . 1 . T h e benefit of a p p l y i n g this t y p e of analysis to the
O B V indicator, r a t h e r t h a n directly to the p r i c e , is that the O B V indicator, as
s u g g e s t e d by its n a m e , w e i g h t s e a c h day's price action w i t h its t r a d i n g v o l u m e
a n d by d o i n g so b e c o m e s a better m e a s u r e of the relative a m o u n t of participation b e h i n d t h e trend.
It also is possible, with the trend creation process still in mind, to approxim a t e a society's short-term willingness to allocate m o n e y (flow of funds) toward

264

PART 4

F I G U R E

High-probability Filters

14.1

The cummulative normal distribution of a dependent variable.

a certain t y p e of investment vehicle by looking at the relative p e r f o r m a n c e of the

investment alternative in question as c o m p a r e d to the interest rate. W h e t h e r the
interest rate is currently increasing or decreasing does n o t matter in this analysis
as long as it does so at a rate different than that of the alternative investment. For
e x a m p l e , if the interest rate is decreasing m o r e slowly (bond p r i c e s m o v e up)
than, for instance, the price of c o m is increasing, but faster than the increase in
the p r i c e for wheat, it is reasonable to a s s u m e that investors prefer to invest in
b o n d s rather than wheat, but also that they favor corn over interest rates. This
m e a n s that, in this particular case, y o u should look for situations to short the
w h e a t a n d g o long the c o m .
For you as a trader, this m e a n s that as long as you can keep a rate of increase
in your investment or trading account that is higher t h a n a combination of the
interest rate at w h i c h the invested m o n e y is a s s u m e d to have b e e n borrowed, the
rate of inflation, and a reasonable p r e m i u m for the risk you are assuming, you are
doing O K . B u t to do better than j u s t O K t o be competitiveyou also must do
better t h a n y o u would have d o n e using any other investment alternative or method.
T h a t is w h a t the rest of this b o o k is all about.
Returning to the analysis of a trend, unfortunately, it is seldom as simple as
j u s t isolating o n e trend from all the others a n d then riding that wave. W h e n sever-

CHAPTER 14

What Makes a Trend

265

al different ideas, preferences, and n o r m s interweave, a larger trend naturally consists of several smaller ones, m a k i n g life difficult for any trader. T h e theoretical
explanation behind these shorter-term trend and m o m e n t u m changes is illustrated
by Figures 14.2 and 14.3. In Figure 14.3, we can see h o w this long-term trend really consists of two distinctly different states of market behavior. In this case they
also are separated by a transition phase, forming the well-known halfway flag of
technical analysts.
To c o m e to grips with this, m a n y technical analysts try to isolate different
trends from each other using, for instance, different moving averages applied
directly to the price, again on the assumption that the longer the average, the m o r e
stable and long-term the trend. In this case, however, we do not concern ourselves
with whether the trend is driven mainly by material or social factors, although
there also is the underlying assumption that the material factors are m o r e long
t e r m than the social factors.
To isolate both the long-term and short-term shifts in the trend caused by, for
instance, sentiment shifts a n d the a m o u n t of trading taking place on margin, m a n y
analysts also use one or several oscillator-type indicators that could be applied
either directly to the price or to any other indicator that attempts to capture
the material or social factors m o r e explicitly. In this case, the longer the lookback

FIGURE

14.2

Two normal distributions intervening with each other.

266

PART 4

F I G U R E

High-probability Filters

14.3

The cummulative effect of two normal distributions phased together.

period for the oscillator, the m o r e significant and reliable the shift is assumed to
be, because it then is a s s u m e d to be the result of m o r e material factors, such as the
creation and flow of funds and/or sentiment shifts a m o n g the larger players, such
as the pension funds a n d insurance c o m p a n i e s .
M o r e formally and in accordance with our reasoning so far, it could be argued
that these stages of unpredictability coincide with periods of major shifts in the
long-term, stable trends and a high amount of margin trading (short-term credit
creation and flow of funds) a m o n g the weaker hands, as these players scramble to
figure out w h e r e the larger players are heading. That is, the shorter the credits, the
higher the degree of uncertainty regarding the material factors and the m o r e unpredictable the markets, because these shorter credits inseminate a larger degree of fear
and panic into the mixture of social factors affecting the traders.

PART

FOUR

A Few Final Thoughts About Part 4

This section started out by taking a closer look at how to filter out those situations
w h e n the m a r k e t s seem to greet us with a particularly friendly environment, suitable for the entry/exit triggers we currently are working with. W h a t it all c o m e s
down to is to increase your chances to place a winning bet, both in the short and the
long run. Whatever your preferred time perspective it never hurts to get an understanding of w h i c h underlying forces are at play to create a n d maintain the moves
we are interested in and are trying to capitalize from.
We n e e d to understand that the predictability and profitability is there if we
could only find a way to m e a s u r e it a n d transform it into dollars m a d e . T h e n you
" o n l y " will n e e d to be able to k n o w at w h a t scale you are looking so that you k n o w
h o w to distinguish one seemingly r a n d o m m o v e from another. Granted, the technical analysis and statistical tools we are using are very coarse a n d clumsy, and for
the p u r p o s e very unsophisticated. B u t however clumsy the tools, this is as good as
it gets, a n d at least with a proper understanding for how a trend gets started, the
right framework for analyzing it a n d a set of generalized rules, as those in a
m e c h a n i c a l trading system, we can b e c o m e good e n o u g h in recognizing the general differences between an uptrend and a downtrend, and if the market seems to be
in a state of predictable change or not.
For the long-term perspective, you can look at it this way: Pretend that the
e c o n o m y (or the part of the e c o n o m y that you are interested in) is a super tanker
h e a d i n g in a certain direction on the ocean. But instead of being full of oil, it is full
of people, single individuals, a m o n g w h o m you are one, w h o in one way or another have an interest in what is going on, j u s t as you d o . It could be other traders, b r o kers, analysts, farmers, regular workers, and all these people's families, relatives,
a n d friends.
267

268

PART 4

High-probability Filters

Pretend further that the person in charge of this ship, for instance Alan
Greenspan, has decided that the ship is heading in the w r o n g direction, or that he
can see trouble ahead, and decides to change course. Even if he puts the engines
into full reverse, the sheer mass and m o m e n t u m of the ship force it to continue in
the original direction for many miles to c o m e . Not even if everybody on the ship
agrees with his decision and wishes to turn around right away will the ship be able
to turn around any sooner. Wishful thinking has nothing to do with where the markets are heading or the o u t c o m e of your trading.
Now, pretend that you, as a trader, instead are standing on the shore, looking
at this ship on the horizon, holding exactly the s a m e information as everybody on
board. W h a t would you bet on: that it will continue in the same direction for
another couple of miles, or minutes, or days, or that the next thing you know, it
will have turned around? Of course, you are better off betting it will continue in
the s a m e direction. You do not mm a super tanker on a dime; not even if you close
your eyes hard and wish. A n d the s a m e goes for the economy, or even any specific smaller part of it, like an individual stock.
To be sure, in the short run strange things happen every now and then
exceptions to the rule bring short-lived fame and fortune to a lucky few. But they
are just thatexceptions to the rule. Generally speaking, you are m u c h better off
going with the current flow of events. On the financial markets, this m e a n s placing your trades in the s a m e direction as the trend is heading. Place a long-term bet
that tries to catch the beginning of the move, but be prepared to get out as soon as
it is clear it is not heading w h e r e you expected or w h e r e the situation is less clear,
as indicated by high short-term volatility levels.
Or to continue the analogy, w h e n you can see bad weather approaching or
w h e n mist or fog m a k e s it difficult to assess the situation, that is when your filters
should signal to stay out. To do that, you m u s t k n o w what to look for, how to measure it and h o w to m a k e the most of it, w h i c h s u m s up the p u r p o s e of this book and
all the hard work w e ' v e d o n e leading up to and including the "optisizing" process
in Part 5.

PART

FIVE

Money Management and Portfolio Composition

Now that we have finished building the engines, let us move over to the intricate
subject of putting together a gearbox and a chassis to get our babies going. But first
I need to point out that there is no way I would have been able to c o m e up with the
math a n d the logical reasoning behind this section myself. For this I rely solely on
Ralph Vince, w h o also coined m o s t of the terminology I use in this section, in his
b o o k s Portfolio Management Formulas (John Wiley & Sons, 1990) and The
Mathematics of Money Management (John Wiley & Sons, 1992). I have merely
taken what is in these b o o k s and tweaked it around a little bit to suit my needs. If
you are interested in learning m o r e about what is covered in the following section,
d o n ' t ask m e g o buy Mr. Vince's b o o k s .
W h e n it c o m e s to the gearbox (the m o n e y m a n a g e m e n t ) , it is quite simple
really. Because in a well-working system you really have no idea about whether
your next trade will be a winner or a loser, you will be best off, in the long run,
always and for each trade, to risk the same fraction of your account. That is, if your
account size going into a particular trade is $100,000 and you have decided that if
the trade goes w r o n g , you can stand to lose 2% of that amount, you will be willing
to lose $2,000. But if your account size is only $50,000, a 2% loss equals only
$1,000, w h e r e a s a $2,000 loss equals a fraction of 4 % .

269

CHAPTER

15

Money Management

The w h o l e idea of m o n e y m a n a g e m e n t is to always invest that fraction, or percentage, of your account going into each trade, m o s t suitable to you, considering such
factors as account size, expected worst-case trades, drawdowns, flat times, growth
rates, a n d dependency a m o n g trades. To be able to do thisto " o p t i s i z e " your tradi n g y o u m u s t start to think in fractions a n d percentages, rather than in flat dollar
rates. Otherwise, y o u will not be able to practice the very s a m e techniques that
already are in practice w i t h professional m o n e y m a n a g e r s .
A n o t h e r important point is that although these techniques add m o r e to your
b o t t o m line than your actual systems, or entry and exit triggers, they do not p e r f o r m
miracles. For these techniques to work, your s y s t e m s m u s t have positive m a t h e matical expectations in the first place. T h a t is, they must have a profit factor above
1 and a positive value for the average trade to be worthwhile. If they just have that,
if they are only marginally positive, w i t h the p r o p e r m o n e y m a n a g e m e n t you can
w o r k wonders.
On a separate note, however, it is w o r t h m e n t i o n i n g that this only h o l d s true
for the special case of one m a r k e t - o n e system. As soon as you trade a portfolio of
m a r k e t s and/or systems, it is fully possible to have individual market/system c o m binations that have a negative mathematical expectancy (a profit factor below 1 a n d
a negative value for the average trade), w h i c h m a y raise the positive expectancy for
the trading strategy as a w h o l e . T h e trick is to put together a portfolio of systems
and/or m a r k e t s with correlations as low as possible to each other, w h e r e a few m a r kets still zig w h e n others zag, b u t with the zigging greater than the zagging, generating a small but steady profit in the long r u n . T h e negative expectancy
markets/systems help to hold up the b o t t o m line while you wait for the positive
271

PART 5

272

Money Management and Portfolio Composition

expectancy markets/systems to m o v e b a c k in line a n d start performing again.

Before we go on there are a few things you must fully understand:

Price does not behave in a rational manner.

Potential gain is not a straight-line function of potential risk; it is not true that
the m o r e you risk, the m o r e you are likely to gain.

T h e a m o u n t of risk you assume has nothing to do with which type of vehicle

you are trading.

Diversification does not necessarily have to reduce drawdown.

Of these four things, we immediately must address the fact, that for money
m a n a g e m e n t purposes, w h e n price does not behave in a rational m a n n e r it means
that if it goes from, for instance 1,350 to 1,000, it does not have to stop at every
tick in between. If you are caught on the w r o n g side of such a m o v e you will suffer a severe, if not totally devastating, loss, because there will be no way out even
if you had a stop loss at 1,345. F u r t h e r m o r e , because these moves happen, the only
thing that is certain in trading is that we all will be blown out of the markets sooner or later. It is just a matter of time. Or to use a simple analogy: if you can only
die by from being hit by a truck, eventually you will die from being hit by a truck.
It is as simple as that.
In the world of trading, however, w h a t might be devastating for some may
only be a freak occurrence but still a fully expected outcome for others. From this
it follows that to fit as m a n y outcomes as possible into the boundaries of what
could be considered expected, you m u s t have considered and treated the whole
subject of trading very defensively. Or, to use the old start-a-new-business cliche,
m o n e y m a n a g e m e n t ( m a k i n g a business plan) is to h o p e for the best but prepare
for the worst. Still, for some the catastrophe might c o m e first thing tomorrow, no
matter h o w m u c h they have tried to protect themselves. If it h a p p e n s to you, I ' m
sorry, but that's j u s t the n a m e of the g a m e .
Now, let us consider the next point in the list above. "Potential gain is not a
straight-line function of potential risk." If it is not, then what is it? To answer that
question we first m u s t define a few t e r m s .
We need to understand the t e r m holding period return ( H P R ) . T h e H P R is the
percentage gained or lost for each specific trade, plus one. For e x a m p l e , if a trade
ends up with a 5% profit, the H P R is 1.05 (1 + 5% = 1 + 0.05 = 1.05). On the
other hand, if a trade ends up with a 5% loss, the H P R is 0.95 (1 + ( - 5 % ) = 1 +
( - 0 . 0 5 ) = 0.95).
T h e second t e r m we must understand is terminal wealth relative ( T W R ) . The
T W R is the geometrically c o m p o u n d e d H P R for all trades, which m e a n s that we
are using multiplication rather than addition. For example, if you have two winning
trades, with H P R s of 1.05 and 1.10, respectively, the T W R is 1.155 (1.05 X 1.10 =
1.155). A T W R of 1.155 means that after the two trades, you have 1.155 times your
original equity, or your total profit is 1 5 . 5 % (1.155 - 1 = 0.155 = 15.5%).

CHAPTER 15

Money Management

273

Note that a profit of 5% and a profit of 1 0 % do not add up to a profit of 1 5 % ,

but slightly m o r e . This is because you were able to use the extra m o n e y you m a d e
in trade one going into trade two. You were able to reinvest your winnings. This
clearly demonstrates the importance of always risking the s a m e fraction of your
account, as represented by your expected worst-case scenario for that particular
trade. To understand this better, consider instead if the two trades above t u m e d out
to be two losers of 5% e a c h w h i c h was the fraction you were willing to risk.
T h e n your T W R would be 0.9025, your equity would have decreased to 0.9025
times the original equity and your total loss is 9 . 7 5 % (0.9025 1 = 0.0975 =
9.75%). T h u s , two 5% losses in a row d o n ' t equal a total loss of 10%, but slightly less than that.
Two major benefits of working with a fixed fractional m o n e y m a n a g e m e n t
strategy, where you always reinvest your profits and losses, is that your account
m e a s u r e d in dollars g r o w s geometrically, and at the same time it keeps you in the
g a m e for m u c h longer during drawdown periods, because the strategy forces you
to scale back and trade fewer and fewer units (contracts, stocks, etc.). (By the way,
if you can m a k e your account grow geometrically, it will take you, on average, the
s a m e a m o u n t of time [trades] to take it from $10,000 to $100,000, as it will to take
it from $100,000 to $1,000,000, or from $1,000,000 to $10,000,000.)
As your account grows, you should trade m o r e and m o r e units, to keep the
relative a m o u n t risked at a constant level. It has always a m a z e d me how most
traders simply do not concern themselves with this, but instead put all their attention in whether or not they m a n a g e d to call the right direction of the move. For
m o s t traders, being right about the direction of the move seems to be m o r e important than actually m a k i n g any m o n e y out of it, not to mention m a k i n g the optimal
amount of money. But, because we have no control over whether the next trade
will be profitable or not, but do have control over the quantity we can put on,
w o u l d n ' t it be better to focus our attention on "optisizing" that amount, once we
have a system that we can trust and with a positive expectation in the long run?
T h e third t e r m we m u s t understand is the v a r i a b l e which is simply the fraction of account size. T h e variable can be optimized in several different ways:
what might be optimal for you might not be optimal for m e . If we only decide to
optimize in relation to the speed with which we wish to m a k e the account grow, it
is called the optimal . With this value for in our hand, we then calculate the "optisized" n u m b e r of units (stocks or contracts) we can put on for any given trade,
given our expected worst-case scenario and current account status. For instance,
suppose we already have calculated to be 0.5 (we will see how to find the optimal value shortly), that we currently have $100,000 in our account, and that the
worst-case scenario is a 1 0 % loss (which in this case happens to equal $10,000 per
unit traded). In this case, we can put on 5 units (100,000 X 0.5 / 10,000 = 5). Had
instead b e e n 0.6, we could have put on six units (100,000 X 0.6 / 10,000 = 6).
T h u s , the higher the the m o r e units we can trade.

274

PART 5

Money Management and Portfolio Composition

To be sure, there are other ways of c o m i n g up with an optimal n u m b e r of

units to trade. For instance, e x a m i n e if there are any type of serial correlations or
dependencies a m o n g the outcomes of your trade. O n e way of examining for
dependency is to do a runs test, which tells you if your system has m o r e or fewer
streaks of consecutive w i n s or losses than w h a t could have been expected if the
trades had been randomly distributed. A n o t h e r way is to look at the linear correlation coefficient.
As we learned in Part 3, if a system has any dependencies or correlations
a m o n g the outcomes of the trades, there is information there that we are failing to
exploit at the system building stage. That is, we should have formulated our entry
and exit rules in such a way that the information the market w a s giving us, by signaling dependency, would have been dealt with at that time. T h i s illustrates the
paradox that with a system working optimally, there is no way to k n o w if the next
trade will be a w i n n e r or a loser.
Furthermore, even if your systems seem to exhibit some serial correlation or
dependency, you probably still are m u c h better off ignoring it as long as the evidence is not truly and really overwhelming "beyond any reasonable doubt." This is
because the evidence is just thatevidence, not proofand for a m o n e y management strategy based on dependency and correlations to prove itself m o r e profitable
than any other type of m o n e y m a n a g e m e n t regimen (or even just profitable), the
dependency needs in fact to be there. If it is not, your m o n e y m a n a g e m e n t strategy
will be suboptimal. To understand this, consider the following example: suppose
you consider trading a system that shows evidence of dependency a m o n g the outc o m e s of the historical trades. You form a hypothesis that states that the dependency is in fact there and, therefore, the system should be traded with this in mind.
But what if it turns out that the d e p e n d e n c y were not there? T h e n you will
have c o m m i t t e d a type II error (statistical jargon for accepting a hypothesis that in
fact should have been rejected) and will lose m o n e y from trading the system either
t o o aggressively or not aggressively enough. Had you done it the other way
around, rejected the hypothesis when you in fact should have accepted it (a type I
error), and traded it according to a fixed fractional m o n e y m a n a g e m e n t strategy,
although in hindsight you would have been better off trading it according to some
d e p e n d e n c y strategy, you still would have m a d e money, albeit at a slower rate.
T h u s , the penalty for m a k i n g a type II error would be a complete wipe out, while
for a type I error it would only be a slower growth of equity. T h e above reasoning
also applies to the special case of scaling back faster in times of drawdown than
what is dictated by the fixed fractional strategy.
This also relates to what we learned about the system building process and the
importance of keeping it as simple as possible, based on as few rules as possible.
This method results in a system that not only trades profitably on historical data,
but also continues to do so in the future as well. By adding rules to build away the
(presumed) dependency, we are curve fitting the system and making it less likely to

CHAPTER 15

Money Management

275

work in the future. Therefore, my recommendation is not to bother with the dependency or serial correlation at all. W h e t h e r it's there or not, build your system with as
few rules as possible that m a k e sense to you, then trade it with a fixed fractional
m o n e y m a n a g e m e n t strategy. T h e price of not doing so is simply too high.
Now, suppose that the trade in the previous example ended up in a profitable
m o v e of 5 % , equal to a profit of $5,000 p e r contract, or a total profit of $25,000
or 2 5 % . T h e following formula tells us that the H P R is 1.25:
HPR = 1 + f x ( - P F I T / WCS)
w h e r e = the value we are using for f = 0.5; P F I T = the profit or loss, per contract, on the trade, with the sign reversed = $5,000; a n d W C S = the worst-case
scenario (always a negative n u m b e r ) = $10,000.
Notice that the only difference between this and the previous formula for the
H P R is that we now multiply the percentage m a d e on the trade with the factor
and the worst-case a m o u n t is the a m o u n t risked. In the previous e x a m p l e we really didn't state how m u c h we risked or lost in dollar terms, but just stated the percentage value as a result of the formula - P F I T X 100 / W C S . To do this in Excel,
type in t h e following formulas and values into an empty spreadsheet:
In cell B 3 : 100,000
I n cell C 3 : - 1 0 , 0 0 0
In cell D 3 : 0 . 5
In cell B 5 : 5,000
Type in the following n u m b e r sequence (one below the other) into cells B6
t o B 1 4 : 2,000, 7,000, - 4 , 0 0 0 , - 2 , 0 0 0 , 6,000, 2,000, - 1 0 , 0 0 0 , - 7 , 0 0 0 and
-3,000.
In cell C 5 : = 1 + D $ 3 * ( - B 5 / C $ 3 ) (then drag down to fill all the cells C5 to
C14)
I n cell D 5 : = C 5
In cell D 6 : = D 5 * C 6 (then drag d o w n to fill all the cells D6 to D14)
I n cell E 3 : = C O U N T I F ( B 5 : B 1 4 , " < > 0 " )
In cell E 5 : = D 5 * B $ 3 (then drag d o w n to fill all the cells E5 to E l 4 )
A

I n cell F 3 : = D 1 4 ( 1 / E 3 )
I n cell F 5 : = B 3 / ( C $ 3 / - D $ 3 )
In cell F 6 : = E 5 / ( C $ 3 / - D $ 3 ) (then drag down to fill all the cells F6 to F14)
W h e n you are done, your spreadsheet will resemble that in Figure 15.1.
To the e x a m p l e above, I also have added another very useful variable, the
geometric mean, w h i c h is akin to the average H P R or growth factor per play. To
arrive at any final T W R , you could either multiply all the H P R s , or you could take
the geometric m e a n raised by the n u m b e r of trades (more about this shortly).

276

PART 5

FIGURE

Money Management and Portfolio Composition

15.1

How to use Excel to calculate your holding period returns and terminal wealth relative.

You can manually change the value in cell D3 between 0 and 1 to try to find
the highest value for T W R in cell D 1 4 . For instance, if you change the value in cell
D3 to 0 . 1 , the value in cell D 1 4 changes to 1.0889 (if y o u ' r e displaying four decimals). T h e value in cell D3 that c o r r e s p o n d to the highest value in cell D 1 4 is the
optimal f for the system. A n o t h e r a n d quicker way is to use the Solver Add-in
function in Excel, u n d e r the Tools m e n u (Figure 15.2).

FIGURE

15.2

Use the solver add-in function in Excel to calculate your optimal f.

CHAPTER 15

Money Management

277

N o t e also that in this particular case we a s s u m e d the worst case to be constant and equal to the worst case in our sequence of trades. There are several other
ways to do this, including calculating the standard deviation of all o u t c o m e s and
setting the worst trade to equal a certain distance from the average trade, or letting
the stops of your system (as they pertain to each individual trade) determine an
individual worst-case scenario for each trade. This is how I prefer to do it. Figure
15.2 also hints at the fact that you could perform m o r e complex optimizations by
adding a few other constraints, such as not letting the m a x i m u m d r a w d o w n or the
standard deviations of the individual outcomes grow past specific values.
In this case, the optimal equals approximately 0 . 3 1 , with a final T W R of
approximately 1.18. This m e a n s that for any particular trade, you should not risk
m o r e than 3 1 % of your total equity going into the trade. This is the optimal for
this system w h e n we work under no other constraints than trying to m a x i m i z e the
growth rate, not taking into consideration any costs or the fact that we cannot trade
fractions of contracts. A n d j u s t as we do not take slippage and c o m m i s s i o n into
consideration when we are looking for the optimal variable setting for a system (so
we do not c o m e up with a suboptimal solution), this is also how you should look
for your system's o p t i m a l R e m e m b e r that we first and foremost want the system
to be well w o r k i n g rather than profitablewe want it to catch a certain type of
market m o v e as efficiently as possible. As long as we do that, eventually we also
will be able to apply it to a market w h e r e it will be profitable as well. Besides,
when it c o m e s to trading with a fixed fractional m o n e y m a n a g e m e n t , seldom will
you be able to trade your system at its optimal level anyway, as will b e c o m e evident shortly.
Consider the following expectancies for a system you are thinking about
trading real-time:

T h e percentage of winning trades are over 7 2 % .

For every winning trade you win, on average, 1.52%.

For every losing trade you lose, on average, 1.19%.

T h e profit factor is 3.36.

This is a fantasy system similar to that depicted by the performance s u m m a ry in Table 12.12. With a system like this, what could go w r o n g ? T h e losers are
considerably smaller than the winners a n d with plenty m o r e w i n n e r s than losers,
at first glance it looks like you should bet almost your entire equity each time, just
saving a little in case you happen to get into a drawdown, which you should m a k e
up for easily considering the high percentage winning trades.
Because the data that is exported into Excel has all outcomes measured in percentages rather than dollars, as is the case in the example above, the calculations look
a little different. In this case, I have used the trade-by-trade export function from Part
1, which you can use for single market/system combinations. (We will learn how to
build a bar-by-bar export function for multiple market /system combinations and bar-

PART 5

278

Money Management and Portfolio Composition

by-bar portfolio analysis later.) In this case, you must insert a few empty rows at the
top of the spreadsheet and a few empty columns directly to the right of the "Profit"
column, as shown in Figure 15.3. If your spreadsheet looks like that in Figure 15.3
(remember, this is a fantasy system that you can substitute for any system of your
choice), you can then type in the following formulas or values:
I n cell F 3 :
=MIN(F5:F219)/100,
w h e r e F219 denotes the last data row.
In cell G 5 :
= 1 +G$3*(-(F5/100)/F$3) (then drag down to fill all the cells G5 to G219)
In cell H 3 :
A

= H 2 1 9 ( 1 /COUNTIF(F5 :F219," < >0"))

In cell H 5 :
=G5
In cell H 6 :
=G6*H5 (then drag down to fill all the cells H6 to H219)
T h e n manually enter into cell G3 all values for f between 0 a n d 1, in steps of
0.05, and copy each value into a separate c o l u m n . W h e n done, you should have no
problem creating a chart like that in Figure 15.4.
Sure enough, the optimal for this system turns out to be as high as 0.90,
which generates a T W R of 3,325. That is, h a d you been able to trade this system
at its optimal
level, you would have m a d e a w h o p p i n g 3,325 times your original
equity, corresponding to a total profit of equally w h o p p i n g 3 3 2 , 4 0 0 % ((3,325 - 1)
* 100). Figure 15.4 shows the T W R you would have been able to achieve depending on what level you traded.

FIGURE

15.3

Using Excel to calculate TWR as a function of f.

CHAPTER 15

Money Management

FIGURE

279

15.4

The TWR of a fantasy system in relation to the fraction of your capital risked per trade.

As y o u can see, trading at an level higher than the optimal does not m a k e
your account grow any faster. Instead you are a s s u m i n g m o r e risk than necessary
to a c c o m p l i s h the same growth rate as you could have d o n e trading at a lower
and you are also taking on a higher likelihood for going b r o k e the m o r e trades you
place. (See point t w o in the list at the beginning of this section.) B u t there is m o r e
to it than that. If you cut your risk (your f) in half, you will cut the expected profits by m u c h m o r e than that. In fact, if y o u change your arithmetically, profits
change geometrically.
T h e paradox is that the better the system a n d the higher the optimal the
larger the drawdowns you suffer. This is because the optimal is the percentage
amount you are willing to lose on one trade, b a s e d on h o w large you expect the
largest loser to be. W h e n that loser hits you, you lose that percentage a m o u n t of
your equity, as suggested by the level that you are using. A n d this is for one trade
only. If the next couple of trades turn out to be losers as well, the drawdown is even
m o r e severe than that.
This relates to points three a n d four in the list at the beginning of this chapter, because this has nothing to do with h o w m a n y markets or what type of markets y o u trade (i.e. h o w well diversified you think you are, or h o w y o u think that
you can keep a lower risk by only trading stocks and/or not selling short). Your

PART 5

280

Money Management and Portfolio Composition

worst d r a w d o w n will still be at least as large as your largest loser multiplied by the
n u m b e r of contracts it takes to assume the total risk required by the level you are
trading at. Figure 15.5 shows what the development of the T W R would have
looked like for this system, traded at the optimal
As you can see, at about trade 50 we m a d e approximately 10 times our initial equity. B u t then a series of b a d trades, w h i c h includes the largest loser, hits us,
taking the equity below the initial level. F u r t h e r m o r e , if you were to trade at the
optimal level, h o w sure could you be that this would be the m o s t optimal level in
the future as well? Because you can't, you are better off staying to the left of optimal because trading on a lower level results in the s a m e equity growth as trading at a level higher than the optimal but with a lower risk. A n d as things might
turn out, even if you a i m to trade to the left of the theoretical optimal you calculated using historical data, in real-time trading, unknowingly, you might still end
up trading to the right of the true optimal because it m i g h t have shifted to the left
as well.
F r o m this perspective, optimal is a purely a c a d e m i c figure with little practical value. B u t to go from that conclusion to reject the entire t h e o r y behind fixed
fractional trading is to take matters a little lightly. Did you, for instance, k n o w that
we have been working with several of these m o n e y m a n a g e m e n t parameters

10000

F I G U R E

15.5

The growth in equity for a fantasy system traded with an f of 0.9.

CHAPTER 15

281

Money Management

already throughout the entire b o o k ? If you have not thought about it, take a look
at the export function for the performance s u m m a r y for all systems based on the
R A D contract and any of the spreadsheets you might already have produced with
this function.
Into the individual profits c o l u m n , titled "Profit" by the export function, we
essentially have exported the unweighted H P R for each trade, except we have presented it as a percentage figure instead of a multiplier. By unweighted I m e a n that
we have not taken the largest loser and the level into consideration when calculating it, as we did, for instance, in cells C5 to C 1 4 in the e x a m p l e above. T h e s a m e
goes for the cumulative profit, in the " C u m . prof." column, which corresponds to
the T W R variable. To transform these percentage n u m b e r s (X) into multipliers (Y)
do the following:
Y = X /100 + 1
To transfer a multiplier to a percentage figure:
X = (Y - 1) * 100
Take a look at the performance s u m m a r y in Table 12.14. It shows a final
cumulative profit of 5 0 . 0 2 % , which equals a T W R of 1.5002 (50.02 / 100 + 1). To
calculate the geometric m e a n , raise it by 1 divided by the n u m b e r of trades for the
system and you will get 1.00442 (1.5002 (1 / 92)). This translates to approximately 0 . 4 4 % ((1.00442 - 1) * 100). This is the return per trade you could expect
to m a k e h a d you b e e n able to reinvest all previous profits.
We also can estimate the geometric m e a n by taking the average profit
(0.46%) and transforming it into a multiplier (1.0046). Let us call this multiplier
A P M and the estimated geometric m e a n E G M . Then raise that by 2, before we
deduct the standard deviation ( S D ) , also raised by 2. Finally, raise everything by 1
divided by 2. Hence, in this case we calculate E G M to be 1.00447 ( ( 1 . 0 0 4 6 2 0 . 0 1 6 3 2 ) ( l / 2 ) ) , or approximately 0 . 4 5 % . From here, we can continue to estimate the T W R by raising everything by the n u m b e r of trades (please bear with
m e ) . Let us call the estimated T W R ETWR and the n u m b e r of trades N. That is:
A

ETWR = EGM N, where

A

EGM = (APM 2 - S D 2 ) ( l / 2 ) , then

A

ETWR = (APM 2 - SD 2) (1/2) N = (APM 2 - SD 2) (N/2)

This final equation is so important that Ralph Vince dubbed it the fundamental equation of trading. It is The Equation, if you wish, the one that explains
everything. Perhaps not, but it does explain a lot. For instance, if the first parenthesis does not end up with a n u m b e r greater than 1, the system will not m a k e any
money. T h e only way for this to happen is for A P M to be sufficiently larger than

282

PART 5

Money Management and Portfolio Composition

SD. Of course, the larger the A P M in relation to the SD, the better, but what really matters is how sure you can be that this relationship will hold up in the future.
B e c a u s e as long as you can be sure of that, the only thing that will affect your bott o m line is h o w frequently you can trade. That is, m a k e N as large as possible. You
do that by m a k i n g sure that the system works well and/or trades profitably on as
m a n y markets as possible. A n d h o w do you do that?
O n c e again, this relates t o w h a t w e l e a r n e d a b o u t the system-building
p r o c e s s a n d the i m p o r t a n c e of k e e p i n g it as s i m p l e as p o s s i b l e , b a s e d on as few
r u l e s as possible. T h i s is the only way to c o m e close to a s y s t e m that not only
trades profitably on historical data, but c o n t i n u e s to do so in the future. By
a d d i n g rules to build away individual m a r k e t d i s c r e p a n c i e s , we c u r v e fit the
s y s t e m a n d m a k e it less likely to w o r k in the future. B u t by w o r k i n g with techn i q u e s s u c h a s r a n d o m entries a n d p e r c e n t a g e - b a s e d stops a n d exits, w e make
sure that the s y s t e m will w o r k well in any m a r k e t , in any t i m e frame, a n d almost
at a n y t i m e .
T h e fundamental equation of trading also relates to what we discussed earlier about drawdown. A l t h o u g h diversification does not necessarily reduce drawdown, a fixed fractional trading strategy will d a m p e n it by forcing you to cut back
on the n u m b e r of contracts traded, while a larger n u m b e r of markets traded will
help you to recover from drawdown at a faster rate, simply because of the additional n u m b e r of trades e a c h n e w market brings to the portfolio.

PRACTICAL APPLICATIONS FOR MONEY MANAGEMENT

All that we have discussed so far leaves us at a crossroads. Because the same system traded at different markets will c o m e up with a different optimal for each
market, it s e e m s we should trade each market individually on a separate account,
or at least trade e a c h market on a separate level of aggressiveness, as suggested by
each market/system combination's optimal Furthermore, if we blend in the
Capital Asset Pricing M o d e l ( C A P M ) and the Efficient Market Hypothesis ( E M H )
we also should weigh each market/system combination against all others so that
each market/system combination is delivering the s a m e risk-adjusted return as all
others and so that the entire portfolio of market/system combinations balances on
the
efficient frontier.
However, after we have gone to great lengths not to treat any of the markets
any differently from all the others, why should we start doing it n o w ? Because if
we did, once again we would be curve fitting. This time we w o u l d n ' t be curve fitting a specific system to a specific market, but instead a specific blend of separately weighted market/system combinations to the overall result of our portfolio.
If we decide to do this, it seriously contradicts what we have done so far. That
is, to m a k e sure that each and every one of our systems, on average and over a long
period of time, works equally well on all markets, and that we have treated each

CHAPTER 15

Money Management

283

and every market the same as all the others, basically stating that a time series is
a time series is a time s e r i e s . . . . N o t e that the key words here are "equally as well,"
not "equally as profitably." If we only wanted to optimize historical profits, we
would have curve fitted a system to a specific market, calculated the optimal
level to 0.99 or higher, then traded it real-time and g o n e broke tomorrow.
R e m e m b e r that we are not solely interested in optimizing our equity, but to optisize our betting size in accordance with a n u m b e r of constraints, a m o n g w h i c h a
desired ending equity is one.
Another key phrase in the p a r a g r a p h above is "over a long period of time."
Of course we will end up with different optimal fs for different market/systems
combinations within a limited t i m e frame, say the last 20 years or so, which is
basically how far back we have tested the systems in this book. But that is just pure
coincidence. Over the next 20 years there might be other market/system combinations with higher fs than those that seem to be doing the best right now, while the
average for the entire portfolio of all market/system combinations will still be
approximately the same.
It also is possible to do the s a m e type of logical reasoning as we did w h e n we
decided not to m a k e the amount invested dependent on any serial correlations
a m o n g the trades. In this case the hypothesis, based on historical observations, will
be, "that each market has its own specific features and statistical traits, and, therefore, should be traded at its o w n / " If this turns out to be false, then we are making
a type II error by accepting a hypothesis that in fact should have been rejected. We
will trade the system too aggressively when we shouldn't and not aggressively
enough when wc should. In the long run, we will lose m o n e y and increase the likelihood of going bust. Had we, on the other hand rejected the hypothesis when we
shouldn't have (a type I error) by trading all the market/system combinations with
the s a m e as suggested by, for instance, the lowest or the for the portfolio as a
whole, we still would have m a d e money, only at a slower rate.
To keep it simple, I therefore suggest that for each portfolio of markets to
which you apply the s a m e system, you should treat each trade the s a m e way, no
matter in which market that particular trade h a p p e n s to be, and trade it at a fixed
fractional level that m a k e s the m o s t sense to you, in regards to expected equity
growth, drawdowns, flat times, and worst-case scenarios. Exactly the s a m e goes
for a situation where you have several systems applied to one market. Finally, if
you would like to combine all your portfolios into one, continue to do the same.
If your systems are thoroughly researched and you " k n o w " they will continue to
work, you still will m a k e tons m o r e m o n e y trading at a fixed fractional level as low
0.015, not risking m o r e than 1.5% of your total equity per trade, than you would
on a fixed contract basis, depending on how you choose to define the constraints
against which you are maximizing your utility.
For instance, using the fantasy system above, if we were to risk only 3 % , the new
T W R would equal 1.4478, which means a total equity growth of a mere 4 5 % . Not too

284

PART 5

Money Management and Portfolio Composition

good, because the original system without optimal managed to produce a total of
more than 2 1 6 % . But what if instead we change the expected worst-case scenario to
equal the stop loss for the system at 1.1%, by typing " 0 . 0 1 1 " into cell F 3 . For each
new trade I put on, this is as bad as I expect things to go. Now, I know that sometimes
things will go worse than that, but that is as bad as I expect them to go. The fact is that
I know if I do this long enough there will come along one trade that will blow me away
completely. So from that point of view, whether I choose to go with the largest loser
(as suggested by the historical data) or with what I expect to lose the most in any particular trade (as suggested by the stop loss in the system) makes little difference.
D o e s n ' t it m a k e sense then to go with what I expect to lose in this particular
trade, not w h a t I might have lost in the past? Besides, if you trade the same system on a portfolio of m a r k e t s you w o u l d n ' t be able to trade it at the optimal level
anyway, as the following e x a m p l e shows.
If you trade 10 markets and it h a p p e n s that they all signal entry the same day,
you w o n ' t be able to allocate m o r e than 1 0 % of your total capital, on average, to
each market, no matter what the o p t i m a l / m i g h t be for each individual market/syst e m combination. A n d by allocating the m a x i m u m 1 0 % per m a r k e t you will go
broke if all trades go against you. N o t to risk going b r o k e you m u s t allocate less
than that, say 2% p e r market, w h i c h will give you a total risk of 2 0 % . Granted,
rarely will you enter all markets at o n c e a n d rarely will they all go against you, but
on the other hand, trading only 10 m a r k e t s isn't m u c h either.
Anyway, with the worst-case scenario set to equal the stop loss, the new optim a l / f o r this overall strategy c o m e s out to 0.10, with the relationship between different s and the T W R resembling Figure 15.6. As you can see, n o w it becomes
even m o r e important that you do not overtrade, because already at an equal to
0.15, the likelihood for going b r o k e is 100%.
By risking only 3% instead of 1 0 % of total equity I believe I have given
myself all the slack I need. In that case, the n e w T W R equals 19.44, which equals
a total equity growth of 1,844%. That is about 3 5 % per year, continuously compounded, over the last 10 years. As a comparison, the best CTAs average around
2 0 % to 2 5 % per year, while the average return from the stock market over the last
15 to 20 years is no m o r e than 1 2 % or so. Figure 15.7 shows you what the equity
g r o w t h for this strategy would look like. Now, I will not be m a k i n g 3 3 2 , 4 0 0 % , as
suggested by the original optimal but this fantasy figure describes the power of
this method. In Figure 15.7, notice especially that the drawdown around trade 50
(although the scales are different) is no worse than any of the others.
We can do a statistical type I/type II error analysis to confirm the importance
of trading at an lower than the optimal This time the hypothesis is "the true
for the future will be higher or the same as the optimal of the past; therefore we
should not use an lower than that " If we accept this hypothesis and it turns out
to be false, then we again c o m m i t a type II error. We will trade at an that is too
high, m a k i n g it likely that we will go bust sooner rather than later. H a d we instead

CHAPTER 15

285

Money Management

FIGURE

15.6

The TWR of a fantasy system using an individual worst-case scenario for each trade.

rejected the hypothesis by trading at an sufficiently lower than the historical optim a l f, w h e n it turned out that we shouldn't have, well, then we w o u l d n ' t have m a d e
as m u c h m o n e y as we could have, but we also would have increased the likelihood
of still being in the g a m e .
Next, we m u s t decide the "optisized" amount of contracts or shares we
should feasibly trade. A n d we will look at actual dollar figures instead of theoretical percentage n u m b e r s .

SHORT-TERM SYSTEMS
For each market/system combination, we have put together a spreadsheet like that
shown in Figure 15.8, w h i c h contains a daily s u m m a r y for that particular market/system combination. For e x a m p l e , Figure 15.8 shows that on D e c e m b e r 24,
1985, the system signaled to go long with a m a x i m u m risk of $578 per contract.
T h e fixed fractional m o n e y m a n a g e m e n t strategy attached to this system told us
to put on 56 contracts. On the close of that day, this market/system combination
w a s d o w n $232.50, on a one-contract basis and -$13,020 for the overall portfolio.
T h e position was closed out on D e c e m b e r 30, with a $315 profit per contract, for
a total profit of $13,440.

PART 5

286

FIGURE

Money Management and Portfolio Composition

15.7

The growth equity for a fantasy system with an individual worst-case scenario for each trade.

In Figure 15.8, the data in the five c o l u m n s under the "1 Contract from T S "
h e a d e r are imported directly from Trade Station, while the data in the three
c o l u m n s u n d e r the "N contractscalculated" are calculated in Excel. All other
cells also are imported directly from TradeStation, but can be altered within Excel
as well. To perform the necessary calculations in columns F to H, type in the following formulas (disregard that some formulas hold references to other cells, outside of Figure 15.8):
In cell F 3 6 :
0
In cell F 3 7 :
=IF(AND(C37<>C36,C37<0),H$32*H$33*MAX(INT($AN36*
( $B$47100)/C37), 1 ),IF(E37=0,F36,0))
then drag d o w n to fill all the cells in that column.
To calculate the open profit of all contracts:
In cell G 3 6 :
=ROUND(D36*F36,0)

CHAPTER 15

Money Management

FIGURE

287

15.8

Using Excel to calculate an "optisized" number of contracts for a fixed fractional trading strategy.

then drag d o w n to fill all the cells in that c o l u m n .

To calculate the closed out profit from all contracts; in cell H 3 6 :
=ROUND(IF(E36<>0,(E36-$B$3)*F35,0),0)
then drag d o w n to fill all the cells in that column.
To export t h e data needed, p u t t o g e t h e r a n o t h e r export function in
TradeStation. We use the M e a n d e r system as an e x a m p l e :
Inputs: MarketName("MaxlOChar."), SystemName("MaxlOChar."),
Direction("Both"), MarginReq(l), DollarValue(l), MarketWeight(l),
SystemWeight(l);
Vars: VSStd(2), SumVS(O), AvgVS(O), DiffVS(O), StdVS(O), SetArr(O),
SumArr(O), DiffArr(O), VSLow(O), VSMid(O), VSHigh(O), TrueEntry(O),
FName(""), TradeStr2( "), Vacuum(O), Missing(O), FillDay(O), RiskValue(O),
PosProfit(O), ClosedProfit(O), MP(0), TotTr(O);
M

Array: VS[20](0);
For SetArr = 0 To 4 Begin
VS[SetArr * 4 + 0] = (0[SetArr] Data2 - C[SetArr + 1] Data2) /
C[SetArr + 1] Data2;
VS[SetArr * 4 + 1] = (H[SetArr] Data2 - QSetArr + 1] Data2) /
C[SetArr + 1] Data2;
VS[SetArr * 4 + 2] = (L[SetArr] Data2 - C[SetArr + 1] Data2) /
C[SetArr + 1] Data2;
VS[SetArr * 4 + 3] = (C[SetArr] Data2 - C[SetArr + 1] Data2) /
C[SetArr + 1] Data2;
End;
For SumArr = 0 To 19 Begin
IfSumArr = OThen
SumVS = 0;

PART 5

288

Money Management and Portfolio Composition

SumVS = SumVS + VS[SumArr];

If SumArr = 19 Then
AvgVS = SumVS / 20;
For DiffArr = 0 To 19 Begin
If DiffArr = OThen
DiffVS = 0;
DiffVS = DiffVS + Square(VS[DiffArr] - AvgVS);
If DiffArr = 19 Then
StdVS = SquareRoot(DiffVS / 20);
End;
End;
VSLow = C Data2 * (1 + (AvgVS - StdVS * VSStd));
VSMid = C Data2 * (1 + AvgVS);
VSHigh = C Data2 * (1 + (AvgVS + StdVS * VSStd));
If MarketPosition = 0 Then Begin
Buy ("Go long") tomorrow at VSLow limit;
Sell ("Go short") tomorrow at VSHigh limit;
End;
If MarketPosition = 1 Then Begin
ExitLong ("Long Target") at EntryPrice * (1 + (2.8 * 0.01)) limit;
If Close > EntryPrice * (1 + (0.6 * 0.01)) Then
ExitLong ("Long Trailing") at EntryPrice * (1 + (0.6 * 0.01)) stop;
ExitLong ("Long Loss") at EntryPrice * (1 - (1.1 * 0.01)) stop;
End;
If MarketPosition = -1 Then Begin
ExitShort ("Short Target") at EntryPrice * (1 - (2.8 * 0.01)) limit;
If Close < EntryPrice * (1 - (0.6 * 0.01)) Then
ExitShort ("Short Trailing") at EntryPrice * (1 - (0.6 * 0.01)) stop;
ExitShort ("Short Loss") at EntryPrice * (1 + (1.1 * 0.01)) stop;
End;
If BarsSinceEntry > = 8 Then Begin
ExitLong ("Long Time") at Close;
ExitShort ("Short Time") at Close;
End;
{***** Export function starts here *****}
MP = MarketPosition;
TotTr = TotalTrades;

CHAPTER 15

Money Management

289

If CurrentBar = 1 Then Begin

FName = "D:\Temp\MM-TM-" + LeftStr(GetSymbolName, 2) + ".csv";
FileDelete(FName);
TradeStr2 = "Market" + "," + "System" + "," + "Direction" + "," +
"Margin ($)" + "," + "$ Value" + "," + "," + "Market weight (%)" +
"," + NumToStr(MarketWeight, 2) + NewLine;
FileAppend(FName, TradeStr2);
TradeStr2 = LeftStr(MarketName, 10) + "," + LeftStr(SystemName, 10) +
"," + LeftStr(Direction, 10) + "," + NumToStr(MarginReq, 0) + "," +
NumToStr(DollarValue, 0) + "," + "," + "System weight (%)" + "," +
NumToStr(SystemWeight, 2) + NewLine;
FileAppend(FName, TradeStr2);
TradeStr2 = "1 Contract fromTS" + "," + "," + "," + "," + "," +
"N contracts - calculated" + NewLine;
FileAppend(FName, TradeStr2);
TradeStr2 = "Date" + "," + "Position" + "," + "Risk ($)" + "," +
"Open ($)" + "," + "Close ($)" + "," + "Contracts" + "," + "Open ($)" +
"," + "Close ($)" + NewLine;
FileAppend(FName, TradeStr2);
End;
Vacuum = DateToJulian(Date) - DateToJulian(Date[l]);
For Missing = 2 To Vacuum Begin
FillDay = JulianToDate(DateToJulian(Date[l]) + (Missing-1));
If DayOfWeek(FillDay) > 0 and DayOfWeek(FillDay) < 6 Then Begin
TradeStr2 = NumToStr(FillDay, 0) + "," +
NumToStr(MarketPosition, 0) + "," + NumToStr(RiskValue, 2) +
"," + NumToStr(PosProfit[l], 2) + "," + "0" + NewLine;
FileAppend(FName, TradeStr2);
End;
End;
RiskValue = 0;
PosProfit = 0;
ClosedProfit = 0;
If MP <> MP[1] and MarketPosition <> 0 Then Begin
TrueEntry = C Data3 * (EntryPrice / C);
RiskValue = TrueEntry * (1.1 * 0.01) * BigPointValue;
End;
If MarketPosition = 1 Then

290

PART 5

Money Management and Portfolio Composition

PosProfit = ((C / EntryPrice) - 1) * TrueEntry * BigPointValue;

If MarketPosition = -1 Then
PosProfit = -((C / EntryPrice) - 1) * TrueEntry * BigPoint Value;
If TotTr <> TotTrfl] Then Begin
If MarketPosition(l) = 1 Then
ClosedProfit = ((ExitPrice(l) / EntryPrice(l)) - 1) * TrueEntry[l] *
BigPointValue;
If MarketPosition(l) = -1 Then
ClosedProfit = -((ExitPrice(l) / EntryPrice(l)) - 1) * TrueEntry[l] *
BigPointValue;
End;
TradeStr2 = NumToStr(Date, 0) + "," + NumToStr(MarketPosition, 0) + "," +
NumToStr(-RiskValue, 2) + "," + NumToStr(PosProfit, 2) + "," +
NumToStr(ClosedProfit, 2) + NewLine;
FileAppend(FName, TradeStr2);
We will work with dollar values now, because we would like to build hypothetical track records that can be compared to similar track records from different
CTAs, fund m a n a g e r s , and/or other system builders like ourselves.
R e m e m b e r from Part 1 that we learned we cannot m a k e any percentagebased calculations on the point-based back-adjusted contract and no dollar-based
calculations on the R A D contract. This causes a problem now, when we would like
to have our results in dollars, but still are dependent on a set of percentage-based
exit techniques to maintain them. Therefore, to transfer our system's results back
to the dollar-based world, we m u s t use some ingenuity and incorporate the unadjusted time series, where all individual front m o n t h s are attached to the time series
one after the other. In this example, we have added the daily data for the unadj u s t e d time series as data 3.
With this time series in place, we can use percentage values to calculate the
n u m b e r of points at risk and the open and closed-out profits in dollars for the unadjusted time series, and then export that value into the spreadsheet program. In the
code above you can see that this has been done in a variable called "TrueEntry,"
which is calculated at the same time as the open and closed-out profits.
A n o t h e r important feature of this code is the loop function, w h i c h makes
sure that you also export data for all days missing in your database or during holidays (except weekends). W h e n you start to put together portfolios, all the data
from all different markets m u s t be in sync.
N o w that we k n o w h o w to export the data, look again at Figure 15.8. As you
can see, row 33 holds information on which market and system this export stems,
as well as information about point value and margin requirements. This information can be typed into the spreadsheet directly, or as inputs to the code. I also have

CHAPTER 15

291

Money Management

prepared both the export function and the spreadsheet for the possibility of experimenting with different systems and market weights. To give all market/system
combinations an equal weighting keep the setting at 1, if you don't want to turn off
that particular market/system combination completely, in which case you can set
either o n e of the weights to 0.
O n c e you have typed in the necessary information and formulae in c o l u m n s
F to H, following Figure 15.8, you can either continue to add m o r e markets in adjacent c o l u m n s before you finish with the portfolio s u m m a r y calculations or, if you
only are interested in this particular market, m o v e on directly to the portfolio s u m m a r y in c o l u m n s J to Q, as shown in Figure 15.9.
To build the portfolio s u m m a r y type in the following formulas:
In cell J36:
=A36
then drag d o w n to fill all the cells in that column.
In cell K 3 6 :
=SUM(G36)
then drag down to fill all the cells in that column.
In cell L 3 6 :
=SUM(H36)
then drag down to fill all the cells in that column.
In cell M 3 6 :
=B2
I n cell M 3 7 :
=M36+L37+(K37-K36)
then drag down to fill all the cells in that c o l u m n .

FIGURE

15.9

Using Excel to calculate the aggregate results from trading several markets In a portfolio.

292

PART 5

Money Management and Portfolio Composition

In cell N 3 6 :
=ROUND(K36* 100/M3 6,2)
then drag down to fill all the cells in that c o l u m n .
I n cell 0 3 6 :
=M36
I n cell 0 3 7 :
=MAX(M37,036)
then drag down to fill all the cells in that c o l u m n .
In cell P 3 6 :
=ROUND((M36-036)* 100/036,2)
then drag down to fill all the cells in that c o l u m n .
In cell Q 3 6 :
=IF(P36=0,0,Q35+1)
then drag down to fill all the cells in that c o l u m n .
Notice especially the use of the s u m function in c o l u m n s K and L. This is
because h a d you also added data from other markets, that data would have been
a d d e d to the overall portfolio results.
In several of the above Excel formulae there are references to a set of cells
in the B column, because this is w h e r e I stored the initial p a r a m e t e r settings, like
the initial account balance, the dollar value to be deducted for slippage and commission, and the percentage risked per trade. This can be seen in Figure 15.10. In
this case, we start out with $1,000,000 in initial equity, of which we will risk
3 . 2 5 % p e r trade. Before we add the closed-out profit to the overall equity of the
portfolio, we deduct $75 p e r contract for slippage a n d commission.
Now, with the data and all the formulas in place it is t i m e to build a performance s u m m a r y that resembles that in Figure 15.11. In this particular case it
shows the results from trading the M e a n d e r system on the S & P 500, over the period M a r c h 1996 to October 1999, with $75 deducted for slippage a n d commission
per contract traded. T h e fixed fraction of the total equity risked per trade was set
to 5% and the initial capital w a s $1,000,000.
As you can see, in this particular case we took the initial capital to an ending
capital of $ 3 , 2 0 3 , 9 7 1 , w h i c h equals a yearly percentage return of approximately
3 6 . 4 % . T h e " p r i c e " for doing this c a m e in the form of having to spend approximately 16 out of every 100 days in the market, a m a x i m u m drawdown of close to
2 8 % , and a longest flat t i m e of 317 days.
Before we go on, let us take a closer look at h o w each performance figure is
calculated and what you need to type into each cell.

CHAPTER 15

293

Money Management

FIGURE

15.10

Example of initial parameter setting for all Excel calculations.

In cell E 3 :
=ROUND(AN(X),0),
w h e r e column AN refers to the column for w h e r e the portfolio equity is calculated ( c o l u m n M, in Figure 15.9, above) and Xrefers to the last row of data.
In cell E 4 :
=ROUND((E3/B2 - 1 ) * 100,0)
In cell E 5 :
A

=ROUND(((E3/B2) (l/B5)-1)* 100,2)

FIGURE

15.11

Strategy summary for a portfolio during the "optisizing" process in Excel.

PART 5

294

Money Management and Portfolio Composition

In cell E 6 :
,,

,,

=ROUND(SUM(SUMIF(H36:H989,">0"),SUMIF(Q36:Q989, >0 ),
SUMIF(Z36:Z989,">0"),SUMIF(AI36:AI989,">0"))/ABS(SUM(SUMIF
(H36:H989; <=0 ),SUMIF(Q36:Q989, <=0"),SUMlF(Z36:Z989,"<=0 ),
SUMIF(AI36:AI989,"<=0"))),2),
,

,,

where columns H, Q, Z, and AI refers to the closed out profit for each market/ syst e m in the portfolio (in this case I p r e p a r e d the fixed fractional portfolio calculations for four markets/systems, but temporarily turned off the other three).
In cell E 9 :
=ROUND(MIN(AQ36: AQ989),2),
w h e r e c o l u m n AQ refers to the portfolio d r a w d o w n ( c o l u m n P, in Figure 15.9,
above).
I n cell E l 0 :
=ROUND(MIN(AM36:AM989),0),
w h e r e c o l u m n AM refers to the closed out equity for the portfolio (column L, in
Figure 15.9, above).
I n cell E l l :
M

,,

=ROU]S0^(SUM(COUNTIF(H36:H989, >0"),COUNTIF(Q36:Q989, >0 ),

COmTIF(Z36:Z989,*>0'0,COUNTIF(AI36:AI989,">0 ))*100/H3,2)
,,

In cell H 3 :
,,

,,

=SUM(COUNTIF(H36:H989, <>0"),COTJNTIF(Q36:Q989,"<>0 ),
M

COUNTIF(Z36:Z989,"<>0"),COUNTIF(AI36:AI989, <>0"))
In cell H 4 :
=ROUND(SUM(SUM(H36:H989),SUM(Q36:Q989),SUM(Z36:Z989),
SUM(AI36:AI989))/H3,0)
In cell H 5 :
=ROUND(H3/B5/B7,l)
In cell H 6 :
=ROUND(H5/12,l)
In cell H 9 :
=MAX(AR36:AR989),
w h e r e c o l u m n AR refers to the flat time for the portfolio ( c o l u m n Q, in Figure
15.9, above).

CHAPTER 15

295

Money Management

In cell H 1 0 :
=COUNTIF(AL36:AL989,"<>0")*100/COUNT(AL36:AL989),
w h e r e column AL refers to the total open equity for all open trades in the portfolio (column K, in Figure 15.9, above).
In cell HI 1:
M

=ROUND(COUNTIF(AL36:AL989, <>0")/COUNTIF(AM36:AM989,">0 ),0)

With all the formulae in place, it is easy to put together a set of charts like
those in Figure 15.12 through 15.14, which show the total equity, the drawdown,
and the open profit, respectively. Figure 15.12 shows the equity curve for a portfolio consisting of only one market/system combination. It corresponds to the performance s u m m a r y in Figure 15.11. Because of the very long initial flat period, it
is not difficult to see that this equity curve is no good, despite the fact that the overall growth rate is as high as 3 6 . 5 % p e r year. At this point, it also is important to
understand that the percentage moves of the market are not the same as the percentage increase or decrease of our account. That the system will have us stopped
out if the market moves against the position by 1.1%, is not the s a m e as if we are
risking only 1.1% of our capital at that trade. No matter h o w m u c h the market

FIGURE

15.12

The equity curve from trading Meander on the S&P 500 with a 5% fixed fractional setting.

296

PART 5

FIGURE

Money Management and Portfolio Composition

15.13

The drawdown curve from trading Meander on the S&P 500 with a 5% fixed fractional setting.

moves, it is still possible to lose m o r e or less than that, in percentage t e r m s . In this

case, we m a k e sure that we have e n o u g h m o n e y going into the trades so that we
can afford to lose 5% of it if the market moves against us by 1.1%.
T h e dangers of trading this market/system combination alone are confirmed
by Figure 15.13, which shows the different drawdown periods. However, when
looking at the open equity of all trades at each day, in Figure 15.14, we can see that
one m a i n reason for this long flat period is not that the system is performing badly,
but instead that it doesn't trade often enough; we have to suffer the pain of one bad
trade for an exceptionally long time. To shorten this period of torture, we must
diversify by trading the s a m e system on other m a r k e t s and/or apply other uncorr e c t e d systems to the s a m e market.
By the way, Figure 15.14 fills an important function in that it allows you to
keep track of the total open equity from all open positions in relation to the total
equity (both closed-out and open) of the entire portfolio and compare it to historical values. Positive values m e a n that the profit is positive, while negative values
obviously indicate the opposite. If the current open profit gets out of line when
compared to what is expected, you k n o w that one or several market/system combinations have developed in such a way that you n o w are steering the entire portfolio
into unfamiliar territory. In this case, very seldom will the profit exceed 10%, or

CHAPTER 15

297

Money Management

FIGURE

15.14

Profits per trade for Meander traded on the S&P 500 with a 5% fixed fractional setting.

surpass a loss of 5 % (because that should have us closed out). Therefore, when
this happens, you know this is an exceptional occurrence. A n d because exceptional occurrences, whether for better or worse, add m o r e to the insecurity and risk than
to the bottom line, you might be better off taking a profit or cutting the loss, overriding the signal from the system for the best of the portfolio as a whole.
As already concluded, the performance s u m m a r y in Figure 15.11 and the
development of the total equity as depicted by Figures 15.12 and 15.13 are not all
that impressive. But this portfolio consisted of only o n e market/system combination. To better our results we need to diversify into other market/system combinations. Table 15.1 shows the result of one such diversification. This time the portfolio consists of three different market/system combinations, the three short-term
systems we built in the previous section of this book, all applied to the S & P 500
c o m m o d i t y futures contract, with the m a x i m u m percentage risked per trade set to
only 1.5%. T h e initial capital was set to $1,000,000, with $75 deducted p e r contract traded. Notice that we n o w are m a k i n g over 5 0 % p e r year, with a largest
d r a w d o w n of less that 2 0 % a n d a longest flat time of less than 6 months ( 1 1 3 / 2 1
= 5.4). A n d we are only risking 1.5% p e r trade!
Figures 15.15 and 15.16 show the a c c o m p a n y i n g equity curve a n d historical
drawdown for this short-term strategy. As you can see, although from the total

298

PART 5

TABLE

Money Management and Portfolio Composition

15.1

1.5% fixed fractional portfolio performance summary for three systems trading S&P 500.
Strategy summaries
Profitability
End Eq ($)

Trade statistics
4,827,166

Total (%)

383

Year(%)

52.17

P factor

1.56

No Trades

316

Avg Tr ($)

12,773

T r / Mark /Year
Tr / Month

Risk measurers

28.1
2.3

Time statistics

Max DD (%)

-19.47

Lrg Loss ($)

-307,050

Winners (%)

56.96

Lng Flat (d)

TIM (%)
Avg Days

113
54.82
3.00

equity chart it looks like the worst d r a w d o w n in dollar t e r m s h a p p e n e d sometime

during the s u m m e r of 1999, the worst drawdown in percentages h a p p e n e d late in
1998. Providing the overall strategy is robust, and rests on sound and simple principles, as the equity continues to grow, the worst drawdown, in dollar terms, will

F I G U R E

15.15

The combined equity curve from trading three different market/system combinations.

CHAPTER 15

299

Money Management

FIGURE

15.16

The combined drawdowns from trading three different market/system combinations.

continue to grow as well, while the drawdown as a percentage of your equity is

likely to stay the same. If it h a p p e n s to increase, this will have nothing to do with
the level the m a r k e t currently is trading, if the m a r k e t trend is up or down, the size
of your account, or any other reason. T h i s is not the case w h e n using any of the
commercially available system testing p a c k a g e s . With the technique illustrated
here, you k n o w that the only reason why you suddenly get a higher drawdown figure is that your strategy is no longer working as it should (or at least is behaving
in an unexpected m a n n e r ) . If and w h e n this h a p p e n s , y o u no longer need to worry,
because having r e a d the previous sections, you k n o w exactly h o w to go back and
correct what is w r o n g . You m a y even c h o o s e to build a new system, especially suited to trade with a portfolio of other systems and markets, all of w h i c h are develo p e d solely to work well together u n d e r a fixed fractional m o n e y m a n a g e m e n t regimen, thus forming an overall trading strategy w h e r e the whole is greater than the
s u m of its parts.
W i t h the above paragraph in mind, look at Figures 15.17 a n d 15.18, w h i c h
show two different ways of m e a s u r i n g the performance of the individual trades.
C o m p a r e t h e m to Figures 2.4 and 2 . 5 . In Figure 2.5 and 15.18 there seem to be an
u p p e r and lower limit to h o w big each bar is likely to b e . This is because in both
charts the m a g n i t u d e of the bars is expressed in percentage t e r m s . In Figure 2 . 5 ,

PART 5

300

FIGURE

Money Management and Portfolio Composition

15.17

The dollar value for each trade from three different market/system combinations.

the percentage is calculated as the value of a specific m o v e in relation to where the

market currently is trading. In Figure 15.18, it is calculated as the value of a specific trade in relation to the total equity of the portfolio.
Figures 2.4 a n d 15.17, on the other hand, do not use these relative comparisons, but chart the dollar values directly. In Figure 2.4, we are looking at the dollar value from trading each m o v e with one contract. In Figure 15.17, we are looking at the dollar value from trading each m o v e with an increasing n u m b e r of contracts. T h e main difference between Figures 2.4 and 15.17 is that in Figure 2.4 the
dollar value per trade is increasing because the market is trading higher, while in
Figure 15.17 the dollar value is increasing because our equity is increasing.
However, whether our equity is increasing or not has nothing to do with whether
the market(s) are trending up, down, or sideways, but rather with the fact that our
strategy is working as it should, catching whatever type of trend or m o v e happens.
We know this now, but there was no way of knowing this from just looking at
Figure 2.4, w h e n we started out putting our strategies together.
Although Figure 15.17 can be fun to look at as long as everything is going
the right way, the most important chart to keep track of is Figure 15.18, for exactly the s a m e reasons as those for Figure 15.14, above. T h a t is, if the closed-out profits seem to get out of line w h e n c o m p a r e d to w h a t could be expected, you know

CHAPTER 15

301

Money Management

FIGURE

15.18

The percentage return for each trade from three different market/system combinations.

that one or several market/system combinations have developed in such a way that
you n o w are steering the entire portfolio into unfamiliar territory. In this case, very
seldom will the closed-out losses be any worse than j u s t a few percentage points.
Therefore, if this starts to h a p p e n m o r e and m o r e often, you k n o w that o n e or several of your market/system combinations have started to behave in ways not in line
w i t h w h a t could be expected and perhaps even against the logic of the system.
W h e n working with the daily and bar-by-bar data there is one m o r e important
chart to consider. That is the margin to equity ratio, which is shown in Figure 15.19.
This chart shows how m u c h of your total equity you actually need to keep on account
with your broker to do the necessary trades. In the case of the S & P 500 portfolio we
have been working with so far, I have assumed a margin of $22,000 per contract. In
this case, as time moves on, the average margin to equity ratio should probably even
out just below 2 0 % , with an expected m a x i m u m around 3 5 % . This m e a n s that of the
total equity we have at stake, we only need to keep, say 5 0 % just in case, on account
with our broker. T h e rest of it we can have elsewhere, where we h o p e it can generate a higher return than the interest rate our broker can offer. It is important to realize, though, that you m u s t have this money readily available and committed to your
strategy as well. Otherwise, it would be as if you were trading your portfolio with
only half the equity, but twice the a recipe for disaster, indeed.

302

PART 5

FIGURE

Money Management and Portfolio Composition

15.19

The margin to equity ratio for a portfolio.

Before we m o v e on to the long-term systems, let us compare the performance summaries in Figure 15.11 and Table 15.1 with Tables 15.2 and 15.3 which
show the results from trading both these portfolios on a one-contract basis. If a
fixed fractional trading strategy is not doing any better than a one-contract, then
what is the point? But we do not have to worry. Table 15.2 shows that the single
market/system portfolio would only have m a d e $88,500 on a one-contract basis,
while the three-market/system portfolio would have m a d e $331,000, or approximately 8%> p e r year on an initial equity of $1,000,000. T h a t is quite a difference
c o m p a r e d to the 5 2 % per year we could have m a d e only risking 1.5% of total equity p e r trade.
O t h e r interesting o b s e r v a t i o n s are that the profit factor, the value of the
average trade, the p e r c e n t a g e profitable trades, a n d the d r a w d o w n all are better
in F i g u r e 15.11 a n d Table 15.2 than they are in Tables 15.1 and 15.3. For the
average s y s t e m builder, these are all i m p o r t a n t figures. H a d we not k n o w n what
we n o w know, c h a n c e s are that we m i g h t h a v e ditched the three-market/system
portfolio for t h e b e n e f i t of the s i n g l e - m a r k e t / s y s t e m c o m b i n a t i o n , a n d then traded that o n e m o r e aggressively instead, if we h a d derived these n u m b e r s directly
from T r a d e S t a t i o n ' s or any other c o m m e r c i a l l y available p r o g r a m ' s performance
summary.

CHAPTER 15

Money Management

TABLE

15.2

Single contract performance for the Meander system.

Strategy summaries
Profitability
End Eq ($)

Trade statistics
1,088,541

No Trades

73

AvgTr($)

Total (%)

Year (%)

2.29

Tr/Mark/Year

6.5

P factor

1.86

Tr / Month

0.5

Risk measurers

1,213

Time statistics

Max DD (%)

-1.68

Lrg Loss ($)

-13,148

Winners (%)

57.53

Lng Flat (d)

275

TIM (%)

16.35

Avg Days

4.00

There are several reasons w h y a portfolio of several market/systems combinations does better than only a single market/sy stem combination, even if we try
to c o m p e n s a t e by trading the sole combination m o r e aggressively. The m o s t obvious reason, as we learned from the fundamental equation of trading, is that the best
way to speed up the equity growth rate for a trading strategy with a positive
expectancy is to trade it as frequently as possible. A n d no matter h o w aggressively we trade a single market/system combination, the actual n u m b e r of trades will
not increase, as it will if we trade several market/system combinations simultaneously. Furthermore, by letting two or several market/system combinations with
positive expectancies share the s a m e account, there is a positive expectancy that
there will be m o r e m o n e y to play with, going into each individual trade.

TABLE

15.3

Single contract portfolio performance summary.

Strategy summaries
Profitability
End Eq ($)

Trade statistics
1,331,013

Total (%)

33

No Trades

316

Avg Tr($)

1,079

Year(%)

7.92

Tr/Mark/Year

P factor

1.59

Tr/Month

Risk measurers

28.1
2.3

Time statistics

Max DD (%)

-3.85

Lrg Loss ($)

-32,656

Winners (%)

56.96

Lng Flat (d)

TIM (%)
Avg Days

112
54.82
3.00

304

PART 5

Money Management and Portfolio Composition

This is true as long as the different market/system combinations are not perfectly (100%) correlated and with each one having a positive expectancy. It could
also be true even if one or several of the market/system combinations have a negative expectancy, as long as they are uncorrelated e n o u g h with the rest of the combinations and the portfolio as a w h o l e . T h e r e is a price for doing this, however, and
that is that the overall efficiency of the portfolio diminishes the m o r e market/syst e m combinations you add. For instance, trading two market/system combinations
that are not perfectly correlated, with an optimal of 0.5 each, results in an optim a l of less than 0.5 for the strategy as a w h o l e . This m e a n s that there also is a
limit to h o w m a n y market/system combinations it is meaningful to trade.
But, to take the reasoning through yet another circle, the decrease in efficiency as a result of trading several uncorrelated market/system combinations also
has a major benefit. T h e lower the optimal or the fixed fractional trading level,
the lower the risk and the shallower the expected d r a w d o w n s . Consider, for
instance, two perfectly correlated market/system combinations. W h e n o n e of them
has a losing trade, so has the other, w h i c h m e a n s that all losing trades leading into
a new drawdown from a recent equity high hit you full force, w h e r e a s a drawdown
for an uncorrelated portfolio of market/system combinations is likely to be dampened because the combinations are low correlated, but also because they don't
trade at the s a m e time. This m e a n s that each losing trade decreases the n u m b e r of
contracts that can be traded going into each consecutive trade.
All the above leads to one important conclusion: the optimal is not a measure of how high a certain portfolio's return is likely to be, but instead of how risky
it is. To get an indication of h o w high the return will be, we must look at the estim a t e d growth rate, as measured by the estimated geometric mean ( E G M ) , given the
a s s u m e d risk as dictated by the largest expected amount to lose p e r contract and the
largest expected amount to lose per trade (the fixed fractional trading level).
Finally, just for the h e c k if it, it could be fun to k n o w h o w m u c h we could
have m a d e h a d we traded the three-market/system portfolio at or near its optimal
Table 15.4 and Figure 15.20 reveal that we would have taken the initial million
to $114,148,000, w h i c h corresponds to an average percentage return of 2 5 4 % per
year. However, the price for this would have been to suffer through a drawdown of
m o r e than 7 5 % of account equity a n d a largest single loss of close to $37,000,000.
(By the way, w h i c h one of all the drawdowns in Figure 15.20 do you think is the
worst, as a percentage of total equity?) T h e s e are clearly not tolerable or tradable
n u m b e r s , w h i c h also is illustrated by the fact that the m a x i m u m n u m b e r of contracts traded at o n e t i m e would have b e e n m o r e than 10,000 and that the average
margin-to-equity ratio would have been close to 2 0 0 % . We would have h a d to borrow the s a m e a m o u n t as we h a d on the account, j u s t to cover margin. A n d who in
his right m i n d would have lent us that k i n d of m o n e y w i t h a track record showing
d r a w d o w n s upwards of 8 0 % ? N o , although the theory surrounding optimal is
beautiful, to trade at an unconstrained optimal level simply is not feasible. (The

CHAPTER 15

305

Money Management

TABLE

15.4

7% fixed fractional (optiomal f) performance summary for a three market/system combination.

Strategy summaries
Trade statistics

Profitability
114,147,973

No Trades

316

Total (%)

11,315

Avg Tr ($)

451,612

Year(%)

253.72

Tr/Mark/Year

P factor

1.33

End Eq ($)

Tr / Month

28.1
2.3

Time statistics

Risk measurers
Max DD (%)

-75.67

Lrg Loss ($)

-36,724,875

Winners (%)

56.96

Lng Flat (d)

TIM (%)
Avg Days

160
54.93
3.00

worst d r a w d o w n as a percentage of total equity h a p p e n e d during the fall of 1997,

and shows up as a little blip about halfway t h r o u g h the time series in Figure 15.20.)

THE LONG-TERM STRATEGIES

W h e n it c o m e s to fixed fractional investing, the long-term systems have one major
disadvantage c o m p a r e d to their short-term siblings: their lower frequency of
trades, which, as we have learned from the fundamental equation of trading, is a
key factor in the success of a fixed fractional trading strategy. To overcome this
w e , therefore should use the system together with as m a n y markets as possible to
speed up the trading frequency. However, in doing so it is vital that the system be
robust a n d likely to trade profitably on as m a n y markets as possible. As we have
seen, to achieve this the system m u s t work the same, no matter what the current
m a r k e t conditions are like or what market we are trading. To be consistent, this
also m e a n s that we a s s u m e no significant statistical differences between the m a r kets and that they are all treated identically and traded with the same even though
there might be significant differences between the fs w h e n we c o m p a r e different
market/sy stem combination with each other.

The SDB Strategy

During the entire testing procedure for the S D B system we only used data up until
October 1992. T h e rest of the data we saved for some out-of-sample testing. Table
15.5 shows the result of trading the entire portfolio on the in-sample data, using the
same fixed fractional m o n e y m a n a g e m e n t for all markets, a n d risking no more than
1.5% of total portfolio equity p e r trade, going into the trade. T h e starting balance

PART 5

306

F I G U R E

Money Management and Portfolio Composition

15.20

The equity curve from trading three different market/system combinations at their optimal f.

w a s set to $1,000,000, with $75 deducted for slippage and c o m m i s s i o n per contract
traded.
Just to m a k e sure that it is indeed beneficial to use a fixed fractional trading
strategy, even if we keep t h e / a s low as 0.015, you can c o m p a r e the results in Table

TABLE

15.5

1.5% fixed fractional trading with the SDB strategy (in-sample period).
Strategy summaries
Trade statistics

Profitability
End Eq ($)

7,462,552

424

No Trades

14,614

646

AvgTr($)

Year(%)

17.6

Tr/Mark/Year

2.1

P factor

2.04

Tr / Month

2.8

Total (%)

Time statistics

Risk measurers
Max DD'(%)

-17.17

Lrg Loss ($)

-148,124

Winners (%)

41.75

Lng Flat (m)

25.9

TIM (%)

99.81

Avg Days

20.00

CHAPTER 15

307

Money Management

15.5 with those in Table 15.6, w h i c h shows how m u c h you would have m a d e trading o n e contract only. T h e only major drawback with the statistics in Table 15.5 is
a little too long flat period. However, research and experimentation show that it is
m o r e a result of b a d portfolio composition than of the system itself, although at
this stage all three variablesthe system, the m o n e y m a n a g e m e n t , and the portfolio compositioninterfere and influence each other, m a k i n g all three parts
equally responsible for results. Nonetheless, in this case the longest flat period
r e m a i n s the s a m e even if is increased to 2 . 5 % or lowered to 1%. T h e only thing
that h a p p e n s is that with a higher c o m e s a higher drawdown and a higher return,
and vice versa.
By the way, t h e o p t i m a l for this portfolio can be found a r o u n d 1 4 % ,
w h i c h p r o d u c e s a final profit of m o r e than $ 2 1 . 5 billion, c o r r e s p o n d i n g to a
yearly rate of return close to 1 2 5 % , b u t also a d r a w d o w n close to 9 5 % , a largest
loss close to $4.5 billion, a n d a l o n g e s t flat t i m e of m o r e than 40 m o n t h s . As
s h o u l d be clear by now, the final r e t u r n is only a function of h o w well you
w o u l d like to sleep at night. In this case we w o u l d be m a k i n g s o m e t h i n g like
$ 1 . 7 5 billion a year, on a v e r a g e i f y o u believe that is w o r t h sleepless n i g h t s ,
b e m y guest.
N o w that we k n o w the S D B system works fine within a m o r e complete trading strategy incorporating consistent m o n e y m a n a g e m e n t on a whole portfolio of
market/system combinations, perhaps it is time to look at how the same strategy
would have performed on previously unseen data. T h a t we will do, but first we will
add another few analysis techniques to our portfolio s u m m a r y spreadsheet.
We will put together a set of formulae that produce monthly results, resembling those in Figure 15.21, w h i c h we will use for further input into a new set of
formulae that produce results like those in Tables 15.7 and 15.8.

TABLE

15.6

Single-contract trading with the SDB strategy (in-sample period).

Strategy summaries
Trade statistics

Profitability
End Eq ($)

1,329,060

No Trades

424

AvgTr($)

765

Total (%)

33

Year(%)

2.32

Tr/ Mark /Year

2.1

P factor

1.99

Tr / Month

2.8
Time statistics

Risk measurers
Max DD (%)

-2.53

Lrg Loss ($)

-6,726

Winners (%)

41.75

Lng Flat (m)

20.81

TIM (%)

99.81

Avg Days

20.00

308

PART 5

FIGURE

Money Management and Portfolio Composition

15.21

Using Excel to calculate monthly returns.

To start, highlight the cell in the column, o n e c o l u m n away from your portfolio calculations a n d at the s a m e row for w h e r e you start the portfolio calculations. In Figure 15.21, that is cell E Z 3 0 .
In cell E Z 3 0 :
=EO30
w h e r e column EO denotes the daily updated dates.
In c e l l E Z 3 1 :
=IF(INT(E031/100)<INT(EO32/l 00),EO31,"")
then drag down to fill all the cells in that c o l u m n .
In cell FA30:
=ER30,
w h e r e column ER the daily updated total equity.
In c e l l F A 3 1 :
=IF(INT(EO31/100)<INT(EO32/100),ER31,"")

TABLE

15.7

Cumulative percentage period returns, n months rolling window (in-sample period).

Cumulative

12

24

36

48

60
159.88

-0.94

3.95

26.37

39.14

54.52

79.46

119.78

1.43

4.41

8.77

17.65

39.97

69.48

110.03

159.17

Best

13.25

27.13

49.21

62.70

116.21

162.76

228.11

256.53

Worst

25.23

Most recent
Average

-9.09

-12.34

-10.97

-6.54

-11.42

-1.67

8.71

St. dev.

3.92

7.21

11.26

15.42

29.05

44.40

60.53

69.04

EGM

1.35

4.16

8.19

16.64

36.92

63.56

101.12

149.81

Sharpe ratio

0.36

0.61

0.78

1.14

1.38

1.56

1.82

2.31

Avg. winning

3.57

7.48

12.17

20.80

44.63

70.11

110.03

159.17

-2.33

-3.52

-3.63

-3.35

-4.26

-1.67

N/A

N/A

Avg. losing

CHAPTER 15

309

Money Management

TABLE

15.8

Annualized percentage period returns, n months rolling window (in-sample period).

Annualized
Most recent
Average

12

24

36

48

60

-10.71

16.76

59.69

39.14

24.31

21.52

21.76

21.05

18.58

18.84

18.31

17.65

18.31

19.23

20.38

20.98
28.95

Best

345.1

161.21

122.64

62.7

47.04

37.99

34.59

Worst

-68.13

-40.95

-20.74

-6.54

-5.88

-0.56

2.11

4.6

58.63

32.11

23.79

15.42

13.6

13.03

12.56

11.07

St. dev.
Sharpe
% winners

0.32

0.59

0.77

1.14

1.35

1.48

1.62

63.76

72.11

78.47

86.96

90.48

99.12

100.00

1.9
100.00

then drag down to fill all the cells in that column.

You m u s t delete all the blank cells in columns EZ and FA and line up all the
other cells containing information underneath each other so that the end result will
look like Figure 15.21. This can be done either manually or with the help of a macro.
W h e n you are done, continue to calculate the latest equity high and drawdown.
In cell F B 3 0 :
=FA30
In cell F B 3 1 :
=MAX(FA31,FB30)
then drag d o w n to fill all the cells in that c o l u m n .
In cell F C 3 0 :
=(FA30-FB30)*100/FB30
then drag down to fill all the cells in that c o l u m n .
In c o l u m n s FD and F E , we calculate the rolling monthly and quarterly
returns. In cell F D 3 1 :
=(FA31/FA30-1)*100
then drag down to fill all the cells in that column.
In cell F E 3 3 :
=(FA33/FA30-1)*100
then drag down to fill all the cells in that column.
In columns FF to FK (not shown in Figure 15.21) continue to calculate the
rolling 6-, 12-, 24-, 36-, 4 8 - , and 60-month rolling returns. With the above formulae
and all other formulae we have used throughout the book, it should be no problem
to go on and produce a table like Table 15.7.

310

PART 5

Money Management and Portfolio Composition

To transform the cumulatively c o m p o u n d e d values in Table 15.7 to annualized dittos in Table 15.8, simply raise each value by (12/PL), w h e r e PL equals the
length of the period. For instance, if the m o s t recent monthly return is stored in cell
K 3 , to transfer it into an annualized figure in cell K 1 3 , use the following formula
i n cell K l 3 :
A

=ROUND(((K3/l 00 +1 ) ( 12/K2) -1 )* 100,2),

w h e r e cell K2 denotes the length in m o n t h s of the original period.
F r o m Tables 15.7 and 15.8 we can, for instance, see that during no 12-month
period did we lose any m o r e than 6 . 5 4 % and never did we produce a losing period lasting longer than 3 years. We also can see that although the percentage of
profitable trades w a s less than 4 2 % , the percentage of profitable m o n t h s was as
high a s 6 4 % .
This is such an important finding that it m u s t be explained further for a deeper understanding. Because we have constructed the system to cut our losses short,
using such techniques as time-based stops we accomplish two things. First, a single loss w o n ' t go on forever, weighting down the results for several m o n t h s in a row.
Second, other trades to c o m e will have an easier time m a k i n g up for the recently
produced loss. In the end, this might m e a n that a lower percentage of profitable
trades will produce a higher percentage of profitable months. O n c e you have taken
your analysis to this point, I think you will agree when I say that it is far more
important to have a high percentage of profitable m o n t h s than it is to have profitable trades. To m a k e sure that the n u m b e r of winning m o n t h s will be high, you
must have taken that into consideration at step one of your system-building process.
This is the s a m e type of trade-off we m u s t do w h e n we compare the average
trade to the standard deviation. T h e trick is not to have as high as possible an avera g e trade, but only to m a k e sure that it is sufficiently larger than the standard deviation; every change we m a k e to the system that decreases the average trade is all
right, as long it decreases the standard deviation even m o r e . (Within reason that
i s t h e mathematical expectancy still must be positive and the average trade large
e n o u g h to m a k e it worthwhile trading the system.) T h e positive side effect from all
this is that we also are likely to increase the n u m b e r of trades, which boosts our
profits even further according to the fundamental equation of trading.
A n o t h e r way to illustrate several of the n u m b e r s in Tables 15.7 and 15.8 is to
create charts like those in Figures 15.22 and 15.23.
Let's take a l o o k at h o w the S D B strategy w o u l d have fared both during the
i n - s a m p l e p e r i o d from J a n u a r y 1980 to O c t o b e r 1992, a n d the out-of-sample
period from O c t o b e r 1992 to O c t o b e r 1999. As before, the starting equity was
set to $ 1 , 0 0 0 , 0 0 0 , a n d $75 w a s d e d u c t e d for slippage a n d c o m m i s s i o n p e r contract traded. Table 15.9 reveals that the e n d i n g equity n o w w o u l d have been close
to $ 2 0 , 0 0 0 , 0 0 0 , c o r r e s p o n d i n g to a yearly return of 1 6 . 6 8 % . T h e fact that this is
slightly lower t h a n for the i n - s a m p l e period alone indicates that the strategy has

CHAPTER 15

311

Money Management

FIGURE

15.22

A graphical presentation of cumulative monthly returns (in-sample period).

TABLE

15.9

1.5% fixed fractional trading with the SDB strategy (both sample periods).
Strategy summaries
Trade statistics

Profitability
End Eq ($)

19,949,789

730

No Trades

25,184

Total (%)

1,895

AvgTr($)

Year(%)

16.68

Tr/Mark/Year

2.4

P factor

1.52

T r / Month

3.1
Time statistics

Risk measurers
Max DD (%)

-27.3

Lrg Loss ($)

-568,453

Winners (%)

39.59

Lng Flat (m)

25.9

TIM (%)

99.88

Avg Days

19.00

312

PART 5

FIGURE

Money Management and Portfolio Composition

15.23

A graphical presentation of annualized monthly returns (in-sample period).

not w o r k e d as well during recent years. T h i s is further c o n f i r m e d by a lower

profit factor, a larger m a x i m u m d r a w d o w n , and a lower p e r c e n t a g e profitable
trades. As you can see, the average trade m e a s u r e d in dollars is m u c h higher in
Table 15.9 than in Table 15.5, but, as s h o u l d be clear by now, this has n o t h i n g to
do with t h e strategy's p e r f o r m i n g any better, b u t simply b e c a u s e the available
equity is so m u c h h i g h e r w e can have m o r e m o n e y at stake at e a c h individual
trade.
From Table 15.10 we also can see that the average trade, measured in percentage terms, has decreased while at the same time the standard deviation has
increased. This is not a good sign, because it implies that not only has the strategy
turned less profitable, but also it has turned m o r e risky. T h e percentage of winning
m o n t h s also has decreased somewhat. All in all this is a decrease in performance,
but still not too far away from what could be considered tolerable for a professional m o n e y manager. In fact, the only thing that most definitely is not tolerable is the
longest flat time, although it has stayed intact during the out-of-sample period.
Figures 15.24 and 15.25 give you essentially the same type of information as Tables
15.7 and 15.8, and figures 15.22 and 15.23.

CHAPTER 15

313

Money Management

TABLE

15.10

Monthly returns for the SDB strategy (both sample periods).

Most recent
Average
Best
Worst

-7.12
1.40
15.58
-13.63

St. dev.

4.70

Geo Mean

1.29

Sharpe ratio (Annualized)

Avg. winning

0.3
4.18

Avg. losing

-2.86

% winners

60.52

The Directional Slope Strategy

W h e n we built the Directional Slope System, we used weekly data to filter out some
of the daily noise. Unfortunately, to be able to track the system in real time, we now
need to transfer it back to a daily strategy by multiplying all the input variables by
five. My initial research shows that this inevitably lowers the results somewhat, but
so far I have found no proof for ditching this system-building concept altogether,
provided that original system logic has been sound and simple and that every effort
has been m a d e to keep the system as robust as possible. I must point out however,
that my research in this area is still at its initial stages and that only time will prove
me right or wrong. Nevertheless, the point here is not to give you the very best systems around, but simply to outline a few tips and tricks on h o w to go about building
one yourself, orperhaps even more importantto describe the underlying reasoning that you must consider before you sit down in front of the computer.
Table 15.11, and Figures 15.26 a n d 15.27 show the result of trading the directional slope system on all 16 markets used during the system-building process. As
before, we started out with $1,000,000 in initial equity, deducting $75 for slippage
and c o m m i s s i o n per contract traded. As y o u can see, these results aren't too tantalizing, mainly because of a largest d r a w d o w n closing in on 6 0 % and a longest flat
period of m o r e than 28 m o n t h s .
However, one major reason for this is that, during the optimization process,
I deliberately used markets that I knew from experience would not sit well with a
trend-following strategy, such as the S & P 500 index and the C R B index. I did this
to pollute the final parameter setting with as m a n y market conditions as possible
to increase the likelihood for the m o d e l to hold up in the future. Inevitably there
will c o m e a day w h e n such trend-following workhorses such as the Japanese yen
will stop working a n d start behaving like the C R B index (which is a notoriously
difficult market with any strategy). W h e n that day c o m e s I want to have m a d e sure
that I have taken every possible action to keep myself from going broke.

PART 5

314

FIGURE

Money Management and Portfolio Composition

15.24

The cumulative monthly returns for the SDB strategy (both sample periods).

To get a feel for how m u c h each market is contributing to the end result you
can put together a table like Table 15.12, w h e r e I simply divided the total closedout profit for each market by the total profit for the portfolio as a whole. In this
case, we can see that the S & P 500 lowers the result by close to 2 5 % , while the crude

TABLE

15.11

1.5% fixed fractional trading with the directional slope strategy.

Strategy summaries
Profitability

Trade statistics
No Trades

2,885

Total (%)

1,601

AvgTr($)

4,420

Year(%)

15.73

Tr/Mark/Year

P factor

1.06

End Eq ($)

17,011,080

Tr / Month

Risk measurers

9.3
12.4

Time statistics

Max DD (%)

-57.56

Lrg Loss ($)

-2,374,435

Winners (%)

29.71

Lng Flat (m)

28.24

TIM (%)

98.06

Avg Days

7.00

CHAPTER 15

315

Money Management

FIGURE

15.25

The annualized monthly returns for the SDB strategy (both sample periods).

oil alone is m a k i n g up for m o r e than 9 0 % of the final profit. W h e n interpreting

these numbers, however, one m u s t be very careful not to m a k e any hasty conclusions. For one thing, just because one market h a p p e n s to have a low or negative contribution factor doesn't m e a n it is inferior to other markets; it could be that particular market (no matter how profitable it seems to be by itself) simply happens to
have all its profitable trades w h e n the portfolio as a whole is in a drawdown, which
will result in a lesser n u m b e r of contracts traded w h e n compared to a market that
always happens to work at its best when the portfolio has reached a new equity
high. Similarly, it could very well be that a market with a negative expectancy contributes positively to the portfolio as long as its winning trades tend to happen at the
m o s t opportune m o m e n t . Furthermore, even if it is reasonable to assume that a
market will have a negative contribution to the portfolio result, it still might add
positively to the very s a m e bottom line by being uncorrelated enough to markets to
keep the equity level up when the other markets are in a slump. A l s o , note that both
the winning and losing percentages are surpassing 100%. This is because these
n u m b e r s are picked from the aggregate portfolio, in which ending equity is not a
straight line function of the equity of the individual markets.
Anyway, j u s t for the heck of it, let us take a look at Table 15.13 and Figure
15.28 to see h o w the directional slope system would have performed, had we only
traded only the nine markets that a d d e d the m o s t to the bottom line. T h e nine mar-

316

PART 5

FIGURE

Money Management and Portfolio Composition

15.26

The equity curve for the directional slope strategy.

TABLE

15.12

Individual contribution factors for the directional slope strategy.

Market

Contributing

Corn

-24.98

S&P 500

-24.82

Juice

-59.71

Live cattle

-23.61

Lumber

4.77

Coffee

49.77

Japanese yen

51.33

Copper

12.25

Gold

-3.30

Eurodollar
Dollar index
Cotton

0.16
-4.49
7.40

CRB

-5.52

Crude oil

91.46

Canada dollar
T-bonds
Total

7.34
21.96
100.01

CHAPTER 15

Money Management

F I G U R E

317

1 5 . Z 7

The drawdown curve for the directional slope strategy.

kets were lumber, coffee, Japanese yen, copper, dollar index, cotton, crude oil,
C a n a d a dollar, and T-bonds. As you can see, this time the m a x i m u m drawdown has
decreased to 2 7 . 5 % , which is considerably better than it was from trading all markets. Unfortunately, this is counterbalanced by a m a x i m u m flat period that is m u c h
t o o long. A n o t h e r interesting observation that can be m a d e from comparing
Figures 15.26 and 15.28 is that Figure 15.26 at o n e point h a d a m a x i m u m total
equity of close to $40,000,000, before the result started to turn south in the middle of 1998.
Honestly now, pretend it is early 1998 and that you just finished building this
very s a m e system using these very s a m e markets. Would you then decide to trade
it with all the available markets or only the nine markets that, in hindsight, turned
out to p r o d u c e the best result, considering the deep drawdown that started later that
s a m e year?
I bet m o s t of you would have gone with the original portfolio, o n e reason
being that c o m m o n knowledge dictates that, "with a long-term strategy you should
trade as m a n y markets as possible, because you never k n o w w h i c h one will take
off next, p r o d u c i n g that h u g e w i n n e r that will m a k e it possible to build a house on
the m o o n . " This is very b a c k w a r d reasoning, however, because, for o n e thing, the
m o r e markets you trade the m o r e likely it is that you will be caught on the wrong
side w h e n panic strikes any o n e of them. Additionally, because all our funds are

318

PART 5

FIGURE

Money Management and Portfolio Composition

15.28

The equity curve for the directional slope strategy traded on the nine best markets.

limited, margin constraints might keep us from trading as m a n y markets as we

wish. Therefore, it m a k e s sense only to trade those that can provide us with the
highest likelihood of success. Just because we used markets like the S & P 500 and
C R B index to build the system does not m e a n we have to trade t h e m if there are
m o r e viable alternatives.
I am aware that this last statement to some extent contradicts my previous
statements that there is no difference between any two time series and that we
should place as m a n y trades as possible. But that is in the long run, and is why we
use markets like the S & P 500 and C R B index when we build the system, to m a k e
sure that when the Japanese yen starts to perform just as " b a d l y " at least I have
built in s o m e sort of catastrophe protection. A n d that is why we are not trading at
each market/system combination's historical optimal
In the short run, however, I know that there are s o m e differences (indicated
by the different optimal fs). It would be foolish to try to squeeze in markets like
those mentioned, or m a r k e t s that are too correlated with each other, as long as I
can't justify it for any other reasons, such as the likelihood of minimizing the
d r a w d o w n or flat time by increasing the n u m b e r of trades.
In ten years' time, however, the picture might look very different and perhaps
then I will be better off substituting all markets completely. The argument against
this last sentence is that, I would be leaving a lot of m o n e y on the table while wait-

CHAPTER 15

Money Management

TABLE

319

15.13

Results for a preferred portfolio with the directional slope strategy.

Strategy summaries
Profitability
End Eq ($)

Trade statistics
No Trades

1,646

Total (%)

20,709,401
1,971

AvgTr($)

10,867

Year(%)

16.91

Tr/Mark/Year

5.3

P factor

1.23

Tr / Month

7.1

Risk measurers

Time statistics

Max DD (%)

-27.54

Lrg Loss ($)

-1,509,982

Winners (%)

31.83

Lng Flat (m)

26.33

TIM (%)

98.06

Avg Days

11.00

ing for the evidence to be strong enough to warrant the switch. True, but my answer
to that is the same as for the type I and type II errors: "Better safe than sorry."
Furthermore, you have to decide what it is you are trying to achieve. We are
discussing trend-following strategies, consisting of a m o n e y m a n a g e m e n t regim e n , a portfolio of markets, and a trend-following system, and the s a m e reasoning holds true for the complete strategy as it does for the single system. That is,
we are not trying to pick tops and bottoms, b u t rather to j u m p on a market in
motion once the trend has proved itself and then ride that m o v e until our stops and
exit techniques tell us that it is no longer there.
A n o t h e r practical reason why we shouldn't trade every market is that the
m o r e markets we trade the m o r e inefficient the strategy will be, lowering the optim a l for the portfolio as a w h o l e . A n d because we like to have plenty of r o o m
b e t w e e n the we are using a n d the true optimal we had better not b o g down the
portfolio too m u c h . My research shows that a portfolio of 12 to 18 different, lowcorrelated markets, possibly traded with as m a n y as three different systems, seems
to work best. But this conclusion is b a s e d on my personal preferences, and I urge
you to do your own research to c o m e up with a portfolio that m a k e s the m o s t sense
to you.

The DBS Strategy

For the D B S system the results are pretty m u c h the s a m e as those for the directional slope strategy, with the s a m e markets producing similar results. This is indicated by Table 15.14, which shows that both the flat time and the drawdown are a
little t o o steep. Rather than once again going through w h y this is the case, only
repeating the discussion regarding the directional slop strategy, let's j u s t move on.

320

PART 5

TABLE

Money Management and Portfolio Composition

15.14

1.5% fixed fractional trading with th DBS strategy.

Strategy summaries
Profitability
End Eq ($)

Trade statistics
No Trades

1,581

Total (%)

21,787,934
2,079

AvgTr($)

11,425

Year (%)

17.21

Tr/Mark/Year

5.1

P factor

1.11

Tr / Month

6.8

Risk measurers

Time statistics

Max DD (%)

-42.37

Lrg Loss ($)

-1,123,734

Winners (%)

33.9

Lng Flat (m)

30.67

TIM (%)

99.98

Avg Days

11.00

CHAPTER

16

Portfolio Composition

In m a n y ways, we have already covered a large part of the complexity of portfolio

composition. We discussed finding m a r k e t s that are not too correlated with each
other, but also that a market with a negative mathematical expectancy still might fill
an important roll for the portfolio as a w h o l e . A n o t h e r important factor to consider
is to use the s a m e m a r k e t s in several different market/system combinations to cover
a w i d e r range of scenarios a n d o u t c o m e s . Consider, t o o , not picking too m a n y markets out of the s a m e group of markets, such as the currencies, and to diversify over
different time frames j u s t as we should over different markets a n d systems. In this
chapter we look at three different ways of putting together a portfolio, do some testing on historical data, a n d discuss a few pros and cons associated with e a c h method.

TOTAL EQUITY CONTRIBUTION I

O n e of the m o s t obvious (but m a y b e also one of the m o s t naive) ways of picking
the m a r k e t s to trade in the future is simply to look at h o w they have performed in
the past and select a g r o u p of top performers. For instance, if we decide to pick the
12 t o p markets from our long-term portfolios (four from each portfolio, without
picking the s a m e m a r k e t twice), we end up with the following 12: r o u g h rice, natural gas, D m a r k (Euro), dollar index, Japanese yen, coffee, cotton, T-bonds, orange
j u i c e , crude oil, copper, a n d C a n a d a dollar.
Tables 16.1 a n d 16.2, and Figures 16.1 a n d 16.2 show the result from trading
a portfolio consisting of all the above 12 markets, each o n e in combination with the
three different long-term systems, plus four short-term systems applied to the S & P
500 (the two entry techniques in the original Black Jack system w e r e divided into
321

322

PART 5

FIGURE

Money Management and Portfolio Composition

16.1

The equity curve for Charity I.

two separate systems), for a total of 40 different market/system combinations. The

time period covered is January 1980 to October 1999 and the fixed fractional trading level is set to 0 . 5 % . As usual, $75 p e r contract traded was deducted for slipp a g e and c o m m i s s i o n . I call this portfolio Charity I.
As you can see Charity I m a d e close to 3 3 % per year, for a total ending equity of close to $250,000,000. From Table 16.1 we can see that the m a x i m u m intramonth drawdown peaked at close to 24 %, while Figure 16.2 shows that the lowest
end-of-month drawdown did not grow past 1 5 % . These are very good numbers,

FIGURE

16.1

0.5% fixed fractional trading with Charity I.

Strategy summary
End Eq ($)
Total (%)
Year (%)

244,040,528
24,304
32.76

Max DD (%)

-23.81

Lng Flat (m)

17

CHAPTER 16

Portfolio Composition

TABLE

323

16.2

Percentage period returns, n months rolling window for Charity I.

Cumulative
Most recent
Average
Best

12

24

-11.06

-3.95

-11.76

16.03

15.67

97.60

131.19

196.04

2.53

8.07

16.88

37.20

93.34

180.59

309.77

500.63
1120.37

36

48

60

21.98

59.01

92.24

131.99

257.59

425.66

744.94

-11.06

-12.41

-11.76

-8.22

0.84

18.15

19.35

48.66

St. dev.

5.52

11.31

17.07

28.14

60.48

109.41

175.53

253.37
444.57

Worst
EGM

2.38

7.48

15.63

34.28

83.64

158.38

270.27

Sharpe

0.46

0.71

0.99

1.32

1.54

1.65

1.76

1.98

Avg. winning

4.98

11.80

20.30

39.30

93.34

180.59

309.77

500.63

-3.04

-3.31

-4.03

-2.97

N/A

N/A

N/A

12

-14.89

-22.14

Avg. losing
Annualized

Most recent

-75.5

Average

34.96

36.4

36.61

16.03
37.2

N/A
24

36

48

60

7.55

25.49

23.31

24.24

39.05

41.04

42.28

43.13

89.1

73.87

70.49

64.93

Best

985.08

539.29

269.56

131.99

Worst

-75.5

-41.14

-22.14

-8.22

0.42

5.72

4.52

8.25

53.51

37.05

28.14

26.68

27.94

28.84

28.72

St. dev.
Sharpe
% winners

90.55
0.39

0.68

0.99

1.32

1.46

1.47

1.47

69.53

75.32

85.96

95.05

100.00

100.00

100.00

1.5
100.00

indeed, once again showing the importance of trading as frequently as possible. The
only not-so-good n u m b e r is the flat period that might be considered a little too long,
especially since that we are working with such a favorable portfolio of markets.
T h e reason behind the relatively long flat time probably can be found in this
portfolio's greatest drawback: the high correlation and similarity between several
of the markets. For o n e thing, there are as m a n y as four currencies and two energies. It is doubtful this would be a preferable solution for a real-life portfolio. But,
on the other hand, there are several traders w h o use only o n e system, trading nothing but a few currencies, or nothing but the S & P 500. So, in that regard, at least
Charity I is far m o r e diversified than m a n y other real life examples.
O n e good thing is that we are not risking any m o r e than 0 . 5 % per trade,
w h i c h m u s t be considered very low. This m e a n s that at any particular time we
never risk m o r e than 1.5% of the total equity going into the trade, per market ( 2 %
for the S & P 500), which is no m o r e than in any of the single system portfolios discussed earlier. Despite the very low risk and the fact that we are only trading 2.5
t i m e s as m a n y system/market combinations as c o m p a r e d to the original long-term
portfolios, we are still raising the total equity tenfold or m o r e .
Unfortunately, however, one other thing that increases exponentially is the computational time. I did all this work on a computer with two Pentium III, 500 M H z

PART 5

324

FIGURE

Money Management and Portfolio Composition

16.2

The drawdown for Charity I.

processors and 256 MB memory, but it still can take as long as half an hour or more
for the computer to do all the necessary calculations. This is not something I recommend you to do unless you have plenty of computational power.

CORRELATIONS AND COVARIANCES

A m o r e sophisticated m e t h o d than just picking out the m o s t profitable markets
could be to look at their correlations and covariances with each other. T h e trick
then is to put together a portfolio consisting of markets with as low correlation to
or covariance from all the others as possible, while at the s a m e t i m e m a k i n g sure
that the m a r k e t s suggested by the correlation analysis also are likely to k e e p their
positive m a t h e m a t i c a l expectations and profit potentials when traded on previously unseen data. To analyze for correlations and covariances you could put
together a spreadsheet like that in Figure 16.3. In this case, I c h o s e to look at
weekly data.
T h e correlation between crude oil a n d British p o u n d in cell G 1 0 :
=CORREL(Data! $L3 :$L 1036,Data!F3 :F 1036)
T h e covariance between crude oil and British Pound in cell J 7 :

CHAPTER 16

Portfolio Composition

FIGURE

325

16.3

Using Excel to calculate correlations.

=CORREL(Data! $F3 :$F 1036,Data!L3 :L 1036)*

STDEV(Data!$F3:$F1036)*STDEV(Data!L3:L1036)
With the help of the spreadsheet depicted in Figure 16.3,1 put together a second
portfolio that I traded the very same way as Charity I. The long-term markets I selected were corn, Canada dollar, Eurodollar, heating oil, coffee, Nikkei index, orange
juice, platinum, Swiss franc, T-notes, and wheat. The criteria for selecting the markets
w a s i n order of importanceto pick only two markets from the same group of markets, to try to find previously unused markets, and to keep the correlations and covariances between different markets as low as possible. I called this portfolio Charity II.
You can see the hypothetical results from trading Charity II in Table 16.2, and Figures
16.4 and 16.5.
I wanted to c o m e up with a m o r e or less new set of markets to see what the
results would have looked like when traded on previously unseen and, in hindsight,
not necessarily the most profitable markets. If you wish, you can look at this portfolio as a worst-case scenario of what we might have been able to produce had we failed
to pick the most profitable markets but still used some c o m m o n sense in the portfolio construction process.
As you can see from Table 16.3, we still would have made more than 2 0 % per year,
even with these secondary markets. Again, however, the flat time is a little too long and
this time the maximum drawdown also has moved closer to what could be considered
tolerable. Still, however, there are plenty of professional traders and analystsmyself
includedwho would be very happy for numbers like these, if produced in real-time
trading. Taking $1,000,000 to close to $38,000,000 in less than 20 years is not bad at all
and plenty more than what the stock market has produced over the same time period.

TOTAL EQUITY CONTRIBUTION II

Instead of j u s t looking at the profit contribution, it also is possible to look at how
m a n y of the m o v e m e n t s in a t i m e series can be explained by the movements in

PART 5

326

FIGURE

Money Management and Portfolio Composition

16.4

The equity curve for Charity II.

another t i m e series. To do this we use the Pearson correlation coefficient which

returns a value between 1 and 1, very m u c h like the correlation coefficient. T h e
difference between the Pearson correlation coefficient and the n o r m a l correlation
coefficient is that the Pearson correlation coefficient requires that one of the variables be dependent on the other variable.
T h i s m e a n s that if two markets have a Pearson correlation coefficient close
to 1, almost the entire m o v e of the dependent market can be explained by what is

TABLE

16.3

0.5% fixed fractional trading with Charity II.

Strategy summary
End Eq ($)
Total (%)
Year(%)

37,797,489
3,680
20.59

Max DD (%)

-28.89

Lrg Loss ($)

-1,690,564

Lng Flat (m)

23.29

CHAPTER 16

Portfolio Composition

FIGURE

327

16.5

The drawdown for Charity II.

going on in the independent market. W i t h a Pearson coefficient close to 0, the two

markets m o v e independently of each other. Tests like these are very c o m m o n in the
stock market, w h e r e m a n y analysts look at h o w m a n y of the m o v e m e n t s in a particular stock can be explained by the movements of the market as a w h o l e .
N o t e that this is a way of trying to explain h o w m u c h of a particular stock's
m o v e can be explained by the m o v e in the market, which is not the s a m e as a
stock's beta, w h i c h indicates h o w m u c h a stock is likely to move, provided a certain m o v e in the market. For instance, a stock w i t h a beta higher than 1, say 2, is
likely to m o v e twice as m u c h in the same direction as the rest of the market, b u t
not all of that m o v e m u s t be explained by the m o v e of t h e m a r k e t w h i c h is w h a t
we m e a s u r e w i t h the Pearson coefficient.
In this case, we do it the other way around and look at h o w m u c h of the total
equity can be explained by the total equity from a certain market/system combination. This is an especially good m e t h o d to use w h e n we have the possibility to trade
the same market with several different systems and would like to narrow down the
portfolio a little bit. W h a t we look for is, first of all, markets that influence the total
equity positively. We are also careful to choose markets that seem to be too good to
be true, because this might m a k e the portfolio too dependent on the performance of
a few markets, w h e n traded on previously unseen data in real-time trading. A major

328

PART 5

Money Management and Portfolio Composition

difference between this m e t h o d and the correlation/covariance method is that with

this m e t h o d we are, so to speak, working backwards, because we first trade every
market/system combination we can think of in one m e g a portfolio and then deduct
all the unwanted market/system combinations from that portfolio, one at the time.
Before we get started, however, I need to throw in another word of warning.
Just as with the other techniques, this m e t h o d is very computationally intensive
and time c o n s u m i n g once you have everything up a n d running in Excel. N o t only
that, but to get everything up and running you also need to export from
TradeStation the trading results for every market/system combination you can
think of. Now, this w o u l d n ' t be that bad if it w e r e n ' t for the fact that TradeStation
does n o t have a way to apply the s a m e system to several markets (or otherwise
manipulate several markets) all at once. At this point, it takes me about a minute
(and I ' m b e c o m i n g pretty quick at this) to set up one market/system combination
from a set of predefined workspaces that I have built. I m a g i n e doing this on several 100-market/system combinations.
It is a drag, to say the least. A n o t h e r way to illustrate this include the fact that
there seems to be no way to set the starting date once and for all, or once a n d for
all to define a particular contract as a futures contract, or to default as h i d d e n all
data streams but the first one. To do all this it takes me at least 20 m o u s e clicks
(conservatively counting) for each market to set up the chart with the necessary
system. I will tell you, I am not writing a b o o k like this j u s t because I think it is
fun to tinker around with TradeStation.
Nonetheless, Table 16.4 shows the Pearson correlation coefficients together
with the equity contribution factors for 23 different markets traded with the S D B
system over the time period January 1980 to October 1999. Again, the initial markets were picked in such a way as not to use those obvious ones already used in many
of the previous examples. Table 16.5 shows the summarized results for this portfolio. The initial equity was set to $1,000,000 and the percentage risked per trade was
set to 1.5%. As usual, $75 was deducted for slippage and commission. With these
settings, the strategy produced an ending equity of close to $29,000,000, corresponding to a percentage return of close to 1 9 % per year. So far so good, but a m a x i m u m
drawdown of close to 4 0 % and a longest flat time of m o r e than 47 months would
have m a d e this a very difficult portfolio to trade in real life.
From these 23 markets, I deducted the six m a r k e t s with the lowest (Au dollar, soybeans, platinum, live cattle, cocoa, a n d British p o u n d ) and the three markets w i t h the highest (rough rice, Eurodollar, and cotton) Pearson correlation coefficients, ending up with a portfolio consisting of sugar, N i k k e i index, natural gas,
municipal b o n d s , lean hogs, lumber, coffee, copper, gold, feeder cattle, crude oil,
C a n a d a dollar, corn, and wheat. T h e s e I then traded with a fixed fractional setting
of 2 % . (With fewer markets, we can increase the risk p e r trade somewhat, because
the strategy n o w should be a little m o r e efficient, while at the s a m e time our overall m a r g i n requirements have decreased.)

CHAPTER 16

Portfolio Composition

TABLE

329

16.4

Initial markets for Charity III.

Market

Contributing

Pearson

-5.46

-0.6592

Au dollar
Sugar
Soybeans
Rice

0.66

0.0561

-2.52

-0.4445

15.85

0.9508

-22.93

-0.8640

Nikkei index

7.55

0.8741

Natural gas

12.86

0.9137

Platinum

Municipal bonds

3.80

0.8302

Lean hogs

-0.13

-0.3184

Live cattle

-2.09

0.1241

8.83

0.6967

Coffee

28.58

0.7810

Copper

5.40

0.9064

Gold

5.19

0.6305

Feeder cattle

1.84

0.8413

Eurodollar

0.64

0.9099

Cotton

20.97

0.9098

Crude oil

32.29

0.8362

Canada dollar

-1.82

0.2626

Cocoa

-9.60

-0.6489

Lumber

Corn

6.79

0.7143

British Pound

-4.17

-0.5608

Wheat

-2.54

0.4263

Table 16.6 shows the result from trading this portfolio. T h e ending equity is
close to $40,000,000, corresponding to a percentage return of m o r e than 2 0 . 5 0 %
p e r year. T h e d r a w d o w n has decreased to a tolerable 2 9 % . T h e longest flat time
also has decreased considerably but is still, at close to 31 m o n t h s , a little too long.

TABLE

16.5

Initial strategy summary for Charity III.

Strategy summary
End Eq ($)
Total (%)
Year(%)

29,016,313
2,802
18.96

Max DD (%)

-39.29

Lrg Loss ($)

-823,050

Lng Flat (m)

47.57

PART 5

330

TABLE

Money Management and Portfolio Composition

16.6

Modified portfolio, strategy summary for Charity III.

Strategy summary
End Eq ($)
Total (%)
Year (%)

38,565,334
3,757
20.72

Max DD (%)

-29.07

Lrg Loss ($)

-1,579,683

Lng Flat (m)

30.81

All in all, however, this is not too b a d for a o n e - s y s t e m strategy, traded on a

b u n c h of markets that were basically chosen because they h a d not been used
before or because they were plain average.

CHAPTER

17

Increasing Your Confidence

Let us briefly look at a few techniques for h o w you can increase your confidence
in your paper tigers. T h e most obvious m e t h o d is to save some of your data for outof-sample testing. If the system still holds up w h e n tested on this part of your data,
chances are very good that you have m a n a g e d to c o m e up with a m o d e l that does
not j u s t m a n a g e to catch a p h e n o m e n o n only associated with the data you used for
your testing and optimization. T h e second method, and the one I prefer, is to alter
the different input variables one at a t i m e by approximately 5 0 % . If the result differs too m u c h from the original result, chances are that particular variable might be
too sensitive to be trusted with real-time data, and you m u s t go back to the drawing board. To be sure, I also run a best- and a worst-case scenario a n d will say m o r e
about this shortly.

OUT-OF-SAMPLE TESTING
For the first method, we specifically put some data aside when we built the Standard
Deviation Breakout (SDB) system. Table 17.1 shows how the original S D B system
(without exits and filters) fared w h e n tested on the out-of-sample data, covering the
period October 1992 to October 1999, trading one contract only. This table is directly comparable to Table 10.18. If you compare the two, you find that 12 markets had
both a higher profit factor and a lower standard deviation when tested on the out-ofsample data, and that nine markets also had lower drawdowns, but that only four markets managed to increase the value for the average trade. For the in-sample period, all
16 markets were tradable according to our rule-of-thumb, but for the out-of-sample
period, only nine markets were tradable. Overall, this indicates that the system is still
331

332

PART 5

TABLE

Money Management and Portfolio Composition

17.1

The SDB system tested on out-of-sample data.

Original system without exits and filters
Market

P. factor

Avg. trade

2 St. dev.

Drawdown

Crude oil

4.29

1,479.39

7,057.42

-2,978.23

T-bonds

1.84

959.06

8,535.63

-14,922.51

T-bills

0.78

-53.19

1,082.75

-2,918.79

Rough rice

2.35

469.66

4,414.97

-2,471.24

Nikkei index

1.16

278.92

10,217.26

-13,980.23
-9,231.34

Natural gas

1.27

370.80

8,909.49

Live cattle

0.45

-302.45

1,924.89

-6,976.45

Lumber

0.75

-290.64

4,713.38

-14,987.61

Coffee

2.28

2,677.13

26,860.39

-15,845.70

Japanese yen

5.16

3,240.19

14,366.39

-6,369.00

Copper

1.18

93.38

3,015.47

-5,342.02

Gold

1.13

47.16

2,083.29

-4,622.61

Dollar index

2.47

1,060.05

7,014.18

-9,012.39

D mark (Euro)

1.90

536.96

4,705.28

-3,672.00

Cotton

1.26

203.75

6,475.78

-8,008.77

Wheat

1.40

124.61

2,077.13

-2,711.16

doing a good j o b keeping down the risk and the cost of doing business, but that it
also is not as market insensitive as we would have liked.
Tables 17.2 and 17.3 show h o w the S D B system p e r f o r m e d on the out-ofsample data, with the stops a n d exits added. Table 17.2 also is comparable to
Table 10.19. O n c e again we m a n a g e d to improve the profit factor for 12 markets,
but unfortunately, the value of the average trade continued to decrease for 11 markets, resulting in only 6 tradable m a r k e t s according to our rule of t h u m b , leaving
only a few m a c r o e c o n o m i c oriented m a r k e t s with high point values as tradable
alternatives. T h i s is, of course, no g o o d a n d m i g h t indicate that we have set the
stops and exits too tightly. In the best of worlds, we should go b a c k to the drawing b o a r d to e x a m i n e w h a t h a p p e n s if we exclude or modify one or several of the
exit techniques.
On the m i n u s side is also the fact that we only m a n a g e d to decrease the drawd o w n for four markets, which confirms the fact that the stops are not doing their
j o b s as intended. T h e n it does not matter that the standard deviation decreased for
a total of nine markets, indicating a lower risk. If the profits are not there, something must be done. Hopefully, a filter adds to the performance again.
Tables 17.4 and 17.5 show the out-of-sample results for the final system,
with both the exits and the filter added. With the filters added, we once again man-

CHAPTER 17

333

Increasing Your Confidence

TABLE

17.2

The SDB system tested on out-of-sample data.

Modified system with exits
Market

P. factor

Avg. trade

2 St. dev.

Drawdown

Crude oil

3.05

1,044.50

6,317.99

-2,287.13

T-bonds

2.09

1,245.85

10,007.90

-14,927.71

T-bills

0.73

-69.61

1,190.71

-3,477.18

Rough rice

0.95

-18.33

1,616.97

-2,578.88

Nikkei index

1.34

596.12

11,407.45

-11,504.12

Natural gas

0.74

-333.44

6,770.50

-16,311.91

Live cattle

0.51

-271.59

2,053.99

-6,430.07

Lumber

0.94

-49.30

4,132.89

-12,826.20

Coffee

0.46

-724.16

4,815.83

-16,986.35

Japanese yen

2.79

2,131.80

12,359.79

-8,681.72

Copper

0.85

-88.57

2,617.81

-7,656.22

Gold

0.97

-11.36

2,035.93

-5,796.95

Dollar index

2.80

1,250.31

7,839.86

-9,211.56

D mark (Euro)

1.55

411.75

5,445.90

-4,172.52

Cotton

1.29

209.09

5,914.57

-9,608.86

Wheat

1.03

13.43

2,149.37

-3,231.83

aged to improve the profit factor for 12 markets, but the value of the average trade
also continued to decrease for 10 markets. This time, however, the decrease c a m e
in markets that already w e r e quite profitable, which in the end resulted in a total
of 9 tradable markets.
If we c o m p a r e the final system, traded on out-of-sample data (Table 17.4) to
the very original system, tested on in-sample data (Table 10.18), we can see that
all m e a s u r e s are both lower in value and m o r e uniform (i.e. lower standard deviations) for the final out-of-sample system, than for its original in-sample counterpart. This also s e e m s to hold true if we c o m p a r e the final in-sample system (Table
13.7) with the final out-of-sample system (Table 17.4).
Of course there are plenty m o r e research and comparative analyses that
c o u l d a n d s h o u l d g o into this before you finally decide to start trading your
m o d e l , but in conclusion we can say that the fact that the system seems to m o v e
toward a m o r e uniform behavior w h e n comparing the different markets is a good
thing, as are the facts that the drawdown and standard deviation of the value of the
average trade seem to stay at relatively low levels. An ever-increasing profit factor
for most markets also is a positive. T h e b a d n e w s is that several markets did not
m a n a g e to p r o d u c e a tradable average profit p e r trade a n d that s o m e even turned
negative. B u t such are the traits of a n o n - m a r k e t - s p e c i f i c m o d e l . It will work well,

334

PART 5

TABLE

Money Management and Portfolio Composition

17.3

Differences between the original system and system with exits.

Differences
Market
Crude oil
T-bonds
T-bills

P. factor

Avg. trade

2 St. dev.

Drawdown

-28.77%

-29.40%

-10.48%

-23.21%

13.64%

29.90%

17.25%

0.03%

2
1

Better

-5.59%

30.87%

9.97%

19.13%

-59.79%

-103.90%

-63.38%

4.36%

Nikkei index

15.42%

113.72%

11.65%

-17.71%

Natural gas

Rough rice

-41.72%

-189.92%

-24.01%

76.70%

Live cattle

12.33%

-10.20%

6.71%

-7.83%

Lumber

26.14%

-83.04%

-12.32%

-14.42%

Coffee

-79.78%

-127.05%

-82.07%

7.20%

Japanese yen

-45.87%

-34.21%

-13.97%

36.31%

Copper

-28.18%

-194.85%

-13.19%

43.32%

Gold

-14.47%

-124.09%

-2.27%

25.40%

13.00%

17.95%

11.77%

2.21%

Dollar index

-18.44%

-23.32%

15.74%

13.63%

Cotton

D mark (Euro)

2.40%

2.62%

-8.67%

19.98%

Wheat

-25.99%

-89.22%

3.48%

19.20%

Better

12

on average, on all markets over a longer period of time, but there invariably will
be times w h e n s o m e markets start to m o v e against you.

ALTERING THE INPUTS

Altering the inputs is another g o o d way to check the robustness of your system. As
is the case with the out-of-sample testing, this can be done both before a n d after
you have attached the m o n e y m a n a g e m e n t . In this case, we will look at how the
Gold Digger system would have performed on the S & P 500, on a one-contract
basis, over the period January 1995 to October 1999. T h e parameters we will
c h a n g e are the breakout filter (60 days 3 0 days), the profit target ( 2 . 8 % 1.4
percentage points), the trailing stop ( 0 . 6 % 0.3 percentage points), and the stop
loss ( 1 . 1 % 0.5 percentage points).
We will change one p a r a m e t e r at the time keeping all the others constant on
their original values. W i t h four parameters to e x a m i n e , this m e a n s that we must
r u n the system test eight times. With these performance s u m m a r i e s as a base, we
then look at w h i c h individual parameter settings produced the best and the worst
results, and alter all the parameters at once to c o m e up with the worst and the best

CHAPTER 17

335

Increasing Your Confidence

TABLE

17.4

The SDB system tested on out-of-sample data.

Modified (final) system with exits and filter
Market

P. factor

Crude oil

3.11

2 St. dev.

Drawdown

1,160.48

6,926.93

-2,485.61
-8,757.69
-1,278.81

Avg. trade

T-bonds

2.57

1,588.33

10,332.34

T-bills

0.82

-37.29

1,048.49

240.89

3,548.54

-3,438.43
-13,084.91
-8,022.83

Rough rice

1.57

Nikkei index

1.42

774.75

12,214.28

Natural gas

0.92

-95.87

6,898.78

Live cattle

0.52

-265.61

2,130.85

-5,406.43
-10,228.79

Lumber

0.93

-58.91

3,895.54

Coffee

1.71

905.00

14,273.04

-11,468.87

Japanese yen

1.82

1,260.72

12,526.50

-14,863.14

Copper

0.91

-52.43

2,726.10

-5,556.51

1.68

235.91

2,528.68

-4,544.11

Dollar index

2.75

1,348.52

8,472.42

-4,956.25

D mark (Euro)

0.90

-120.21

5,763.42

-9,866.05

1.88

587.89

6,767.50

-6,758.33

1.02

6.48

2,158.88

-3,341.48

Gold

Cotton
Wheat

p a r a m e t e r settings, respectively, possible. In this case, all p a r a m e t e r settings are

j u d g e d by the average profit only. Of course, you c a n a n d shoulduse other
evaluation m e a s u r e s as well. A n d of course, y o u w o u l d c o m e up w i t h results that
are m o r e reliable if y o u do the comparisons over a whole portfolio instead of o n e
single market.
Tables 17.6 a n d 17.7 s h o w t h e results from t h e worst a n d t h e b e s t p a r a m eter settings respectively. T h e w o r s t p o s s i b l e setting t u r n e d out t o b e o n e w i t h
a 3 0 - d a y filter, a profit target of 1.4%, a t r a i l i n g stop of 0 . 6 % a n d a stop loss
of 1.6%. T h e b e s t setting i n c l u d e d a 9 0 - d a y filter, a 2 . 8 % profit target, a 0 . 3 %
trailing stop, a n d a 0 . 6 % stop l o s s . T h e results are directly c o m p a r a b l e to Table
12.8.
As y o u can see, even with the worst possible setting, the system is marginally profitable, w h i c h is a v e r y g o o d thing, indicating that even in the worst of times,
we should not lose our shirts. T h e question is, however, if we d e e m it profitable
e n o u g h or if we would have liked the results to be slightly m o r e similar to those of
the original p a r a m e t e r settings. In real life, only you can m a k e that decision and
only t i m e a n d plenty of b a d system-building attempts help you gain the experience
necessary. In this case, I would say that I w o u l d have liked to see the worst-case
scenario a little better than it is. With only o n e m a r k e t tested over a very limited

336

PART 5

TABLE

Money Management and Portfolio Composition

17.5

Differences between SDB system with exits only and final system.
Differences
Market

P. factor

Crude oil

Avg. trade

2 St. dev.

Drawdown

Better

1.94%

11.10%

9.64%

8.68%

T-bonds

22.91%

27.49%

3.24%

-41.33%

T-bills

11.63%

-46.43%

-11.94%

-63.22%

Rough rice

65.73%

-1414.49%

119.46%

33.33%

Nikkei index

6.32%

29.96%

7.07%

13.74%

Natural gas

24.05%

-71.25%

1.89%

-50.82%

2.26%

-2.20%

3.74%

-15.92%

-1.43%

19.49%

-5.74%

-20.25%

Coffee

270.30%

-224.97%

196.38%

-32.48%

Japanese yen

-34.70%

-40.86%

1.35%

71.20%

Live cattle
Lumber

Copper

6.79%

^0.80%

4.14%

-27.42%

Gold

73.64%

-2176.69%

24.20%

-21.61%

Dollar index

-1.68%

7.85%

8.07%

-46.20%

^t2.31%

-129.20%

5.83%

136.45%

Cotton

46.55%

181.17%

14.42%

-29.67%

Wheat

-1.66%

-51.73%

0.44%

3.39%

Better

12

10

D mark (Euro)

t i m e span and using one m e a s u r e r only, however, we do not have e n o u g h data to

c o m e to a definite conclusion.
As for the c o m b i n e d best-case scenario, we can see that it too produces an
average profit lower than that produced by the original p a r a m e t e r settings. This is
actually good n e w s , because it m e a n s that the original settings are doing a good
j o b supporting each other in their respective roles and p r o d u c i n g a final system
w h e r e the whole is larger than its parts, although there might be t i m e s w h e n the
true best-parameter settings differ slightly.

TABLE

17.6

Gold Digger, worst-case scenario, January 1995-October 1999.

Total trades

194

Profit factor

1.05

Lrg winner
Avg winner
Cum profit

Avg profit

0.04%

127

St Dev

1.72%

5,792

Winners

118

60.82%

76

39.18%

5.14%

17,348

Losers
Lrg loser

-8.30%

-28,013

1.26%

4,250

Avg loser

-1.86%

-6,275

4.52%

15,255

Drawdown - 1 9 . 5 2 %

-65,880

CHAPTER 17

TABLE

337

Increasing Your Confidence

17.7

Gold Digger, best-case scenario, January 1995-October 1999.

Total trades

206

Winners

Profit factor

1.31

Lrg winner
Avg winner

Avg profit

0.17%

572

St Dev

1.73%

5,832

C u m profit

10

50.00%

5.14%

17,348

Losers
Lrg loser

10

50.00%

-8.30%

-28,013

1.44%

4,874

Avg loser

-1.11%

-3,730

37.46%

126,428

Drawdown

-14.76%

-49,815

PART

FIVE

A Few Final Thoughts About Part 5

The observant reader probably has noticed that up until Part 5 I have consistently
talked about " s y s t e m s " rather than "strategies," but that in Part 5 I have used the
word "strategy" m o r e frequently. This is because to m e , a " s y s t e m " is nothing but
the basic entry and exit rules. A "strategy," on the other hand, is a complete trading
plan, including a system, a m o n e y m a n a g e m e n t regimen, and a well-thoughtthrough reason for trading a particular set of markets. To m e , a system without a
proper m o n e y m a n a g e m e n t regimen attached to it can never be a strategy.
I have noticed that O m e g a research recently went from using the word "syst e m " to using the word "strategy." That is nothing but another ridiculous way for a
software v e n d o r to try to cash in on a new b u z z word or distance itself from another word that might have lost a little bit of its sex appeal. There is no way you can
use TradeStation as a standalone tool to build a complete strategy. You c a n ' t even
do proper system testing on a single market, not to mention all the hassle you have
to go through if you want to test a portfolio!
In Part 5 we took a closer look at how to put together a fixed fractional m o n e y
m a n a g e m e n t regimen, a n d we did so using percentage-based calculations throughout. It is important to understand that we would not have been able to do this had
we not d o n e our h o m e w o r k properly and g o n e through all the excruciating work in
the previous chapters. It should now be clear that everything affects everything else;
to build a trading strategy that is larger than its p a r t s t h e system, the m o n e y m a n agement, and the portfolio c o m p o s i t i o n w e m u s t know exactly w h a t it is we want
to achieve, and there is no way of cheating.
I am a big fan of Mr. Vince's work, but even so there have been a couple of
occasions throughout this section were I have implicitly argued against him and
339

340

PART 5

Money Management and Portfolio Composition

what seems to have b e c o m e " c o m m o n w i s d o m , " or otherwise argued for a different way of approaching a specific problem or topic. Before I end this book, I
would like to take the opportunity to address these in m o r e detail.
C o n c e r n i n g percentage-based versus point-based data, if I understand Mr.
Vince correctly, it should not matter what data we use. Mr. Vince acknowledges
the problem of using point- or dollar-based data in heavily trending markets, but
also says that if you have to use so m u c h data that the trend starts to b e c o m e a
problem, then you are probably using too m u c h data anyway (The Mathematics of
Money Management, p. 83 ff. John Wiley & Sons, 1992).
N o t h i n g can be m o r e wrong. You should use as m a n y data as you possibly
can to m a k e your results as robust as possible. There is no way around this,
because for a system to work in the future it needs to exploit a type of market
a n o m a l y that is not dependent on w h a t level we are trading, what the most recent
trend looked like, or w h o h a p p e n e d to be president at the time. It m u s t be a p h e n o m e n o n not dependent on any particular circumstances, only referable to that
particular point in time. Furthermore, using percentage-based calculations is the
only way to m a k e your results comparable a m o n g different markets. Mr. Vince
says that for a system to be robust, it should work on several markets, but he does
not address the problem of h o w to get there. (Granted that is not the p u r p o s e of
any of Mr. Vince's b o o k s , but still....)
Another key point throughout this section has been that you do not have to use
your largest historical loser to calculate optimal f. T h e optimal is the that gives
you the highest T W R value in accordance with the constraints you have applied to
your model. Unfortunately, most analysts and system designers consistently refer to
the optimal as the corresponding to the largest historical loser. This has stuck in
people's minds and m a n y have therefore dismissed the entire theory surrounding
fixed fractional investing as something very dangerous. But this is only one way of
many to relate optimal to a specific constraint. That is, we do not want to experience a loser larger than our largest historical loser, the largest historical loser being
the constraint. B u t we could equally as well have said that we do not want to experience a loser larger than our largest historical loser, times two, because this would
result in an even higher optimal than the optimal most commonly referred to.
In this particular case I do not think it w a s Mr. Vince's intention to have
everybody stare themselves blind on his way of doing it, but that is the way people work. T h e y read something a n d take it for face value without putting their own
thinking caps on. True, trading with an / b a s e d on your largest loser is very dangerous, but as we have seen, it is m o r e or less impossible to use an that high anyway if you like to trade a well-diversified portfolio, because of several practical
reasons such as margin constraints, extremely high drawdowns, and an impossible
n u m b e r of contracts traded.
Instead of using your largest historical loser to calculate I suggest that you
simply use the stop loss level for each individual trade and that you keep that stop

PART 5

Money Management and Portfolio Composition

341

loss level the same, no matter what m o d e the m a r k e t is in. O n c e y o u have calculated the optimal in accordance w i t h the constraints of your choice, m a k e sure
that you actually use an far lower than that, b e c a u s e there is no guarantee that the
future, u n k n o w n , w o n ' t be far lower as well, especially for a market-specific syst e m built with point-based data.
M a n y portfolio managers, especially in the stock market, probably also have
tried to combine a fixed fractional m o n e y m a n a g e m e n t regimen with the Capital
Asset Pricing M o d e l ( C A P M ) a n d the Efficient Market Hypothesis ( E M H ) .
However, as far as I understand, the two cannot be combined, because for a system
to work equally as well on several different markets, we have to assume that there is
no difference in the statistical traits of all these markets. This might or might not be
so in real life, but for our purposes, this is what we must assume. A n d because we
assume this, we cannot deviate from it w h e n it is time to apply m o n e y management,
to the system we are building, because it would result in a suboptimal solution. If we
do not believe that the statistical traits for all markets are the same, then there is no
way we can build systems that should work equally as well in all markets; all systems would, out of necessity, have to be market specific, curve fitted, not robust at
all, and very, very dangerous to trade using any m o n e y management regimen.
A few also have argued that one should use different s depending on whether
the overall portfolio is in a drawdown, whether the market is in an uptrending or
downtrending m o d e , or whether we are about to go long or short. However, in doing
so we m u s t assume that we h a d done a b a d j o b constructing the underlying system
and that we think we can increase the return from the individual trades by tinkering
with the m o n e y management. But this is not so, because with a correctly built system there is no way of knowing, or even anticipating, the outcome of each individual trade. A well-working system m a k e s no difference between trying to pick tops or
bottoms, entering on breakouts to the up side or down side, whether the trend is up
or down, or even in which markets all this occurs. This also argues against the p o s sibility, within the confines of a specific strategy, to risk m o r e going long (short) in
an uptrend (downtrend) either by altering the stop loss level or by altering the bet
size. That said, this does not mean that several market-specific or even trend-specific characteristics or anomalies do not exist (in fact, we have stumbled u p o n and
acknowledged several such [short-term] anomalies within this book). If these are the
types of market anomalies you would like to exploit, then you m u s t build a system
that exploits exactly that, one market anomaly at a time. Those systems would, however, no longer be universally applicable and therefore probably m u c h less reliable
and robust than any of the systems we have worked with in this book.
Finally, a few, m o r e theoretically inclined, system builders and analysts have
advocated that rather than j u s t looking at the historical trades, it is better first to
calculate the distribution of the outcomes of the trades, and from there calculate
parametrically. At first glance this m a k e s a lot of sense, because using the m a t h e matical expression for the expected distribution of our future trades would m a k e

342

PART 5

Money Management and Portfolio Composition

our m o d e l s even m o r e forward looking than using the actual, historical trades. We
looked into the distribution of our trades w h e n we put together the stops and exits
for our short-term systems. In this particular case, the distributions we c a m e up
with hardly lend themselves to any parametric calculations of In fact, if they did,
I would say that we had a b u n c h of b a d systems on our h a n d s . Because as soon as
the distribution of our trades starts to resemble something that we can put a n a m e
to, especially if it turns out to be normally distributed, we are probably doing a
very b a d j o b in m a n a g i n g our trades, essentially leaving each trade to its own destiny as soon as we have entered into it, using no stop or exit techniques to streamline the o u t c o m e .
Instead, we should m a k e sure that each trade only has a set of very few, very
specific and clearly distinguished outcomes, one being the stop loss that we also
used for our optimal calculation. Remember, it is thanks to our stop and exit techniques we are able to exploit non-market-specific anomalies. If it were not for these
techniques, we would not be able trust our systems to a fixed fractional m o n e y
m a n a g e m e n t regimen in the first place. That said, there still might be plenty of systems with normally distributed trades that work m u c h better than those in this book.
T h e systems a n d strategies we explored so far are simply "paper tigers."

BACK TO THE FUTURE

T h e financial markets are not r a n d o m and there are ways to take advantage of this
n o n r a n d o m n e s s with the help of mechanical trading strategies. T h e good news is
that you do not have to be a rocket scientist to do so. You only need a tad more
knowledge than you hold about the topic already. Unfortunately, knowing too little
about a specific topic sometimes can be m o r e devastating than knowing nothing at
all. T h e big fish in this industry realize that, so they have no incentive to educate
you, because what is devastating for your economy is a gold m i n e to theirs.
Considering the above, it is no coincidence that our discussion on stop placements has the most exhaustive one, because when it comes to basic trading and
investing, getting out of a position is what nobody really talks about. Go through all
the recommendations from several analyst and brokerage companies and you will
find a multitude of opinions such as "strong buy," "buy," "cheap," "outperform,"
"stronger than average," "long-term buy," "positive outlook," but seldom will you
find the words "sell," "sell if it drops to," or "sell if it reaches." The strongest phrases in this regard seem to be " m e d i u m perform" or "hold." W h y is this? It is because
in this industry, selling something m e a n s bad news and your broker does not want
you to think about bad news or what might go wrong. Transmitting b a d vibes,
whether professionally right or not, is not good for business.
But because neither you nor I can blow hundred dollar bills out of our noses,
to b u y something (or go long or short on a futures contract) usually m e a n s we need
to get out of something else, either with a profit or a loss, but always, always at a

PART 5

Money Management and Portfolio Composition

343

point where we believe our m o n e y can c o m e to better use elsewhere, in accordance

with our research. This m e a n s that getting into a position is a one-time decision and
once it is m a d e it is over a n d done with. B u t the decision to get out of the same position is an ongoing process, w h e r e you constantly ask yourself "will my m o n e y
c o m e to better use elsewhere?" No matter if this " e l s e w h e r e " h a p p e n s to be in the
b a n k account, in another trading or investment opportunity, or under your mattress.
That is why no trading or investment technique is complete without a set of thoroughly researched stops and exits. T h e stops and exits not only form the core in the
trading process but also set the stage for a profitable m o n e y m a n a g e m e n t regimen.
Unfortunately, to research how and w h e r e to place your stops and exits you
m u s t k n o w a whole lot m o r e about your system than its m a x i m u m historical drawd o w n in dollar t e r m s . B e c a u s e this figure tells you nothing about w h a t to expect
in the future, j u s t as it gives y o u no clue whatsoeverabsolutely n o n e a b o u t h o w
to go about doing something about it. For this, you m u s t k n o w h o w to divide the
d r a w d o w n into several different subcategories, such as the start trade drawdown
( S T D ) , the end trade d r a w d o w n ( E T D ) , the closed trade d r a w d o w n ( C T D ) , and the
total equity d r a w d o w n ( T E D ) . A n d as if this w e r e not enough, you also m u s t know
h o w to calculate each trade's m a x i m u m adverse excursion ( M A E ) and m a x i m u m
favorable excursion ( M F E ) .
In Part 3 I showed you h o w to do all that a n d m o r e . For each of our e x a m p l e
systems we put together a set of stop and exit techniques that e n h a n c e d the performance for every single o n e of them. N o t only that, but for our short-term syst e m s these techniques were generic and n o n - m a r k e t - s p e c i f i c best c o m p r o m i s e s ,
principally developed with the help of r a n d o m entries and later applied to a m a r ket other than those used during the testing procedure. This m e a n s that there w a s
virtually no curve fitting involved a n d that we are exploiting are n o n - m a r k e t - s p e cific discrepancies that are likely to occur again a n d again in all markets.
This brings us to two other interesting conclusions:

You get a long-term edge in almost every market if y o u only trade with the
direction of the underlying trend.

If y o u do not k n o w or cannot form an opinion regarding the direction of the

long-term trend, you are better off trading short t e r m (i.e. approximately three
to nine days) using as little historical data as possible for your decision
making.

T h e statistical traits for the markets are pretty m u c h the s a m e whether the
trend is u p , down, or sideways. For instance, by j u s t looking at the last five days
of market action, you likely are not able to tell the direction of the underlying longt e r m trend. Therefore, for practical reasons, the short-term traits can be said to be
static in nature, although the small, nonmeasurable changes that might occur over
time lead up to a n d form the longer-term trend. At that level, the statistical traits
b e c o m e d y n a m i c w i t h measurable differences, distinguishing u p , down, and side-

344

PART 5

Money Management and Portfolio Composition

ways trends from each other. Another major advantage of trading short term, but
only in the direction of the long-term trend, is that you dare to w o r k with limit
orders, trying to pick tops and b o t t o m s , influenced by the increased security you
feel about k n o w i n g the long-term trend.
To figure out w h e r e the market is heading, both in the short and the long run,
and the t h e o r y w h y all this is happening in the first place, was what Part 4 was all
about. W h e n attaching the filters we also might have to go back and adjust the original idea a little, or perhaps even scrap it all together if it turns out that we cannot
c o m e up with a way to fit the filter to the rest of the machinery as efficiently as we
would like to. Not until all the pieces from Parts 1 and 4 fit snugly together can we
attach the m o n e y m a n a g e m e n t that not only transforms the system to a strategy, but
also will m a k e the whole picture larger than any of its parts. To this set of rules and
frameworks we also need to add the understanding of the trading process and the
creation of a rule-based trading strategy as a flow of events and intervening decisions, rather than individual decisions separated from each other in time and space.
Because if we do not understand and acknowledge this, we cannot see the benefits
of the type of trading a n d that it is not a matter of optimizing your system's individual signals to the underlying data, but "optisizing" your trading positions in relation to your overall strategy and other personal constraints and preferences.

INDEX

Note: Boldface numbers indicate illustrations. Italic t indicates a table.

200-day moving average, 70, 87
filtering, 228, 229
absolute vs. percentage values, 64
academic theories of market trends, 256-257
account size vs. return on account, 6-7
accumulation area in market analysis, 263,
264

adoption-diffusion model in market analysis,

263
ADX indicator, 241-242, 250, 252-253
in filtering, 234
arbitrage, 47, 257
asymmetric distribution, 188-189
average dollar value per trade, 20
average losing trade, MAE/MFE, 148,148,
149

average percentage profit, 28

in Black Jack, 209, 210
average percentage winner, 28
average profit, maximum adverse excursion
(MAE), 153-154, 1 5 4 ( 0
average profit multiplier (APM), 281-282
average return, Black Jack, 213-214, 214
average trade value, 7
average true range indicator, 233, 241
average weekly true range (AWTRO/AWTR),
83, 84
average winning and losing trades, 9
back-adjustment method of splicing, 31
best-fit regression lines in, 35
point-based adjusted method of splicing
and, 31, 32-37
ratio-adjusted method of splicing and, 31,
37-38
bar count/calculation in TradeStation,
129-130
bars per trade, average number of, 9-10, 10(f)
basic crossover signals, trends in markets,
86-87

bear markets, depth of historical drawdowns,

23,24
bell curve distribution, 187-188,188
best-fit regression lines, back-adjusted
method, futures contracts, 35
beta value, 327
Black and Scholes pricing formula, 233
Black Jack system, 53, 204-221
average percentage of profitable trades in,
78-80, 7 9 ( 0 , 78
average percentage profit calculation in,
74, 209, 210
average return in, 213-214, 214
average trade in, 216, 217
closed trade drawdown (CTD) in, 208, 209
cumulative profit and equity tops in, 22,
22, 217-218
entry techniques for, 216-217, 220-221
exit techniques for, 69-81, 7 5 - 8 0 ( 0 , 205,
216,216(0,217,219
filtering in, 239-240, 2 4 0 ( 0
Gold Digger system vs., 218-221,
218-221(0
kurtosis in, 211, 211, 2 1 2 ( 0 , 2 1 3 ( 0
maximum adverse excursion (MAE) for,
208, 209, 216
maximum favorable excursion (MFE) for,
208, 209, 216
Meander system combined with, 219-221,
219-221(0
minimum profit level statistics for, 77, 7 7 ( 0
percentage profitable trades in, 74
performance summary of, 217-221,
217-221(0
portfolio composition of, 321
profit factor in, 218
profit target in, 76, 7 6 ( 0 , 213-214, 214,
215-216
random number generator for, 204-205
ratio of average profit to standard
deviation in, 74
ratio of percentage profitable trades and
standard deviation in, 75
345

Index

346

Black Jack system (cont.)

skew in, 211, 212, 2 1 2 ( 0 , 2 1 3 ( 0
standard deviation of percentage profit in,
74
standard deviation of percentage profitable
trades in, 74
standard deviation in, 74, 209, 210, 216,
217,218
stop loss statistics in, 77-78, 7 8 ( 0 ,
205-206, 209,211-212
surface chart for, 209, 210
testing, using historical data, 80-81, 8 0 ( 0 ,
206
time-based stops in, 205, 214-215, 215,
219
time-based trailing stop in, 76-77
trade-length statistics for, 75-76, 7 5 ( 0 ,
205-206,211-212
TradeStation code for, 206-208
trailing stops in, 214-215, 215
uncertainty measures in, 77
Black, Fisher, 257
Black-76 pricing formula, 233
Bollinger-band indicators, 53, 63, 107-113,
108,110,111, 113(0

building a system, 47, 48, 52

Campaign Trading, Tactics and Strategies...,

23,123
Capital Asset Pricing Model (CAPM), 257,
258, 282, 341
catching market moves, 52-53
Charity I portfolio, 322-324, 322, 3 2 3 ( 0 , 324
Charity II portfolio, 325, 3 2 6 ( 0 , 327
Charity III portfolio, 328-330, 3 2 9 ( 0 , 3 3 0 ( 0
Chebchev's theorem, 191, 194
closed trade drawdown (CTD), 5, 23, 121,
122, 123, 134-135,135, 136, 208,
209, 343
commission, 17
Commitment of Traders Report, 262
Commodity Futures Trading Commission
(CFTC), 262
computer software for evaluating
performance, 1-2
consecutive winners/losers, maximum
number, 10-11
correlations, 274, 324-325, 3 2 5 ( 0
covariances, 324-325, 3 2 5 ( 0
Crash of 1929, 261,262
Crash of 1987, 37-38
creation-of-funds analysis, 262-263, 266
credit policies and market trends, 260

crossover signals, trends in markets, 86-87

cumulative profit, 21, 22, 28, 29, 160, 161,
217-218, 281
curve fitting, 115-116, 167, 282-283

data mining, 55-81

absolute vs. percentage values in, 64
better use of data in, 62-69
in Black Jack I system, 69-81, 7 5 - 8 0 ( 0
Bollinger-band type indicators in, 63
calculations for, 58-59
exit techniques and, 69-81
200-days moving average in, 70
drift in, 70
profit target for, 70
stop loss for, 70, 77-78, 7 8 ( 0
time-based stop for, 70
time-based trailing stop in, 76-77
trailing stop for, 70
trend filter in, 70
triggers for, 70
in Gold Digger I system, 59-62, 6 0 ( 0 ,
62(0
in Meander system, 64-69, 67, 6 8 ( 0 , 6 9 ( 0
normalization of market movements and,
64
percentage of moves of certain amplitude
and, 5 8 ( 0
periods of moves of certain lengths and,
57(0
regular moving averages in, 63
summary report for, 5 7 ( 0
for testing a system, 59
for tracking market moves, 56-62, 5 7 ( 0 ,
58(0
trends in markets and, 55-56
uncertainty measures in, 77
data selection for analysis, 52-53
decision to trade, 83-84
DeLuca, Dave, 223
DeMark, Tom, 70
demographics and market trends, 259-260
Dennis, Richard, 97-98
dependencies, 274
depth of historical drawdowns, 23, 24
Directional Slope Method, 87-97, 90-94,
9 3 ( 0 , 9 4 ( 0 , 313-319, 314(0, 316,
316(0,317,318,319(0,313
average profit factor in, 89
code for, 88, 124-129
data sort function in, 89
Dynamic Breakout System(DBS) vs., 105,
107, 174-175, 174(0, 176

Index

347

Directional Slope Method {cont.)

Entry MA for, 90-92, 9 1 , 9 2
Exit MA for, 90-92, 9 1 , 9 2
exit techniques for, 159-160
filtering in, 252-253, 2 5 2 ( 0 , 2 5 3 ( 0
MAE/MFE and, 149
matrix of values for, 89-90, 90
modifications to, 159-160
money management in, 313-319, 3 1 4 ( 0 ,
316,316(0,317,318,319(0
out-of-sample testing of, 95-97, 9 6 , 9 7
performance summaries for, 9 3 ( 0 , 9 4 ( 0 ,
161-165, 161-165(0
ratio of profit factor and standard
deviation in, 89
STD/MAE in, 131, 141
testing of, 88-89, 95-97, 9 6 , 9 7
trend strength vs., 95-97, 9 6 , 9 7
distribution area in market analysis, 263, 264
distribution curves {See also market analysis;
trends in markets), 187-221,188
accumulation area in, for market analysis,
263, 264
asymmetric distribution in, 188-189
bell curve distribution in, 187-188,188
Chebychef's theorem in, 191
distribution area in, 263, 264
distribution curves in, 187-188, 188
Gold Digger II system in, 191-197
kurtosis in, 189
leptokurtic distribution curves in, 188-189
normal distribution curves in, 187-188,
188

out-lyers and, 190

platykurtic distribution curves in, 189
skew in, 189-190
standard deviation in, 188, 190-191
symmetric distribution in, 188
diversification
in future contract data, 4-5, 38-41, 3 9 , 4 0
in money management, 282
risk vs., 272, 279-280, 279
DMI-plus/DMI-minus lines, 234
dollar value of average loser, 29
dollar value of average trade, 29
dollar value of average winner, 29
dollar value of largest drawdown, 30
dollar value of largest loser, 29
dollar value of largest winner, 29
drawdown, 23-25, 160, 161,223,272,273, 343
closed trade drawdown (CTD), 5, 23, 121,
122, 123, 134-135,135, 136, 208,
209, 343
depth of historical drawdowns, 23, 24

drawdown {cont.)

dollar value of largest drawdown, 30

in efficient trades, 121-141
end trade drawdown (ETD), 6, 23, 122,
123, 131-134,132, 133, 138-139,
140, 157-158, 158(0, 170, 179,
181(0, 343
favorable excursion (FE) level (MAEfe),
156-157, 156(0, 168-177,169,
169-176(0, 179, 180(0
flat time, 25
largest percentage drawdown, 29
maximum adverse excursion (MAE), 23,
123-124,124, 130-131,131,
135-136,135,136, 141, 143-166,
167, 177-186,179, 208-209, 216,
343
maximum favorable excursion (MFE), 23,
123, 131-134, 132-134, 136,
138-139,138, 140, 141, 143-166,
144, 167, 208-209,216, 343
maximum intraday drawdown in, 4-5,
5-6, 23
Meander system and, 123
open trade drawdown (OTD), 5, 6, 23
run ups, 25
start trade drawdown (STD), 6, 23,
121-124,124, 130-131,131,
135-136,135, 136, 242, 343
total equity drawdown (TED), 5, 23,
121-122,122, 343
TradeStation and, 123
drift
exit techniques and, 70
in market trends, 87-88
Dynamic Breakout System (DBS), 97-107,
168-177,169, 169-176(0,319,
320(0
code for, 98-99
directional slope method vs., 105, 107,
174-175, 174(0
end trade drawdown (ETD) in, 170
exit techniques in, 101, 168
filtering in, 242-247, 243,244, 245,
245(0, 246, 247(0, 248, 249(0
lookback period in, 99-101,100
MAE in, 168
MAEfe in, 168
money management in, 319, 3 2 0 ( 0
performance summaries for, 106(0,
172-177, 173-176(0
profit analysis for, 170
Standard Deviation Breakout system
(SDB) vs., 180-181

348

Index

Dynamic Breakout System (cont.)

stop loss in, 168
stops in, 99-101,100
test of, 102-107, 103-106(0
time-based stop in, 171-172
trade statistics summary for, 103-105,
103(0, 104(0, 105(0
trailing stops in, 168
triggers of volatility for, 101-102
turtle traders in, 97-98, 107
volatility in, good vs. bad, 99-101,100

EasyLanguage, TradeStation, 1
Eckhardt, Bill, 97-98
efficient frontier, 282
Efficient Market Hypothesis (EMH), 257,
258, 282, 341
efficient trading, 119-142
bar count/calculation in TradeStation,
129-130
closed trade drawdown (CTD), 121, 122,
123, 134-135,135, 136
directional slope system for, 124-129
drawdowns for, 121-141
end trade drawdown (ETD), 122, 123,
131-134,132, 133, 134, 138-139,
140

final profit, 138-139,140

long trades, TradeStation equations for,
119, 120,120
maximum adverse excursion (MAE) in,
123-124,124, 130-131,131,
135-136,135, 136, 141
maximum favorable excursion (MFE) in,
123, 131-134,132, 133, 134, 136,
138-139, 138, 140, 141
Meander system and, 123
overriding filter in, 134
pivotal resistance level/support levels in,
123-124
RINA system and, 130
short trades and, TradeStation equations
for, 119-121
start trade drawdown (STD) in, 121-124,
124, 130-131,131, 135-136,135,
136

STD/MAE in, 138-139,140

total equity drawdown (TED), 121-123,122
TradeStation code for, 119
end trade drawdown (ETD), 6, 23, 122, 123,
131-134,132, 133, 138-139,140,
157-158, 158(0, 170, 179, 181(0,
343

entry points, 274

in Black Jack, 216-217, 220-221
closed trade drawdown (CTD) in, 122,
123, 134-135,135, 136
in Black Jack, 216-217, 220-22lend trade
drawdown (ETD) in, 122, 123,
131-134, 132, 133, 134, 138-139,
140

money management and, 271

start trade drawdown (STD) in, 122,
123-124,124, 130-131,131,
135-136,135, 136, 135
total equity drawdown (TED)in, 123
"equally well" vs. "equally profitably"
defined, 52
Equation, The, in money management, 281-282
Equis (See MetaStock)
equity, underwater equity curve, 23
equity highs/tops, 21, 160, 161
estimated TWR (ETWR), 281
Euro currency, 55
evaluating performance, 1-12
account size vs. return on account, 6-7
average dollar value per trade, 20
average percentage profit, 28
average percentage winner, 28
average trade value, 7
average winning and losing trades, 9
bars per trade, average number of, 9-10
closed trade drawdown (CTD), 5, 23
commission, 17
computer software for, 1-2
cumulative profit, 21, 22, 28, 29
diversity in, 4-5
dollar value of average loser, 29
dollar value of average trade, 29
dollar value of average winner, 29
dollar value of largest drawdown, 30
dollar value of largest loser, 29
dollar value of largest winner, 29
drawdown, 23-25
end trade drawdown (ETD), 6,23
equity tops, 21
flat time, 25
futures contract data, 31-44
gross profit/loss, 8
largest percentage drawdown, 29
largest percentage loser, 29
largest percentage winner, 28
largest winning and losing trades, 7-8
largest winning/losing trade, 20-21
maximum adverse excursion (MAE), 23
maximum consecutive winners/losers,
10-11

Index

349

evaluating performance (cont.)

maximum favorable excursion ( M F E ) , 23
maximum intraday drawdown in, 4 - 6 , 23
measures of performance in, 3 - 3 0
MetaStock program for, 12
number (winning/losing) trades, 9 - 1 0
number of losers, 28
number of trades, 29
open trade drawdown ( O T D ) , 5, 6, 23
percent profitable trades, 10-11
percentage calculation of performance,
13-14, 14-16
percentage of losers, 29
percentage of winning trades, 29
performance summary report,
TradeStation, 6
price vs. value in, 3 - 4
profit factor, 8 - 9 , 30
profit per trade, 1 7 - 1 9 , 1 9
return on account, 6 - 7
run ups, 25
slippage, 17
standard deviation calculation, 2 1 , 28, 29
start trade drawdown ( S T D ) , 6, 23
total equity drawdown ( T E D ) , 5, 23
total net profit in, 45
TradeStation program for, 1-2
underwater equity curve, 23
exit techniques, 6 9 - 8 1 , 117-118, 274,
342-344
200-days moving average in, 70
in Black Jack system, 6 9 - 8 1 , 7 5 - 8 0 ( 0 ,
205,216,216(0,217,219
closed trade drawdown ( C T D ) and, 122,
123, 134-135, 135, 136
in Directional Slope Method, 9 0 - 9 2 , 91,

92
drift in, 70
in Dynamic Breakout System ( D B S ) , 101,
168-177,169, 1 6 9 - 1 7 6 ( 0
end trade drawdown ( E T D ) and, 122, 123,
1 3 1 - 1 3 4 , 1 3 2 , 133,134, 1 3 8 - 1 3 9 ,

140
in Gold Digger I I , 191
long-term exits and, 167-186
in Meander system, 219, 220
money management and, 2 7 1 , 282
multiple markets and, 7 1 - 7 2
profit target for, 70
in Standard Deviation Breakout system
( S D B ) , 177-186, 177
start trade drawdown ( S T D ) in, 1122-124,
124, 1 3 0 - 1 3 1 , 1 3 1 , 3 5 - 1 3 6 , 135,

136

exit techniques (cont.)

stop loss for, 70, 7 7 - 7 8 , 7 8 ( 0
time-based stop for, 70
time-based trailing stop in, 7 6 - 7 7
total equity drawdown ( T E D ) in, 123
trailing stop for, 70
trend filter in, 70
triggers for, 70
uncertainty measures in, 77

fading a system, 167

Fama, Eugene, 257
favorable excursion level ( M A E f e ) , 156-157,
156(0, 1 6 8 - 1 7 7 , 1 6 9 , 1 6 9 - 1 7 6 ( 0 ,
179, 180(0
Standard Deviation Breakout system
( S D B ) , 179, 180(0
Dynamic Breakout System ( D B S ) ,
168-177,169, 1 6 9 - 1 7 6 ( 0
filter techniques, 117-118, 2 2 7 - 2 4 0 , 267
200-day moving average in, 228, 229
A D X indicator in, 234, 2 4 1 - 2 4 2 , 250,
252-253
average true range method in, 233, 241
Black and Scholes pricing formula and,
233
in Black Jack system, 2 3 9 - 2 4 0 , 2 4 0 ( 0
Black-76 pricing formula and, 233
in Directional Slope System, 2 5 2 - 2 5 3 ,
252(0, 253(0
DMI-plus/DMI-minus lines in, 234
in Dynamic Breakout System ( D B S ) ,
2 4 2 - 2 4 7 , 243, 244, 245, 2 4 5 ( 0 ,
246, 2 4 7 ( 0 , 248, 2 4 9 ( 0
in Gold Digger system, 2 3 6 - 2 3 8 ,
237-239(0
high/low breakout indicator in, 231
high-probability, 225
kurtosis in, 235
long-term volatility, 2 4 1 - 2 5 4
lookback periods for, 2 3 5 - 2 3 6 , 2 3 6 ( 0
maximum adverse excursion ( M A E ) in,
242
in Meander system, 199, 2 3 8 - 2 3 9 , 2 4 0 ( 0
moving averages, 2 3 4 - 2 3 5
on-balance volume ( O B V ) indicator in,
229, 230, 2 3 1 - 2 3 2 , 236
oscillators in, 234
overriding, 134
pricing formulas and, 233
profit factor and, 242
RSI indicators in, 234
in short-term systems, 2 2 8 - 2 3 6

350

Index

filter techniques (cont.)

skew in, 235
standard deviation in, 235
in Standard Deviation Breakout (SDB)
system, 233, 247-252, 249-251(0
start trade drawdown (STD) in, 242
T-bill ratio, 231-232
TradeStation code for, 232-233
true range method in, 233
volatility and, 233, 234, 241-254
fiscal policies and market trends, 259-260
fixed fractional money management method,
53, 273,339
flat time, 21, 25
flow-of-funds analysis, 262-263, 264, 266
Freeburg, nelson, 70
futures contract data, 31-44
back-adjustment method of splicing in, 31
best-fit regression lines, 35
point-based adjusted method of
splicing, 31, 32-37
ratio-adjusted method of splicing, 31,
37-38
best-fit regression lines in, 35
multimarket portfolios in, 38-41, 3 9 , 4 0
nonadjusted data in, 31-32
nonadjustment method of splicing
contracts in, 31
perpetual adjustment method of splicing
in, 31
perpetual contracts in, 41-44, 4 2 , 4 3
point-based adjusted method of splicing
in, 31
ratio-adjusted method of splicing in, 31,
37-38
splicing contracts in, 31
volatility in, 35

gain vs. risk, gain vs. risk in, 272, 273,

279-280, 282
geometric mean, 275, 281
global market structures, 258-259
Gold Digger systems, 59-62, 6 0 ( 0 , 6 2 ( 0 ,
191-197
Black Jack vs., 218-221, 218-221(0
Chebchev's theorem in, 194
exit techniques in, 191
filtering in, 236-238, 237-239(0
kurtosis in, 192-193, 193(0, 194, 196,
196(0,198
leptokurtic nature of, 193
mean in, 193
median in, 192-193, 193(0

Gold Digger systems (cont.)

performance summary of, 192, 192(0,
195-197, 195-197(0
profit factor in, 195,218
robustness test of, altering inputs for,
334-336, 336(0, 3 3 7 ( 0
skew in, 192-193, 193(0
standard deviation in, 192, 194, 218
stop loss in, 191
stops in, 195
TradeStation code for, 194-195
gross domestic product (GDP), 259, 261
gross profit/loss, 8

halfway flags in market analysis, 265

high/low breakout indicator, 231
high-probability filters, 225
holding period return (HPR), 272, 275, 276,
281

indicator piling, 227-228

indicators, 51, 227
ADX indicator, 234, 241-242, 250,
252-253
average true range indicator, 241
DMI-plus/DMI-minus lines, 234
high/low breakout indicator, 231
market microstructure indicators, 263
on-balance volume (OBV) indicator, 229,
230, 231-232, 236, 263
oscillators, 234
RSI indicators, 234
sentiment indicators in, 262
T-bill ratio filter, 231-232
interest and amortization in market trends,
260-261,264

kurt function, Excel, 189

kurtosis, 189, 223
in Black Jack, 211, 211, 2 1 2 ( 0 , 2 1 3 ( 0
filtering and, 235
in Gold Digger II, 192-193, 193(0, 194,
196,196(0,198
in Meander system, 199, 201, 203
largest percentage drawdown, 29
largest percentage loser, 29
largest percentage winner, 28
largest winning and losing trades, 7-8, 7
largest winning/losing trade, 20-21, 20

Index

351

Latour, Bruno, 255, 258

least square regression
maximum adverse excursion (MAE), 150,
152.152
maximum favorable excursion (MFE),
152.153
leptokurtic distribution curves, 188-189, 193
Livermore, Jesse, 255
long-term investing systems, 343-344
ADX indicator in, 241-242, 250, 252-253
average true range indicator in, 241
closed trade drawdown (CTD) in, 121
Directional Slope System for, 252-253,
252(0,
253(0,313-319,314(0,
316,316(0, 317,318,319(0
Dynamic Breakout System (DBS) for,
242-247, 243,244,245, 2 4 5 ( 0 ,
246, 2 4 7 ( 0 , 248, 249(0, 319, 320(0
exits for, 167-186
maximum adverse excursion (MAE) in,
242
momentum changes in market trends and,
265-266, 265, 266
money management in, 283,
305-313
optimal fin, 283, 306, 318, 319
profit factor in, 242
Standard Deviation Breakout (SDB)
system for, 247-252, 249-251(0,
305-313, 306-309(0, 308,311,
311,312,313(0,314,315
start trade drawdown (STD) in, 121, 242
total equity drawdown (TED) in, 121-122,
122

TradeStation equations for, 119, 120,120

volatility filters and, 241-254
lookback period
Dynamic Breakout System, 99-101,100
filtering, 235-236, 2 3 6 ( 0
losses/losing trades
average dollar value per trade in, 20
average losing trade in, 148,148, 149
average winning and losing trades in, 9
bars per trade, average number of, 9-10,
dollar value of average loser in, 29
dollar value of average trade in, 29
dollar value of largest loser in, 29
gross profit/loss in, 8
largest percentage loser in, 29
largest winning and losing trades in, 7-8,
20-21
MAE/MFE and, 149-150,151
maximum consecutive winners/losers in,
10-11

losses/losing trades (cont.)

number (winning/losing) trades in, 9-10
number of losers in, 28
percent profitable trades in, 10-11
percentage of losers in, 29
profit factor in, 8-9

MACD indicator, 51
macroeconomic-oriented markets, 83
margin to equity ratio, 301
market analysis (See also distribution curves;
trends in markets), 255-266,
342-344
academic theories of, 256-257
accumulation area in, 263, 264
adoption-diffusion model in, 263
arbitrage principles and, 257
beta value and, 327
Capital Asset Pricing Model (CAPM) in,
257,258
Commitment of Traders Report in, 262
Commodity Futures Trading Commission
(CFTC) and, 262
Crash of 1929, 261,262
creation-of-funds analysis in, 262-263,
266
credit policies and, 260
demographics and, 259-260
distribution area in, 263, 264
Efficient Market Hypothesis (EMH) and,
257, 258
fiscal policies and, 259-260
flow-of-funds analysis in, 262-263, 264,
266
global market structures and, 258-259
gross domestic product (GDP) in, 259,
261
halfway flags in, 265
interest and amortization in, 260-261, 264
market microstructure indicators in, 263
momentum changes in, 265-266, 265,266
money supply (Ml, M2) in, 259
net present value (NPV) in, 259
New Economy and, 261
on-balance volume (OBV) indicator in,
263
Pearson correlation coefficient in,
326-329
portfolio theory in, 257
price vs. value in, 256, 257
random walk hypothesis (RWH) and, 257
real vs. financial markets, outcomes in,
261

352

Index

market analysis (cont.)

rule-based trading in, 256, 258
sentiment indicators in, 262
social factor influencing market trends
and, 257
speculation and, rules of, 256
technical analysis in, 262
wealth of markets and, 256
market microstructure indicators, 263
market-specific vs. diversified portfolios in,
47
Markowitz, Harry, 257
Mathematics of Money Management, The,
269, 340
maximum adverse excursion (MAE), 23,
123-124,124, 130-131,131,
135-136,135,136, 141, 143-166,
167, 177-186,179, 208-209, 216,
343
in average losing trade, 148, 148,149
average percentage move for each bar in,
146-147,148
average profit calculation in, 153-154,
154(0
in Black Jack, 208, 209,216
definition of, 143, 145
directional slope in, 149, 150
in Dynamic Breakout System (DBS),
168-177,169, 169-176(0
end trade drawdown (ETD) in, 157-158,
158(0
favorable excursion (FE) level (MAEfe)
in, 156-157, 156(0
filtering and, 242
least square regression in, 150, 152,152
lifespan and development of trade using,
145
market test for, 145, 146,147
regression equation for, 147-148, 148
rules for, 148
in Standard Deviation Breakout system
(SDB), 177-186,179
standard deviation in, 146-147, 148,
154-155, 155(0
start trade drawdown (STD) in, 145
stop loss and, 152-153, 155
time-based stop and, 157-159, 158(0,
159(0
trailing stop and, 157-159, 158(0, 159(0
trajectory analysis and, 145
winners vs. losers and, 149-150, 151
Maximum Adverse Excursion, Analyzing Price
Fluctuations..., 23, 123
maximum consecutive winners/losers, 10-11

maximum favorable excursion (MFE), 23,

123, 131-134, 132-134, 136,
138-139,138, 140, 141, 143-166,
144, 167, 208-209,216, 343
average losing trade in, 148,148,149
average percentage move for each bar in,
146-147, 148
in Black Jack, 208, 209, 216
definition of, 143, 145
directional slope in, 149,150
least square regression in, 152, 153
lifespan and development of trade using,
145
market test for, 145,146, 147
regression equation for, 147-148,148
rules for, 148
standard deviation in, 146-147,148,
154-155, 155(0
stop loss and, 145, 155
trailing stop and, 145
trajectory analysis in, 145
winners vs. losers in, 149-150,151
maximum intraday drawdown in, 4-5, 5-6, 23
mean
in Gold Digger II, 193
in Meander system, 199
Meander system, 64-69, 67, 6 8 ( 0 , 6 9 ( 0 , 131
Black Jack combined with, 219-221,
219-221(0
drawdowns and, 123
ETD/MFE in, 132-134,132,133,134
exit techniques in, 219, 220
filtering in, 199, 238-239, 2 4 0 ( 0
kurtosis in, 199, 201, 203
maximum favorable excursion (MFE) in,
138-139,139
mean in, 199
money management in, 287-290, 291,
293, 295-302, 2 9 8 ( 0 , 303(0, 306
out-of-sample testing for, 203
performance summary of, 202-203, 2 0 3 ( 0
profit factors in, 199, 203
skew in, 199, 203
standard deviation in, 199
stop loss in, 200-201, 2 0 1 ( 0 , 2 0 2 ( 0 , 202
summarized life of trade in, 139,140, 141
time-based stop in, 219
weekly data for, 197-204, 199(0, 200
measures of performance, 3-30
account size vs. return on account, 6-7
average dollar value per trade, 20
average percentage profit, 28
average trade value, 7
average winning and losing trades, 9

Index

353

measures of performance (cont.)

bars per trade, average number of, 9-10
closed trade drawdown (CTD), 5, 23
commission, 17
cumulative profit, 21, 22
diversity in, 4-5, 4
drawdown, 23-25
end trade drawdown (ETD), 6, 23
equity tops, 21
flat time, 25
gross profit/loss, 8
largest winning and losing trades, 7-8,
20-21
maximum adverse excursion (MAE), 23
maximum consecutive winners/losers,
10-11
maximum favorable excursion (MFE), 23
maximum intraday drawdown in, 4-6, 23
number (winning/losing) trades, 9-10
open trade drawdown (OTD), 5, 6, 23
percent profitable trades, 10-11
percentage calculation of performance,
13-14, 14-16
performance summary report,
TradeStation, 6
price vs. value in, 3-4
profit factor, 8-9
profit per trade, 17-19,19
return on account, 6-7
run ups, 25
slippage, 17
standard deviation calculation, 21, 28
start trade drawdown (STD), 6, 23
total equity drawdown (TED), 5, 23
total net profit in, 4-5
underwater equity curve, 23
median, 192-193, 193(0
Merton, Robert, 257
MetaStock, 1-2, 7
Miller, Merton, 257
Modigliani, Franco, 257
momentum changes in market trends, 265-266,
265,266
momentum indicator, 51
money management, 47, 269, 271-320,
339
average profit multiplier (APM) in,
281-282
capital asset pricing model (CAPM) in,
282,341
correlations in, 274
cumulative profit in, 281
curve fitting, 282-283
dependencies, 274

directional slope strategy, 313-319, 3 1 4 ( 0 ,

316,316(0,317,318,319(0
diversification, 272, 279-280, 282
drawdown, 272, 273
in Dynamic Breakout System (DBS), 319,
320(0
efficient frontier in, 282
efficient market hypothesis (EMH) in,
282, 341
entry points and, 271, 274
estimated TWR (ETWR) in, 281
exit techniques and, 271, 274, 282
fixed fractional strategy for, 273, 339
gain vs. risk in, 272, 273, 279-280, 282
geometric mean in, 275, 281
holding period return (HPR) in, 272, 275,
276, 281
key points in, 272
in long-term systems, 283, 305-313
margin to equity ratio in, 301
in Meander system, 287-290, 291, 293,
295-302, 298(0, 303(0, 306
optimal fin, 273, 275, 277, 279-281, 279,
280, 282, 283-285, 285, 286,
304-306, 318, 319, 340-342
in portfolios vs. one-market systems,
271-272
practical applications for, 282-285
price fluctuations in, 272
profit/loss in, 275
risk, 272, 273, 279-280, 282
rule-based trading systems, 274-275
runs test in, 274
in short-term systems, 283, 285-305, 287,
291,293, 295-302, 2 9 8 ( 0 , 303(0,
306
in Standard Deviation Breakout (SDB),
305-313, 306-309(0, 308,311,
311(0,312,313(0,314,315
standard deviation in, 281-282
terminal wealth relative (TWR) in,
272-273, 275, 276, 277, 278, 279,
280, 280, 283-284, 285, 286
The Equation for, 281-282
type I errors in, 274, 284-285
type II error in, 274, 284-285
money supply (Ml, M2), 259
moving averages, 228
filter using, 234-235
in Standard Deviation Breakout system
(SDB), 177-178
in tracking trends, 86-97
multimarket portfolios, future contract data,
38-41,39,40

Index

354

net present value (NPV) in, 259

New Economy and, 261
nonadjustment method of splicing contracts
in, 31
normal distribution curves, 187-188,188
normalization of market movements, 64
number (winning/losing) trades, 9-10
number of losers, 28

Omega Research (See TradeStation)

on-balance volume (OBV) indicator, 229,
230, 231-232, 236, 263
open trade drawdown (OTD), 5, 6, 23
optimal f, 273, 275, 277, 279-285, 279,280,
285, 286, 340-342
in long-term systems, 306, 318, 319
in short-term systems, 304-305
oscillator-type indicators (See also Bollingerbands), 107, 234
out-of-sample testing, 48, 331-334,
332-336(r)
outlyer trades, 21(f), 190
overriding filter, 134

picking tops and bottoms (cont.)

rate-of-change indicator in, 51
RSI indicator in, 51
stochastic indicator in, 51
technical analysis for, 51
in testing a system, 52
piling, indicators, 227-228
pivot-points in, 53
platykurtic distribution curves, 189
plus and minus DMI indicator, 51
point-based adjusted method of splicing, 31,
32-37
portfolio composition, 269, 321-330, 339
Charity I portfolio, 322-324, 322, 3 2 3 ( 0 ,
324

Charity II portfolio, 325, 3 2 6 ( 0 , 3 2 7

Charity III portfolio, 328-330, 3 2 9 ( 0 ,
330(0
correlations in, 324-325, 3 2 5 ( 0
covariances in, 324-325, 3 2 5 ( 0
Pearson correlation coefficient, 326-329
total equity contribution I, 321-324, 3 2 2 ,
323(0, 324
total equity contribution II, 325-330
Portfolio Management Formulas, 269

Patterns in the Dark, 258

Pearson correlation coefficient, 326-329

percent profitable trades, 10-11
percentage calculation of performance,
13-14, 14-16
percentage of losers, 29
percentage of winning trades, 29
percentage-based analysis, 53
performance summary report, TradeStation, 6
perpetual adjustment method of splicing, 31
perpetual contracts, 41-44, 4 2 , 4 3
Peters, Edgar, 255, 258
picking tops and bottoms, 51-82
in Black Jack, 53
Bollinger bands in, 53
in building a system, 52
catching market moves in, 52-53
data selection for, 52-53
"equally well" vs. "equally profitably"
defined, 52
fixed fractional investing and, 53
indicators of, 51
MACD indicator in, 51
momentum indicator in, 51
optimal f in, 53
percentage-based analysis in, 53
pivot-points in, 53
plus and minus DMI indicator in, 51

portfolio theory, 257

PowerEditor, TradeStation, 1
price fluctuations, 272
price vs. value, 3-4, 256, 257
pricing formula
average true range method, 233
Black and Scholes pricing formula, 233
Black-76 pricing formula, 233
in filtering, 233
standard deviation method, 233
true range method, 233
profit per trade, 17-19, 19
profit/profit factor, 8-9, 30, 160, 161, 275
average percentage profit in, 28
average profit per trade over time in,
85-86, 8 6 ( 0
in Blackjack, 218
commission vs., 17
cumulative profit in, 21, 2 2 , 28, 29
equity tops in, 21
exit techniques in, profit target and, 70
in filtering, 242
in Gold Digger II, 218
gross profit/loss in, 8
MAE/MFE in, average profit calculation
in, 153-154, 154(0
in Meander system, 199, 203
percent profitable trades vs., 10-11
profit factor defined in, 8-9, 30

Index

355

profit/profit factor (cont.)

profit per trade in, 17-19,19
slippage vs., 17
profit target, 70
in Black Jack, 213-214, 214, 215-216
in Standard Deviation Breakout system
(SDB), 180
Pruden, Henry, 255, 263

random entries (See distribution curves)

random number generator, 223
Black Jack, 204-205
random walk hypothesis (RWH), 257
Raschke, Linda B., 70
rate-of-change indicator, 51
ratio-adjusted method of splicing, 31, 37-38
real vs. financial markets, outcomes in market
trends, 261
regression equation, MAE/MFE, 147-148,148
regular moving averages, data mining, 63
return on account, 6-7
RINA system, 119, 130, 242
risk, 272, 273, 279-280, 282
robustness test, altering inputs for, 334-336,
336(/), 337(0
Rothschild, Nathan, 255
RSI indicator, 51, 234
rule building, 116
rule-based trading, 256, 258, 274-275
run ups, 25
runs test, 274

Scholes, Myron, 257

sentiment indicators, 262
Sharpe, William, 257
short-term investing systems, 47-49, 343-344
ADX indicator in, 234
average profit per trade over time in,
85-86, 8 6 ( 0 , 85
Black Jack system in, 239-240, 2 4 0 ( 0
closed trade drawdown (CTD) in, 121
DMI-plus/DMI-minus lines in, 234
filtering in, 228-236
Gold Digger system in, 236-238, 237-239(0
kurtosis in, 235
lookback periods for, 235-236, 2 3 6 ( 0 , 235
margin to equity ratio in, 301
Meander system in, 238-239, 2 4 0 ( 0 ,
287-290, 291, 293, 295-302,
298(0, 303(0, 306
momentum changes in market trends and,
265-266, 265, 266

short-term investing systems (cont.)

money management in, 283, 285-305,
287,291,293, 295-302, 2 9 8 ( 0 ,
303(0, 306
moving averages filter in, 234-235
on-balance volume (OBV) indicator in,
229, 230, 231-232, 236
optimal fin, 283, 304-305
oscillators in, 234
pricing formulas in, 233
RSI indicators in, 234
skew in, 235
standard deviation in, 235
start trade drawdown (STD) in, 121
T-bill ratio filter in, 231-232
total equity drawdown (TED) in, 121-122,
122

TradeStation equations for, 119-121

trends in markets and, 85
volatility and, 233, 234
skew, 189-190, 223
in Black Jack, 211, 212, 2 1 2 ( 0 , 2 1 3 ( 0
filtering and, 235
in Gold Digger II, 192-193, 193(0
in Meander system, 199, 203
slippage, 17
social factor influencing market trends, 257
Soros, George, 255
speculation, rules of, 256
splicing contracts in, 31
standard deviation, 21, 28-29, 188, 190-191
in Black Jack, 209, 210, 216, 217, 218
Chebychef's theorem in, 191, 194
in filtering, 233, 235
in Gold Digger II, 192, 194, 218
MAE/MFE, 154-155, 155(0
maximum adverse excursion (MAE) and
maximum favorable excursion
(MFE), 146-147, 148
in Meander system, 199
money management and, 281-282
Standard Deviation Breakout (SDB),
107-113,108,110, 111, 177-186,
178, 305-313, 306-309(0, 308,
311,311(0,312,313(0,314,315
Dynamic Breakout System(DBS) vs.,
180-181
end trade drawdown (ETD) in, 179, 181(0
exit techniques for, 181-182
favorable excursion (MAEfe) in, 179,
180(0
filtering in, 247-252, 249-251(0
maximum adverse excursion (MAE) in,
178-179, 179

356

Index

Standard Deviation Breakout (cont.)

maximum trade length in, 181, 183(0
money management for, 305-313,
306-309(0, 308,311, 3 1 1 ( 0 , 312,
313(0,314,315
moving averages and exits in, 177-178
out-of-sample testing in, 331-334,
332-336(0
performance summary of, 182-186,
183-186(0
profit target for, 180
stop loss in, 178
testing of, 177
start trade drawdown (STD), 6, 23, 121-124,
124, 130-131,131, 135-136,135,
136, 242, 343
stochastic indicator, 51
stop loss, 70, 77-78, 7 8 ( 0 , 132
in Black Jack, 205-206, 209,
211-212
in Dynamic Breakout System (DBS),
99-101,100, 168-177,169,
169-176(0,169, 169-176(0
in Gold Digger II, 191, 195
maximum adverse excursion (MAE),
152-153, 155
maximum favorable excursion (MFE),
145, 155
in Meander system, 200-201, 2 0 1 ( 0 ,
202(0, 202
in Standard Deviation Breakout system
(SDB), 178
strategies vs. systems, 339
surface chart, Black Jack, 209, 210
Sweeney, John, 23, 123, 141, 143, 167
symmetric distribution, 188
system concepts, 47-50
arbitrage, 47
building a system, 47, 48
market-specific vs. diversified portfolios
in, 47
money management strategies in, 47
out-of-sample testing in, 48
picking tops and bottoms, 51-82
short- and long-term methodologies
defined for, 47-49
testing a system, 47, 48
trends, 47
volatility in, 47
systems vs. strategies, 339

T-bill ratio filter, 231-232

T-bills, perpetual contracts vs., 41-44, 4 2 , 4 3

T-bond, 55
technical analysis of markets, 51, 262
Technical Analysis of Stocks and
Commodities, 263

terminal wealth relative (TWR), 272-273,

275, 276, 277, 278, 279, 280, 280,
283-284, 285,286
testing a system, 47, 48, 52, 59
out-of-sample testing, 331-334,
332-336(0
robustness test, altering inputs for,
334-336, 336(0, 337(0
time-based stop, 70, 157-159, 158(0,
159(0
in Black Jack, 205, 214-215, 215, 219
in Dynamic Breakout System(DBS),
171-172
maximum adverse excursion (MAE) and,
157-159, 158(0, 159(0
in Meander system, 219
time-based trailing stop, 76-77
total equity contribution I, 321-324, 322,
323(0, 324
total equity contribution II, 325-330
total equity drawdown (TED), 5, 23,
121-122,122, 343
total net profit in, 4-5
tracking market moves, 56-62, 57(0, 58(0
trade length, Black Jack, 205-206
TradeStation, 1-2, 1
average percentage profit in, 28
average percentage winner in, 28
average trade value in, 7
bar count/calculation in, 129-130
Black Jack code in, 69-81, 75-80(0,
206-208
Bollinger-bands indicator in, 107-113,
108,110,111, 113(0

cumulative profit in, 28, 29, 160, 161

data mining calculations and, 58-59
directional slope method in, 87-97, 90-94,
93(0, 94(0
dollar value of average loser in, 29
dollar value of average trade in, 29
dollar value of average winner in, 29
dollar value of largest drawdown in, 30
dollar value of largest loser in, 29
dollar value of largest winner in, 29
drawdown calculations in, 25
drawdowns in, 123, 160-161
Dynamic Breakout System (DBS) code,
98, 168-177,169, 169-176(0
efficient trades in, 119
equity highs in, 160, 161

Index

357

TradeStation (cont.)
evaluating performance, code for, 11-12,
26-28
exit techniques in, 160
filtering in, 232-233
flat time calculation in, 26
futures contract data and, 31-44
Gold Digger systems code, 59-62, 6 0 ( 0 ,
6 2 ( 0 , 191-197
largest percentage drawdown in, 29
largest percentage loser in, 29
largest percentage winner in, 28
long trades, equations for, 119, 120,120
maximum intraday drawdown in, 5-6
Meander system code, 64-69, 67, 6 8 ( 0 ,
69(0
number of losers in, 28
number of trades in, 29
open trade and entry price in, 129-130
percentage of losers in, 29
percentage of winning trades in, 29
performance summary report in, 6, 2 8 ( 0
profit factor in, 30
profit in, 160, 161
random number generator in, 204-205, 223
short trades, equations for, 119-121
standard deviation in, 28, 29
trend-tracking calculations in, 58-59
Trade Works Software, 223
trailing stop, 70, 132, 157-159, 158(0, 159(0
in Black Jack, 214-215, 215
in Dynamic Breakout System (DBS),
168-177,169, 169-176
maximum adverse excursion (MAE),
157-159, 158(0, 159(0
maximum favorable excursion (MFE), 145
trajectory analysis, MAE/MFE, 145
trend filter, 70
trends in markets (See also distribution
curves; market analysis), 47, 55-56,
85-113,343-344
200-day moving average in, 87, 228, 229
absolute vs. percentage values in, 64
accumulation area in, 263, 264
average profit per trade over time in,
85-86, 8 6 ( 0
basic crossover signals in, 86-87
Bollinger-band type indicators and, 63,
107-113,108,110, 111, 113(0
calculations for, 58-59
Commitment of Traders Report and, 262
Commodity Futures Trading Commission
(CFTC) and, 262
Crash of 1929 and, 261, 262

trends in markets (cont.)

creation-of-funds analysis in, 262-263, 266
credit policies and, 260
decision to trade vs., 83-84
demographics and, 259-260
directional slope method in, 87-97, 90-94,
93(0, 94(0
distribution area in, 263, 264
dollar value of half average weekly true
range (AWTRO/AWTR) in, 83, 84
drift in, 87-88
Dynamic Breakout Systems and, 97-107
exit techniques and, 6981
filtering and, 228, 229
fiscal policies and, 259-260
flow-of-funds analysis in, 262-263,264,266
global market structures and, 258-259
gross domestic product (GDP) in, 259, 261
indicator piling in, 227-228
interest and amortization in, 260-261, 264
long-term systems and, 343-344
market analysis and, 255-266
market microstructure indicators in, 263
Meander system and, 64-69, 67, 6 8 ( 0 , 6 9 ( 0
momentum changes in, 265-266, 265, 266
money supply (Ml, M2) in, 259
moving averages in, 86-97, 228
net present value (NPV) in, 259
New Economy and, 261
normalization of market movements in, 64
on-balance volume (OBV) indicator in,
229, 230, 231-232, 236, 263
oscillator-type indicators in, 107
real vs. financial markets, outcomes in, 261
regular moving averages and, 63
sentiment indicators in, 262
short- vs. long-term trading and, 85
short-term systems and, 343-344
social factors in, 257
Standard Deviation Breakout system in,
107-113,108,110,111, 113(0,
177-186
tracking market moves, 56-62, 5 7 ( 0 , 5 8 ( 0
turtle traders in, 97-98
volatility in, 83-84
triggers for exit techniques, 70
true range method, filtering, 233
turtle traders, 97-98, 107
type I errors, 274, 284-285
type II error, 274, 284-285

uncertainty measures, 77
underwater equity curve, 23

358

Index

Vince, Ralph, 53, 269, 281, 339-340

volatility, 47, 83-84
dollar value vs. percentage and, 84
Dynamic Breakout System, good vs. bad,
99-101,100
Dynamic Breakout System, triggers of,
101-102
filtering and, 233, 234, 241-254
future contract data and, 35
von Liechtenstein, Max, 255

wealth of markets, 256

weather-sensitive markets, 83
Werner, Richard, 255
winning trades
average dollar value per trade in, 20

winning trades (cont.)

average percentage winner in, 28
average winning and losing trades in, 9
bars per trade in, average number of in,
9-10
dollar value of average trade in, 29
dollar value of average winner in, 29
dollar value of largest winner in, 29
largest percentage winner in, 28
largest winning and losing trades in, 7-8
largest winning/losing trade in, 20-21
MAE/MFE and, 149-150,151
maximum consecutive winners/losers in,
10-11
number (winning/losing) trades in, 9-10
percent profitable trades in, 10-11
percentage of winning trades in, 29

ABOUT

THE

AUTHOR

Thomas Stridsman is a senior editor for Active Trader m a g a z i n e , trading consultant, and popular speaker at industry seminars a n d conferences. Previously, he wrote
for Futures magazine. Before moving to Chicago in 1997, Stridsman operated his
own Web-based trading advisory service in Sweden and w a s chair of the Swedish
Technical Analysts Federation.

Other books in The Irwin Trader's Edge Series

The Encyclopedia of Trading Strategies by Jeffrey Owen Katz and
D o n n a L. M c C o r m i c k
Technical Analysis for the

Trading Professional by Constance B r o w n

Agricultural Futures and Options by Richard D u n c a n

The Options Edge by William Gallacher
The Art of the Trade by R.E. M c M a s t e r