Under the Microscope: The Structure of the

Foreign Exchange Market∗

a
Michael J. Sager and Mark P. Taylorb†
a
Putnam Investments and University of Warwick.
b
University of Warwick and Centre for Economic Policy Research.

April 2005

Abstract
We provide a detailed, up-to-date description of the microstructure of
the foreign exchange market and of the behaviour of participant groups.
In the light of this, we highlight shortcomings in existing theoretical mod-
els of market interaction and present an outline alternative model that
marries theoretical prediction and current market practice.

Keywords: Foreign exchange market; microstructure; order flow.

JEL Classification: F31

∗ While remaining solely responsible for any remaining errors in this paper, the authors

are grateful to two anonymouse referees and to Suzanne Brant, Jack Crawford, Rich Lyons,
Michael Melvin, Dirk Morris, Carol Osler, as well as numerous market participants, for in-
sightful and often extensive comments on earlier drafts.
† Corresponding author: Professor Mark P. Taylor, Department of Economics, University

of Warwick, Coventry CV4 7AL, UK. E-mail: mark.taylor@warwick.ac.uk

1
1 Introduction
Following the breakdown of the Bretton Woods system in the early 1970s, the
foreign exchange market has been the subject of intense academic and practi-
tioner research. In broad terms, two particularly important stylised facts have
emerged from this body of evidence.
First, although the foreign exchange market is highly liquid,1 it appears to
depart significantly from the text-book efficient markets paradigm in the sense
that there are significant deviations from uncovered interest rate parity that
cannot be accounted for by stable models of risk premia (Taylor, 1995; Sarno
and Taylor, 2002) and also in the sense that it is an opaque market, since the
lack of a physical market place makes the process of price-information interac-
tion difficult to understand (Dominguez, 1999; Lyons, 2002b). Both of these
traits result from heterogeneities, including informational asymmetries, differ-
ing reaction speeds to information innovations, and diverse opportunity sets
and risk-return expectations, that exist amongst the various participant groups.
These heterogeneities imply the presence of persistent profit opportunities from
informed trading relative to an underlying benchmark return.
Second, although empirical evidence over the post-Bretton Woods period
suggests that fairly standard macroeconomic fundamentals–such as relative
monetary velocity–may influence the long-run behaviour of real and nominal
exchange rates (for surveys see Frankel and Rose, 1995; Froot and Rogoff, 1995;
Taylor, 1995; Sarno and Taylor, 2002), the quality of shorter-term exchange rate
models and forecasts continues to be an occupational hazard of the international
financial economist, as fundamental variables are poorly correlated with high
frequency exchange rate movements.
Largely motivated by these stylised facts, a growing literature on foreign
exchange market microstructure has emerged in recent years to suggest that the
quality of fundamental-based exchange rate forecasts can be improved by resort
to measures of foreign exchange order flow (Froot and Ramadorai, 2001; Lyons,
2002a; Evans and Lyons, 2002a,b).2 An important conclusion of this literature
is that order flow allows the wider market to learn about the private information
1 Consistent with Kyle (1985), we define a liquid market as one that exhibits the following

characteristics: tightness, so that bid-ask spreads for small transactions are tight; depth, so
that bid-ask spreads for large transactions are small and; resilience, so that deviations of the
spot rate from “fair value” should be corrected quickly. See Daníelsson and Payne (2001) for
a recent assessment of these criteria applied to the interdealer sector of the foreign exchange
market.
2 Order flow may be defined as transaction volume signed according to the initiator of the

trade (Lyons, 2001); positive for a buy order, negative for a sell order. It is distinct from
transaction volume, since the latter is a measure of trading activity between customers and
dealers, or within the interdealer market, over a given period and in a particular exchange
rate without indication of the direction of these transactions. Measured as signed transaction
volume, order flow therefore provides an indication of the relative strength of buy and sell
orders between, say, customers and dealers. In this way, order flow within particular investor
groups will not necessarily sum to zero, but can instead exhibit persistent trends–for example
if customers build a long position in a particular currency relative to an underlying strategic
benchmark position, then order flow in that currency from this participant group will be
positive and rising.

2
and trading strategies of better informed participants, and therefore represents
the conduit through which informational asymmetries become embedded within
market prices (Lyons, 1995; Bjønnes and Rime, 2001b).
The microstructure literature draws support and scepticism in equal mea-
sure (for more sceptical views, see Rogoff, 2002; Fisher and Hillman, 2003). Few
disagree that the central hypothesis of this literature provides an intuitive expla-
nation of the process of price discovery in the foreign exchange market. Dishar-
mony surrounds the assessment of the practical value of this literature and, from
the same quarter, discontent has also been expressed concerning the treatment
given in the microstructure literature to assimilating the true structure of the
foreign exchange market and the extent and motivation of interactions between
the various participant groups. To date, theoretical predictions of this literature
have been based upon highly stylised models of investor interaction (e.g. Evans
and Lyons, 2002a,b). Ultimately, it is only through a detailed understanding of
market structure and practice that one can fully appreciate the different impact
on high frequency exchange rate returns of various categories of order flow.
Foreign exchange market structure and participant group interactions have
altered substantially in recent years. Daily market turnover has increased in
the last three years (BIS, 2004), with the majority of this increase concentrated
upon the various customer groups in the market.3 The motivation for these flows
has also altered, and investors increasingly view the foreign exchange market
as a potential source of persistent returns that can provide important diversifi-
cation benefits when combined in a portfolio with returns to more traditional
assets, such as equities or bonds. The nature of interactions between customers
and dealers has also undergone important changes in recent years, including
rapid growth of electronic trading platforms–traditionally the preserve of the
interdealer market–and important changes in the tactical interplay between
customers and dealers.
The objective of this paper is to provide a comprehensive, up-to-date de-
scription of the structure of the foreign exchange market. In so doing, we marry
the main tenets of existing theoretical models with the practical reality of the
foreign exchange market.
The remainder of the paper is organised as follows. In the next section we
pass a microscope over the foreign exchange market and, in addition to dis-
cussing trends in daily market turnover and geographic organisation, describe
the various participant groups within the market and discuss the motivation
and behaviour of each group, as well as group interaction. We then examine
how closely both traditional asset-price exchange rate models and microstruc-
tural models reflect this structure. Drawing upon all of these aspects, we then
present a thumbnail sketch of an alternative model that we believe would pro-
vide a better approximation to the foreign exchange market and yield testable
implications. In a final section we draw some conclusions and offer suggestions
for future research.
3 The term customer encompasses asset management firms, hedge funds, commodity trading

advisors (CTAs), central banks, corporations and high net worth private individuals. For
expositional purposes, we exclude exchange rate transactions by tourists.

3
2 The Microstructure of the Foreign Exchange
Market
2.1 Basic Features
The foreign exchange market is a decentralised market in the sense that mar-
ket participants are generally separated from one another and transactions take
place through electronic media such as by telephone or through computer net-
works. This is in contrast to, say, the New York Stock Exchange where traders
physically interact with one another. Two implications of decentralisation are
fragmentation and lack of transparency. The foreign exchange market is frag-
mented in the sense that transactions may (and do) occur simultaneously or
near simultaneously in the market at different prices (Sarno and Taylor, 2001).
It is opaque–or lacks transparency–in the sense that the absence of a physi-
cal market place makes the process of price-information interaction difficult to
observe and understand (Dominguez, 1999; Lyons, 2002b).
The foreign exchange market is also the most liquid financial exchange in the
world. Daily market turnover is estimated at approximately $1.9 trillion, includ-
ing spot, forward and swap transactions (BIS, 2004). This compares to daily
turnover in the US government bond market of $500 billion (Federal Reserve
Bank of New York, 2005), and on the New York Stock Exchange of $46 billion
(NYSE, 2005). Although this estimate of daily turnover represents a substan-
tial increase compared with 2001 ($1.2bn; BIS, 2002), part of the improvement
reverses the decline reported between 1998 ($1.5 trillion) and 2001; nonethe-
less a longer-term upward trend in daily turnover is apparent.4 Including only
spot transactions, daily turnover is approximately $621 billion, emphasising the
continued growth in the use of derivative contracts over spot transactions; the
latter accounted for the majority of turnover in the late 1980s (BIS, 1999).5
Although the market is decentralised, it nevertheless has various physical
locations or trading centres throughout the world where many of the most im-
portant market participants–in particular the market-making dealers–tend to
concentrate. Thus, a foreign exchange transaction booked with the London of-
fice of a Japanese or American bank would be recorded as having taken place
in the London market, even if it was conducted over a computer network and
the customer was located elsewhere in the world. In this sense, London has
consolidated its position as the major trading centre for foreign exchange, with
32% of daily market turnover transacted in this location (BIS, 2002). There
are other important centres in New York and Tokyo, and smaller ones in Auck-
land, Sydney, Singapore, Hong Kong, Frankfurt and San Francisco. In terms
of composition, the US dollar is on one side of approximately 90% of all daily
transactions–this has changed little over the last fifteen years–followed by
4 For
instance, daily market turnover was estimated in 1989 at $590 billion (BIS, 1996).
5 Currencyoverlay specialists in asset management firms, as well as Hedge Funds and
CTAs, implement currency hedging decisions primarily through the use of one or three month
forward contracts. Forwards are preferred over spot or futures because they require no upfront
payment and have no associated margin calls.

4
the euro (37%), yen and sterling (approximately 20% each; BIS, 2004). Euro-
dollar is consequently the most liquid exchange rate, followed by yen-dollar and
sterling-dollar. These rankings are also little changed over the last fifteen years,
although the percentage of daily market turnover traded in sterling-dollar has
risen by a third during this period, at the expense of yen-dollar.

2.2 Market Participants: Dealers and Customers

In broad generic terms, there are two types of participants in the foreign ex-
change market: dealers and customers.6 Dealers account for the majority of
market liquidity, with approximately 59% of total daily market turnover occur-
ring within the interdealer market. But the predominance of this group has
been in decline for a number of years, and is down from over two-thirds in
the early 1990s (BIS, 1996, 2002). Customers contribute the remaining 41% of
daily market turnover. Of this figure, financial customer groups contribute the
majority share, equivalent to 28% of total market turnover, and corporations
13% (BIS, 2002). Although a detailed breakdown of financial customer flows is
not publicly available, our (albeit informal) discussions with a number of major
investment banks suggest that asset management firms and hedge funds each
account for approximately one quarter of customer flows, or together about 20%
of total market flows, with commodity trading advisors (CTAs), central banks
and individuals contributing the remaining 8%.7

2.2.1 Dealers
The interdealer market encompasses market-makers, leverage traders, desig-
nated proprietary or “prop” traders, and senior risk takers. Market-makers
continue to perform their traditional core function of facilitating access for cus-
tomers to interdealer liquidity and providing best execution for customer trades.
But their wider role has evolved over recent years, in a number of ways. First,
market-makers are typically allocated a book exchange rate upon which they
focus their main attention. This compares with a few years ago when these
individuals traded in a range of exchange rates. Second, Daníelsson and Payne
(2001) suggest that market-makers often now focus upon one side of the mar-
ket at a time, rather than posting genuine two-sided quotes. Put another way,
market-makers are no longer typically the main source of price discovery in the
foreign exchange market and do not attempt to generate excess profits from
their market activity, but are instead largely facilitators of customer trades; in
a similar vein, there has been a marked increase at many banks in the ratio of
sales people to market-makers.
Prop traders have been characterised as intra-day, or even “nintendo”, traders
(Bjønnes and Rime, 2003), whose investment time horizon extends from minutes
6 Dealers are also known as broker-dealers; the two terms are synonymous.
7 Market share data can vary substantially between investment banks who focus upon
different segments of the customer market. Also, categorisation of flows between the various
customer groups is somewhat arbitrary, with a number of investors spanning more than one
segment. Consequently, these detailed market share data should be interpreted with caution.

5
to hours at most, and for whom there is no capacity to run overnight positions.
In fact, this description is more consistent with the activities of leverage traders
(also known as spot traders). These individuals trade primarily on the basis
of order flow executed by the bank’s trading desk, and typically have an in-
vestment horizon of a few hours, or at most days. The short-term nature of
these traders’ activity represents a key source of total market volatility, with
tight stop-loss levels typically introduced around every position. As a result,
price breaks of key technical levels typically extend further in the short term
than they otherwise would in the absence of leverage interdealer trading (Osler,
2002, 2003).
In reality, prop traders are actually often discouraged by senior management
from trading too actively and instead are encouraged to focus most of their
allocated risk budget upon longer investment horizons than leverage traders–
typically days but sometimes up to three months or even one year, depending
upon the perceived opportunity set at any given time. Prop desk risk budgets
are reduced overnight, reflecting the difficulty of monitoring positions outside
of business hours, but remain positive.
An additional group of market participants–previously overlooked by aca-
demic analyses and surveys of the foreign exchange market–includes senior risk
takers at large investment banks. These individuals perform a similar function
to designated prop traders, but are allocated a much larger risk budget, for
instance $100-$200 million compared with a typical $25-$40 million for des-
ignated prop traders, reflecting their relative seniority, experience and perfor-
mance track record within the market.8 This risk budget will also be reduced
overnight. Senior risk takers are similarly encouraged to focus upon relatively
longer investment horizons, whenever the opportunity set allows.
A number of infrastructural trends are apparent within the interdealer mar-
ket. First, industry consolidation has meant the number of banks that account
for a majority of interdealer flows has fallen since 1998, with 17 banks in London
and 13 banks in the US accounting for 75% of turnover transacted in these lo-
cations. This compares with 24 and 20 banks in 1998, respectively (BIS, 2002).
Second, the general level of risk appetite within the interdealer market has de-
clined in the wake of the 1998 Long-Term Capital Management crisis, leading
to a reduction in the level of risk capital allocated to individual traders. Third,
and related to the previous point, the rigour of risk management procedures
and infrastructure related to dealer activity has improved substantially in re-
cent years, to a generally high standard (Geithner, 2004). The predominant
focus of risk control procedures remains the imposition of maximum intraday
and overnight nominal–dollar–position limits for individual traders. Many
8 Our estimate of the typical prop trader risk budget is smaller in absolute terms than
Cheung and Chinn (1999), who studied the US sector of the market, and Cheung and Wong
(2000), who examined trading activity and practices in Hong Kong and Singapore. But it is
directionally consistent with both of these studies given broad market trends described below.
Exact position limits are considered to be market-sensitive information, and are therefore
confidential to individual banks. Our estimates result from informal conversations with market
participants.

6
banks also make the amount of risk capital available to traders a function of
past performance, rewarding good performance with an increase in risk capital
and penalising bad results. Typically, traders assume responsibility for–or, in
market jargon, “wear”–all losses, so that these reduce next month’s available
risk capital on a commensurate basis. By contrast, profits are typically shared
with the bank, such that risk capital available to a trader in the next month
will increase only by some fraction of last month’s reported profits. Trading
risk is also monitored using a variety of other metrics, including daily maximum
drawdown, or capital loss, and daily Value at Risk (VaR) limits9 ; the use of
VaR limits appears now to be more commonplace than reported by Cheung and
Wong (2000). A loss that trips intraday limits is unlikely to signal the termi-
nation of a trader’s risk budget, but instead will trigger notification of senior
management, require a detailed explanation of the circumstances surrounding
the loss and of remedial steps to be taken, and closer scrutiny of the trader’s
outstanding positions and activity in the immediate future. More generally,
banks also conduct VaR sensitivity analysis on the activity of its trading desk
in aggregate to ensure that senior management have a broad understanding of
the probability of an extreme event that could threaten the solvency of the
institution.
Fourth, most interbank trading occurs electronically, via either the Elec-
tronic Banking System (EBS) or Reuters D3000. Both systems were established
in 1993 and were the primary facilitators of the subsequent marked increase in
market liquidity. Their functionality is essentially equivalent, providing ex ante
anonymous limit order bid-ask pricing to dealers. Combined, these systems ac-
count for approximately 85% of total interbank activity, with EBS dominating
in all exchange rates other than sterling, Canadian and Australian dollar cross
rates.10 Remaining interbank trading activity is shared by the remnants of voice
broker (10%) and secure bank-to-bank chat lines provided by Reuters (5%). The
dominance of EBS in most liquid exchange rates raises a question-mark over the
robustness of existing empirical research which mainly uses data from Reuters
to test the key predictions of the microstructure literature.

2.2.2 Customers
Customers interact with dealers in order to access the liquidity of the interbank
market. Voice trading continues to account for the majority of customer trades.
But three major electronic systems–FX Connect, FXAll and Currenex–have
recently been introduced to the customer-dealer space.11 All three systems are
9 Value at Risk (VaR) is defined as the maximum percentage value of an investment portfolio

that could be lost during a fixed period (e.g. one day) within a certain confidence level (e.g.
95%).
10 Why this delineation by exchange rate originally arose is not clear. Unsuccessful attempts

have periodically been made by both systems to gain market share in additional exchange
rates.
11 Volumes transacted on these systems have grown rapidly since their introduction. For

instance, FXAll reported a rise in volumes of 104% in 2004, to an annual total of $4.9 trillion.
FX Connect transacts a similar annual volume to FXAll, and these two systems combined

7
multi-bank electronic trading portals linked directly to customer trading desks.
They comprise an essential component of the push towards the introduction of
Straight Through Processing (STP) that facilitates the complete automation of
customer foreign exchange management from order creation through electronic
trading portals to the settlement and confirmation of trades. A primary moti-
vation behind the introduction of STP is a reduction in the risk of human error
at various points in this process (Baird, 2002). The emergence of electronic
portals in the customer-dealer space is unlikely to increase the availability of
order flow data, and thereby will not improve the transparency of the foreign
exchange market and the quality of microstructure-based exchange rate models
as electronic order flow is treated as strictly confidential by system governing
boards.12
Except for corporate hedging of the translation risk of international rev-
enues and costs, all other customer activity in the foreign exchange market
relates to the management of currency exposure in investment portfolios.13 By
reputation, customer order flow is the most important source of private infor-
mation in the foreign exchange market (Lyons, 1995; Ito, Lyons and Melvin,
1998; Bjønnes and Rime, 2003), with the sign of customer trades considered
more informative than the associated nominal value of these trades (Bjønnes
and Rime, 2001b). An important reason for this reputation is the heterogeneity
that exists within this segment of the foreign exchange market: informational
asymmetries; different reaction speeds to data innovations; diverse opportunity
sets and risk/return expectations. We discuss each one of these heterogeneities
in turn in the following sections. But in general terms, all of them suggest that
a broad understanding of the behaviour of an exchange rate at a given time
requires knowledge of the types of customers prevalent in the market at that
time and of the ways in which they trade and interact with the wider market.

Passive versus active customers14 It is estimated that only a small fraction

of daily customer flows are initiated by profit-maximising investors, typically in
currency overlay firms, hedge funds and CTAs (Knott, 2002). Instead, the
majority of flows reflect non-price level dependent hedging activity. Customer
foreign exchange exposure occurs in a number of ways, reflecting the presence
in the market of both passive and active customers. Passive customers inherit
foreign exchange exposure from the sale and purchase of underlying assets, such
account for approximately 86% of all electronic customer trades conducted through multi-bank
portals.
12 This view appears contrary to that of Evans and Lyons (2002a).
13 For instance, on behalf of pension funds of private sector corporations, government and

supranational institutions, and wealthy individuals.

14 Our use of the terms ”active” and ”passive” differs from the sense of Evans and Lyons

(2002a,b), but is more consistent with market tradition. Whether a customer is passive or
active is determined not by the group from which they are drawn–for example, corporations
versus asset managers–but by the nature of their activity in the foreign exchange market.
There are similarities in the categorisation of customers in this section with Kyle (1985).
He defines a market in which three types of participants coexist: nondiscretionary liquidity,
discretionary liquidity and informed traders.

8
as international equities or bonds, around a strategic portfolio benchmark, or
from the accrual of international revenues and costs by corporations. This
exposure is addressed by passive investors in one of two ways. First, and relating
to financial customers, it is left unhedged. In this case, customers essentially
adopt a policy of benign neglect towards currency exposure, presumably in the
belief that returns to active currency management average zero over the long
term. That this neutrality hypothesis has proven invalid by observation during
the post-Bretton Woods period (Baldridge, Meath and Myers, 2000; Hersey
and Minnick, 2000; Parker Global, 2004) suggests that the behaviour of these
participants is inconsistent with the Rational Expectations Hypothesis (REH).
Embarrassment risk associated with long-term underperformance of this policy
of benign neglect should be a significant factor encouraging these investors to
switch to the active customer camp; this switch may be encouraged by the
trend towards more intense scrutiny of longer-term investment performance by
pension fund trustees.
The second way that inherited currency exposure is addressed by passive
customers—both financial and corporate customers—is by implementation of hedges
that return exposure back to an underlying strategic benchmark position. These
hedges are typically implemented without tactical consideration of the exchange
rate at which they are conducted.15 As well as emphasizing the lack of a profit
maximising motive on the part of many foreign exchange customers, this be-
haviour also seems to contradict the Efficient Market Hypothesis (EMH) as-
sumption of investor risk neutrality; passive participant groups appear willing
to pay a premium to other, active customer groups in order for these investors to
assume their short-term foreign exchange risk exposure (Kearns and Manners,
2004).
Foreign exchange exposure is introduced by active customers into portfolios
as part of an active currency programme. This behavior is the most consis-
tent with the EMH, although one can reasonably question whether even these
customers, or their associated investors, exhibit risk neutrality given the ex-
tent of evidence against this assumption generally (Lewis, 1994). Within active
currency programmes, appointed currency managers are given discretion by in-
vestors to add value to underlying portfolios by implementation of appropriate
active hedges around a strategic currency benchmark. These hedges are subject
to explicit portfolio guidelines relating to position limits and admissible ex-
change rates that are designed to ensure that the contribution to total portfolio
risk of these hedges is consistent with ex ante investor expectations. Strategic
currency benchmarks have traditionally been defined in relation to a set of assets
or liabilities to which the investor has a long-term exposure; this is the format
of a traditional currency overlay programme. More recently, investor interest
has grown in currency programmes implemented on the basis of an underlying
notional capital value separate to this underlying asset exposure with a bench-
15 This is the case for short-term hedging by corporate customers. This group also under-

takes medium-term hedging that does have regard to exchange rate levels. However, market
anecdote suggests evidence of associated systematic forecasting errors, undermining a key
tenet of the Efficient Market Hypothesis.

9
mark equal either to zero or a cash—risk-free—interest rate; this investor interest
has largely accounted for the recent rise in popularity of currency hedge funds,
and leveraged funds offered by established asset management firms.16

Informed versus uninformed customers An alternative customer classifi-

cation would divide participant groups into informed and uninformed investors.
Which category a particular group belongs to reflects their ability to interpret
the impact of data innovations upon future exchange rate returns. Most obvious
amongst informed customers would be large central banks, who both observe
innovations in fundamental data series relevant to their own currency in ad-
vance of other market participants, and who have access to the best streaming
information on intraday order flow in their sector of the foreign exchange mar-
ket. Reflecting both the sophistication of this customer group, as well as the
magnitude of associated flows, central bank foreign exchange activity is among
the most keenly sought source of private information. This information can
relate either to prospective market intervention or compositional changes in the
currency denomination of official foreign exchange reserves. For instance, on
a number of occasions since the inception of the euro in January 1999, and
particularly during 2003 and 2004, market anecdote has suggested important
portfolio adjustments away from the US dollar and towards the euro amongst
Middle Eastern, peripheral European and Asian central banks. As such activ-
ity occurs transacting dealers gain important information that can be used to
inform their own foreign exchange trading strategies, at least in the short term.
An increasing number of central banks allocate a small part of official foreign
exchange reserves to wealth creation, reflecting both an increase in the stock of
reserves, particularly in Asia, and less need for these reserves to be held in liquid
form due to wider use of derivative instruments. As a result, market anecdote
suggests that central banks are now an active part of daily traded flows, and
that investment activity on this portion of foreign exchange reserves has proven
successful.17

Investment styles Customer heterogeneity, and contradictions of the EMH,

can be observed in the range of investment styles that coexist within active—and
informed—customer groups. These styles exploit different types of information—
as well as the same information in different ways—leading to differences in ex-
change rate expectations.18
16 The long-term performance of active currency managers compares favourably to that of
managers in other asset classes, including equities and bonds (Morris, 2004). This performance
provides a stark, and robust contradiction to the suggestion by Federal Reserve Chairman Alan
Greenspan that the practice of active currency management is akin to a coin tossing contest
(Greenspan, 2004).
17 By contrast, evidence is mixed on whether foreign exchange intervention by central banks

has been a profitable activity during the floating exchange rate era. For instance, see Becker
and Sinclair (2004) and Dalio (2002) for contrasting conclusions.
18 There has been some blurring of differences between investment styles in recent years as

the product ranges of the various active, informed customers have converged.

10
 Many CTAs are pure technical, or “black box”, managers. The associated in-
vestment process will typically comprise a set of optimised technical, or chartist,
trading rules that have no intuitive underlying theoretical economic interpreta-
tion.19 News to this customer group is historical price innovation, over any pe-
riod from minutes, hours, days, all the way out to years, combined with detailed
trading rules related to key support and resistance levels, moving average cross-
over levels, over-bought and over-sold calculations20 and a range of other price
patterns. Publicly announced macroeconomic news is only relevant indirectly,
to the extent that it has some historic price impact. The importance of this
type of technical trading within the foreign exchange market, particularly for
high frequency exchange rate returns, is confirmed by Allen and Taylor (1990),
Taylor and Allen (1992), Cheung and Wong (2000), and Euromoney (2002).
This type of trading is often considered irrational, and therefore inconsistent
with the EMH, as it represents an important source of systematic forecasting
errors. In reality, the extent to which technical investors have demonstrated an
ability to generate persistent excess returns suggests that they are in fact acting
rationally and exploiting the proven unit root properties of spot exchange rates
in the vicinity of equilibrium (Taylor, Peel and Sarno, 2001).
Hedge funds and currency overlay managers initiate order flow predomi-
nantly on the basis of publicly available macroeconomic information, with an
expected pay-off schedule that is likely to stretch from around one to three
months. The investment process of these customers is often highly quantified,
with pre-defined trading rules based upon theoretical relationships between eco-
nomic or financial variables and exchange rates. In this case, public data inno-
vations will directly trigger customer order flow. Most hedge funds and currency
overlay managers also employ technical analysis similar to CTAs21 , and some
also introduce tactical exchange rate hedges into portfolios on the basis of a
purely qualitative interpretation of fundamental and order flow data, on the
rationale that not all events-including central bank intervention-that determine
exchange rate returns can be quantified in a consistent manner over time. The
profitability of qualitative trading will rely upon the skill of portfolio managers
to interpret the impact of order flows and market reaction to these events in
a consistently accurate manner. Associated trades will—by design—typically be
risk-reducing at the total portfolio level, thereby improving the Information
19 Market anecdote suggests that some CTAs have begun to embrace fundamental-based

trading strategies, in an effort to diversify returns. Others have adopted a more discretionary
investment style. Models will be used to determine key technical levels and turning points
but positions will be implemented on a discretionary basis by portfolio managers and traders.
CTAs also appear to be increasing the breadth of portfolios by trading in a greater range
of exchange rates; traditionally, trading activity concentrated on euro-dollar, yen-dollar and
euro-yen.
20 Over-bought and over-sold calculations attempt to define when an exchange rate has

moved too far and fast in either direction. They are typically calculated based on a moving
average of the difference between the number of advancing and declining days over a certain
period of time.
21 A testable implication of the increasing use of technical analysis is that half-lives of ex-

change rate disequilibria should have risen over recent years, particularly for the most liquid
exchange rates. We are grateful to Dirk Morris for this observation.

11
Ratio of the investment strategy.22
Finally, risk-control currency managers introduce option replication strate-
gies into client portfolios in order to minimise the downside risk attached to any
level of foreign exposure over some fixed investment horizon, typically one year
(Layard-Liesching, 2002). Consequently, this type of manager will react to price
and data innovation indirectly to the extent that these affect the downside risk
profile implicit within portfolios.

Reaction speeds Customer heterogeneity implies differences in the reaction

speed of various customers to the arrival of new information. Many hedge funds
are among the quickest to respond to significant data innovations by immedi-
ately trading in size to implement desired changes in portfolio construction. As
hedge funds typically make more use of leverage than other active—or informed—
customers, significant shifts in portfolio positioning will typically represent a
larger proportion of total assets than is the case for other customers. While
this implies that hedge funds can potentially benefit more than most other
customers from significant data innovations, the downside of this approach is
that sudden strategy shifts do not allow hedge fund traders to exploit intra-day
market liquidity—by feeding orders into the market—to ensure that trades are im-
plemented at best price;23 another, more systemic consequence of this tendency
to effect sudden shifts in positioning is that this behaviour of hedge funds can
amplify the impact of shocks to the foreign exchange market during periods of
significant market stress (Geithner, 2004).
By contrast, and at odds with the REH that argues public information is
instantaneously reflected in exchange rates, currency overlay firms typically ex-
ploit investment strategies that build or reduce active hedges gradually over
the course of several trading sessions in response to significant data innovations.
This contradiction is consistent with practical observation that the contempora-
neous relationship between order flow and exchange rates is relatively persistent
(Bjønnes and Rime, 2001a; Froot and Ramadorai, 2001; Fan and Lyons, 2001;
Evans and Lyons, 2002a,b). It may be explained by a desire to minimise the
impact of portfolio shifts on exchange rate returns, and therefore expected prof-
its, given the size of assets under management; this explanation is only valid as
long as one believes that exchange rate returns exhibit significant positive serial
correlation.
Differences in customer reaction speeds can provide dealers with important
information about the size and sign of customer trades, and the likely persistence
of this trading. To the extent that a dealer sees order flow from large active,
22 Information Ratio is defined as a measure of excess return per unit of risk taken, where

excess return is defined as the difference between the currency portfolio’s return and the return
to a benchmark index.
23 It may also be the case that sudden shifts in hedge fund positioning are due to position

liquidation as a result of significant losses; leverage implies that the probability of significant
losses will be higher within the hedge fund community than amongst asset management firms.
The recent trend towards contingent credit agreements mitigates this probability, to some
degree (Geithner, 2004).

12
informed currency customers it is generally reasonable to assume that this is only
a small part of the total trade being executed, and that the remaining orders
are likely to be fed into the market throughout the current trading session (in
the case of hedge funds) or several trading sessions (currency overlay managers,
for instance). This knowledge will allow the dealer either to “piggy-back” on
the trade, committing some of his own risk capital to the same trade (Bjønnes
and Rime, 2001b), or to net off trades from other customers.
Information about large customer order flow is made available by dealers
to other large customers, on a quasi-anonymous basis via direct voice links,
once a dealer has executed a customer order. An indication of the exchange
rate, type of initiating customer, and size and sign of the transaction will be
provided, as well as the ease with which the market absorbed the order.24 This
information helps other customers gauge the extent of technical support and
resistance levels around current spot exchange rates, including any option-based
trading structures, and the probability and likely extent of any break-outs from
prevailing price trend channels.
In a similar manner, ad hoc information concerning the level and volume
of limit orders placed by customers on dealer order books will be provided by
dealers to preferred clients on a daily basis. Osler (2002) finds that information
regarding limit orders has historically had an important explanatory role for
yen-dollar. But the quality of this information is likely to have deteriorated
in recent years as customers have become more reticent to place limit orders
with dealers, reflecting greater cogniscence of their own strategic market im-
pact. For informed customers, this cogniscence has increasingly focused efforts
upon ensuring that internal risk management systems can incorporate appropri-
ate, continuous position gain-loss monitoring procedures to allow creation and
execution of trades as stop levels are approached.

3 Theoretical Models of the Foreign Exchange

Rate
Given the foregoing discussion of the microstructure of the foreign exchange
market, we are now in a position to analyse alternative approaches to modelling
the exchange rate, beginning with traditional asset-price exchange rate models
and then moving to more recent microstructural models.

3.1 Traditional Asset-Price Models

Traditional asset-price models of exchange rate determination take the form:

St = β 0 Ft + αSt+1
e
, (1)
24 For instance, "Asian names" is a market pseudonym for the major central banks in Asia.

But dealers will never explicitly provide the name of customers initiating trades to other
customers, in order to ensure client confidentiality.

13
 e
where St is the spot exchange rate at time t, and St+1 is the one-period-ahead
expected spot rate given information at time t (both expressed as domestic
currency per unit of foreign currency). Ft represents a vector of fundamental
variables that exhibit some persistent explanatory power for the determination
of exchange rates, and β represents a vector of factor loadings. This formulation
is sufficiently general to encompass simple monetary models of the exchange
rate, sticky-price overshooting models and portfolio balance models (see Taylor,
1995; Sarno and Taylor, 2002; Lyons, 2001, chapter 7).
The first thing to note about this class of models is that they are explicitly
macroeconomic and apparently independent of either the way information is pro-
cessed in the market or the institutional structure of the market (Lyons’s “two
i’s” –Lyons, 2001). Thus, asset-price models are essentially equilibrium models
that marginalise the importance given to the means by which that equilibrium
is reached or the institutional setting in which currency prices are determined.
e
Moreover, whose expectation is St+1 ? Many of these models were developed
(or have been interpreted) under the strict assumption that agents are endowed
e
with rational expectations, so that St+1 is the true conditional mathematical
expectation. If that is the case, then news on fundamentals is instantly im-
parted into the market as soon as it is released and the exchange rate will jump
to its new equilibrium level. This cannot, however, be the whole story, since this
would imply a complete consensus of opinion among all market participants as
to the precise change in “fair value” due to any data innovation; furthermore,
our discussion above of the behaviour of the various participant groups in the
foreign exchange market makes clear that this is not the practical reality of
the market. A less restrictive assumption, which has its roots in Friedman’s
(1953) apologia of floating exchange rates, would be that the market in some
average sense conforms to the rational expectations hypothesis and reaches this
equilibrium because uninformed participants are driven out of business (or to
change their behaviour) by informed participants. In this case, equation (1) rep-
resents a snapshot of equilibrium in the market as dependent on macroeconomic
fundamentals, but says nothing at all about how the market gets to that equi-
librium. It is perhaps for that reason that asset-price models have performed
spectacularly poorly in explaining short-run exchange rate movements (Meese
and Rogoff, 1983; Taylor, 1995; Sarno and Taylor, 2002), although there is some
evidence that they may be able to explain longer-run movements. Indeed, given
that the profession seems to have moved towards support for some form of long-
run purchasing power parity equilibrium for exchange rates (see e.g. Taylor
and Taylor, 2004), then grafting this onto an existing consensus of long-run
monetary neutrality implies that the long-run behaviour of the exchange rate
must reflect relative movements in monetary velocity between countries (see e.g.
Frankel and Rose, 1995; MacDonald and Taylor, 1993).
Viewed as a long-run equilibrium relationship, therefore, asset-price mod-
els may be innocuous but they are also anodyne: they tell us nothing about
short-run exchange rate behaviour and, in particular, nothing at all about how
information on macroeconomic fundamentals gets compounded into the current
exchange rate; they draw a veil over the most interesting part of the show.

14
 As Lyons (2001) notes, microstructural approaches to exchange rate deter-
mination are not necessarily at odds with asset-price models. For example,
if–as predicted by most asset-price models–relative monetary velocity affects
the exchange rate, then agents will not wait for this month’s money supply fig-
ures to be released before adjusting the exchange rate accordingly. Rather, an
excess increase in the money supply (decline in velocity) will itself have created
downward pressure on the external value of the currency as interest rates decline
(perhaps as a result of the central bank’s enthusiastic purchase of government
bonds in their open market operations) and net capital inflows diminish. While
the release of the money supply figures at the end of the month will explain
the downward movement in the foreign price of domestic currency (i.e. the
exchange rate), the effect of the excess money supply growth will have been
reflected throughout the month in a rise in negative order flow associated with
the domestic currency; this is one explanation why order flow appears to be
highly correlated with contemporaneous exchange rate movements (Evans and
Lyons, 2002a).
But while customer order flow in particular may represent the missing piece
of the exchange rate puzzle, this does not mean that these data can necessarily
be exploited for the purposes of predicting exchange rate movements even at very
short horizons, since they are only contemporaneously observable on a real-time
basis by a limited number of well-informed market participants. These partic-
ipants include the custodian or investment bank or electronic trading platform
that collects and collates the data, and on an indirect, verbal basis, a select
group of preferred customers of these organisations–for instance large asset
management firms, hedge funds and CTAs.25 Limited access to order flow data
reflects the confidentiality concerns of major market participants whose trading
activity is captured by these data and which could therefore be identified by
competitors, and a wish on the part of the collecting institution to sustain and
profit from this potential information advantage for as long as possible. For
this reason, direct access to proprietary order flow data by preferred customers
is provided only after the raw data have been filtered and collated into an in-
dex to ensure customer anonymity, and with at least some publication lag.26
Dealers in smaller banks see most customer order flow data only indirectly once
they have been embedded in prices. Furthermore, most customers do not gain
direct access to even filtered flow data, and instead have to content themselves
with interpretative analysis provided by dealers that includes qualitative trad-
ing strategies based upon developments in their own collated flow data (HSBC,
25 Preferred clients are those identified by investment banks to be of strategic importance

to their position in the foreign exchange market. As well as best execution, these clients will
be provided with better streaming information than other clients on intraday and daily order
flow and market positioning, and will be allocated resources for collaboration on confidential,
customised research. Importantly, information concerning order flow data made available
to preferred customers never includes details of individual customer transactions, so that
customer anonymity is assured.
26 As we report in a companion paper, filtering and publication lags appear to significantly

reduce the information content of available order flow data, from a customer perspective (Sager
and Taylor, 2005).

15
2003; Citibank, 2003). (We address the issue of the predictive content of order
flow for exchange rate movements in Sager and Taylor (2005).)
In summary, while traditional asset-price exchange rate models may possibly
tell us about equilibrium conditions or “fair value” in the foreign exchange
market, they ignore extremely important aspects of the foreign exchange market
which are evident in our preceding discussion of its microstructure. They can tell
us nothing at all about the way in which heterogenous groups interact with one
another in the market and transmit information to one another. in other words,
they ignore issues of information transmission and institutional structure.

3.1.1 The Evans-Lyons Model

The academic literature presents a number of stylised models of the interaction
between the various customer groups and dealers described above. One of the
most recent and influential is due to Evans and Lyons (2002a,b), which in turn
borrows aspects of Kyle’s (1985) sequential equilibrium model. The model is
consistent with (but does not rely on) equilibrium or “fair value” in the market
being determined in accordance with asset-price models of the kind discussed
above, but also attempts fill in missing gaps in these models in terms of infor-
mation transmission and institutional features. Its broad characteristics may
be set out as follows.
There are N dealers in the market, a continuum of non-dealers (or cus-
tomers), and an infinite number of trading days. Dealers observe a periodic
payoff on foreign exchange, denoted Rt , representing the flow of macroeconomic
information. This payoff is proxied by Evans and Lyons by interest differen-
tials. There are three stages to the foreign exchange trading day. First, and
having observed Rt at the start of each day, dealers independently and simul-
taneously set bid-ask spreads for, and trade with, customers. Second, dealers
trade amongst themselves, independently and simultaneously posting a bid-ask
spread for other traders. These spreads lead to trades, as dealers spread risk
generated by earlier customer trades through the interbank market. Once this
second round of trading is complete, all dealers are able to observe the order
flow, Xt , that has occurred from interdealer trading during that day. These
data are assumed by Evans and Lyons (2002a,b) to convey important informa-
tion about the size and sign of customer trading during Round One. Finally, in
the third round of trading dealers once more trade with customers, in order to
share overnight risk more widely across the market. A crucial assumption here
is that dealers set prices in Round Three such that customers willingly absorb
all dealer inventory imbalances, so that dealers run no overnight open positions.
Consequently, the closing price at the end of each day within the Evans and
Lyons model will be

t
X t
X
Pt = β 1 ∆Rt + β 2 Xt (2)
τ =1 τ =1

16
and the change in prices from the end of day t − 1 to the end of day t can be
written as

∆Pt = β 1 ∆Rt + β 2 Xt (3)

P
where Xt = N i=1 Tit , is the total order flow generated by interdealer trading
1
during the day and Tit = αCit is the sum of customer orders received by indi-
vidual trader i during the first round of trading; α is a constant coefficient. A
key distinction in the model exists between “active” and “passive” customers.
Round One customers are active in the sense that they generate initial imbal-
ances in dealers’ positions, while Round Three customers are passive in that
they passively absorb dealers’ positions at the end of the day.27
Although the Evans and Lyons (2002a,b) characterisation of the foreign
exchange market is a useful taxonomic device for clarifying one’s thoughts
about market microstructure and customer-dealer interactions, we believe that
it should not be interpreted too literally as a description of the foreign exchange
market (as it appears to have been in a number of applied microstructural stud-
ies of the foreign exchange market). In particular, the notion that customers
willingly absorb the daily inventory imbalance of dealers seems at odds with
market practice. Indeed, customers have traditionally paid dealers to assume
this type of price and credit risk on their behalf, thereby preserving customer
anonymity and allowing access to the liquidity of the interbank market through
a relatively small number of contact points.28 This payment is made in soft
terms, with dealers traditionally able to divide customer trades into smaller
tranches as they spread risk through the interbank market. As the width of
bid-ask spreads is a positive function of transaction size for transactions beyond
a minimal value (Bjønnes and Rime, 2001b), tranching saves the dealer trans-
action costs and, along with knowledge of, and trading around customer limit
order books, has traditionally represented a major source of trading profit for
banks (Rime, 2001).
This profit source has recently been undermined by the behaviour of those
customers increasingly aware of the strategic price impact of their trading activ-
27 Evans and Lyons’ use of the terms "active" and "passive" differs from ours above, or
market norms.
28 It is standard practice for a currency overlay manager (the agent) to request prior approval

from a client (the principal) to trade with a maximum of ten dealers on a continuous basis
on the client’s behalf. This ensures that the price and credit risk of the client with respect to
any one of the dealers is minimised and that the overlay manager maximises its opportunity
to achieve best execution for the client. More recently, interest has grown in prime brokerage
agreements, particularly amongst hedge funds. Under these agreements, all of a customer’s
foreign exchange orders are transacted through a single dealer. The prime broker is usually
a large, highly rated institution that allows the customer to initiate trades, subject to credit
limits and collateralisation, with a pre-determined limited group of high quality third party
banks (so-called “Spoke” banks) in the prime broker’s name, rather than the customer’s.
Consequently, customer credit risk is limited to the prime broker. These agreements also
allow for a reduction in operational risk for customers, as all positions are with the prime
broker and only one dealer back office will be involved in trade reconcilliation and netting
(E-Forex, 2003).

17
ity. For example, hedge fund traders will often trade against their own positions
in small size in order to encourage the interdealer community to adopt similar
positions, subsequently reversing this tactic with substantially greater volumes
to effect a short squeeze on the market. In a similar vein, these traders will often
execute many trades simultaneously across the interdealer community, rather
than feeding them into the market, in order to maximise the noise surrounding
their activity and its subsequent price impact. Moreover, currency overlay man-
agers will now regularly engage in the tranching activity traditionally employed
by dealers, dividing trades of $100 million-$1 billion into smaller amounts of,
say, $25-$50 million. These tranches are then spread amongst dealers intermit-
tently during the course of one or more trading sessions. Although consistent
with customer profit-maximisation, this practice undermines the trading profits
of dealers and implies that banks are now accepting risk on behalf of customers
without adequate compensation in return. How dealers can regain payment
for this risk in the future is not immediately apparent, given also the rise of
electronic trading that implies both a compression of dealer bid-ask spreads on
this segment of trading and an increase in ex ante customer anonymity (Portes,
2002).29 Consequently, risk sharing in the foreign exchange market would seem
to be an important area of future research.

3.2 A Thumbnail Sketch of a Push-Pull Model

Given our discussion of the structure of the foreign exchange market, we can
sketch the features of a theoretical framework that would correspond more
closely to the real world than existing theoretical models while paying close
attention to Lyons’s “two i’s” (information and institutions) and having, in
principle at least, have testable implications.
First, there would be only one round to the trading day and customers
would be free to trade at any time. A key distinction between customers would
hinge on whether they are active or passive, and informed or uninformed–in
the sense that we have used these terms in this paper–and therefore whether
they push prices or are pulled into the market by a price movement. Net cus-
tomer orders from either large active customers, or those considered by other
market participants to be particularly well informed, would be expected to ini-
tiate price movements–these would be the so-called “push customers”. Other
customers, perhaps passive or less well informed, or whose investment process
is predominantly driven by technical analysis, are attracted or pulled into the
market by price movements; effectively they exercise an option to trade once the
price crosses their implicit “strike price”. These would be the “pull customers”.
There would be no requirement for dealers to square positions overnight, and no
assumption that customers absorb dealers’ daily inventory. There would also be
evidence of rational errors, no assumption of risk neutrality for any participant
group and, depending upon the particular customer group, no profit motive
from exchange rate hedging.
29 Manipulation of bid-ask spreads in response to larger size, ad hoc customer order flow

remains a potential source of profits for dealers.

18
 Among the various customer groups participating in the market, it would
be possible to test hypotheses concerning whether a certain category could be
described as “push” rather than “pull” by examining the correlation of their
order flow with subsequent price movements: a positive and significant corre-
lation would indicate that the customer is pushing the market; a negative and
significant correlation would be consistent with customers being pulled into the
market to buy as prices drop and to sell as prices rise.30

4 Conclusions
The foreign exchange market incorporates a heterogeneous set of participants.
Heterogeneities include informational asymmetries, different reaction speeds to
significant data innovations, diverse opportunity sets and risk-return expecta-
tions. The behaviour and interaction of market participants, which both gen-
erates persistent inefficiencies and ensures that the process of price discovery is
opaque, is poorly understood. Existing theoretical models of market microstruc-
ture make a number of simplifying assumptions that bear little resemblance to
actual market practice. In particular, although most models assume that cus-
tomer participant groups are the main source of private information in the for-
eign exchange market, theoretical work has concentrated upon the behaviour of
the interdealer market, overlooking the subtleties of the customer segment. On
the other hand, traditional asset-price models of exchange rate determination
ignore crucial issues of information dissemination and institutional structure.
In this paper we set out to improve understanding of foreign exchange mar-
ket structure, and participant group behaviour and interaction. We reviewed
recent structural developments in the market, including trends in daily market
turnover, geographic concentration, liquidity of individual exchange rates and
trading and settlement of customer-dealer transactions. We also presented a
detailed description of the behaviour of individual participant groups within
both the interdealer and customer segments of the market, as well as the ways
in which these various groups interact. This description allowed us to sketch
a model of market behaviour that differs in a number of ways from existing
theoretical models. A model conforming to our “push-pull’ framework would
represent a necessary practical foundation to the market microstructure litera-
ture that emphasises the role of order flow in the determination of high frequency
exchange rate returns.
30 In a recent paper, Carlson and Osler (2005) develop a theoretical and empirical model

that conforms to some of the main features of our suggested framework.

19
References
[1] Allen, H., and M.P. Taylor (1990), Charts, Noise and Fundamentals in the
Foreign Exchange Market, Economic Journal, 100, pp. 49-59.
[2] Baird, D. H. (2002), Integrating eFX with Institutional Investor Needs.
JPMorgan Fleming Asset Management Mimeo.
[3] Baldridge, J., B. Meath, and H. Myers (2000), Capturing Alpha Through
Active Currency Overlay, Frank Russell Company.
[4] Bank for International Settlements (1996), Triennial Central Bank Survey
of Foreign Exchange and Derivatives Market Activity 1995 - Final Results.
www.bis.org.
[5] Bank for International Settlements (1999), Triennial Central Bank Survey
of Foreign Exchange and Derivatives Market Activity 1998 - Final Results.
www.bis.org.
[6] Bank for International Settlements (2002), Triennial Central Bank Survey
of Foreign Exchange and Derivatives Market Activity 2001 - Final Results.
www.bis.org.
[7] Bank for International Settlements (2004), Triennial Central Bank Survey
of Foreign Exchange and Derivatives Market Activity 2003 - Preliminary
Results. www.bis.org.
[8] Becker, C. and M. Sinclair (2004), Profitability of Reserve Bank Foreign
Exchange Operations : Twenty Years After the Float. Reserve Bank of
Australia Research Discussion Paper # 2004-06.
[9] Bjønnes, G.H. and D. Rime (2001a), US Exchange Rates and Currency
Flows. Unpublished mimeo Stockholm Institute for Financial Research.
www.sifr.org.
[10] Bjønnes, G.H. and D. Rime (2001b), Customer Trading and Information
in Foreign Exchange Markets. Working paper # 4. Stockholm Institute for
Financial Research.
[11] Bjønnes, G.H. and D. Rime (2003), Dealer Behavior and Trading Systems
in Foreign Exchange Markets. Norges Bank working paper 2003/10.
http://www.norges-bank.no/publikasjoner/arbeidsnotater/pdf/arb-2003-
10.pdf.
[12] Brown, B.W. and S. Maital (1981), What Do Economists Know? An Em-
pirical Study of Experts’ Expectations.. Econometrica, 49, 2, 491-504.
[13] Carlson, J.A. and C.L. Osler (2005), Short-Run Exchange Rate Dynamics:
Theories and Evidence. Mimeo, Brandeis International Business School.

20
[14] Cheung, Y.-W. and M. D. Chinn (1999), Currency Traders and Exchange
Rate Dynamics: A Survey of the U.S. Market". Journal of International
Money and Finance 20, 439-471.
[15] Cheung, Y.-W. and C. Y.-P. Wong (2000), A Survey of Market Practi-
tioners’ Views on Exchange Rate Dynamics. Journal of International Eco-
nomics, 51, 401-419.
[16] Citibank (2003), Order Book and Option Vols. Report. Citibank.
[17] Consensus Economics (2004), Foreign Exchange Consensus Forecasts.
http://www.consensuseconomics.com/.
[18] Dalio, R. (2002), Inefficiencies and Opportunities in the Currency Markets,
in A Guide to Currency Overlay Management, D. Knott (ed.) Deutsche
Bank. Global Market Research.
[19] Daníelsson, J. and R. Payne (2001), Measuring and Explaining Liquidity
on an Electronic Limit Order Book: Evidence from Reuters D2000-2. LSE
Financial Markets Group Mimeo.
[20] Dominguez, K. M. (1999), The Market Microstructure of Central Bank
Intervention. NBER Working Paper # 7337.
[21] E-Forex (2003). Foreign Exchange Joint Standing Committee E-Commerce
Subgroup Report.
[22] Euromoney (2002), 2nd International Forex Forum. www.euromoney.com.
[23] European Central Bank (2003), Review of the Foreign Exchange Market
Structure. www ecb.int.
[24] Evans, M. and R.K. Lyons (2001), Order Flow and Exchange Rate Dynam-
ics. BIS Papers #2. http://www.bis.org/publ/bispap02.htm#pgtop.
[25] Evans, M. and R.K. Lyons (2002a), Order Flow and Exchange Rate Dy-
namics. Journal of Political Economy, February, 170-180.
[26] Evans, M. and R.K. Lyons (2002b), Informational Integration and FX Trad-
ing. Journal of International Money and Finance, 21, 807-31.
[27] Evans, M. and R.K. Lyons (2003), New Micro Exchange Rate Economics,
typescript, U.C. Berkeley.
[28] Fan, M. and R.K. Lyons (2001), Customer Trades and Extreme Events in
Foreign Exchange. U.C. Berkeley Mimeo.
[29] Federal Reserve Bank of New York (2005). http://www.ny.frb.org/.
[30] Fisher, P. and R. Hillman (2003), Explaining and Forecasting Ex-
change Rates with Order Flows. Bank of France Economic Policy Forum.
http://www.banque-france.fr/fr/fondatio/telechar/paulfisher.pdf.

21
[31] Frankel, J. A. and K. Froot (1985), Using Survey Data to Test Some Stan-
dard Propositions Regarding Exchange Rate Expectations. NBER Working
Paper # W1672.
[32] Frankel, J.A. and A.K. Rose (1995), Empirical Research on Nominal Ex-
change Rates, in G.M. Grossman and K. Rogoff (eds.) Handbook of Inter-
national Economics, III, pp.1689-729. Amsterdam: North-Holland.
[33] Friedman, M. (1953), Essays in Positive Economics. University of Chicago
Press.
[34] Froot, K. A. and J.A. Frankel (1989), Interpreting Tests of Forward Dis-
count Bias Using Survey Data on Exchange Rate Expectations. NBER
Working Paper # W1963.
[35] Froot, K.A. and K. Rogoff (1995), Perspectives on PPP and Long-Run
Real Exchange Rates, in G.M. Grossman and K. Rogoff (eds.) Handbook
of International Economics, III, pp.1647-88. Amsterdam: North-Holland.
[36] Froot, K.A. and T. Ramadorai (2001), Currency Returns, Institutional
Investor Flows and Exchange-Rate Fundamentals. State Street Associates
Working Paper # 5.
[37] FXAll (2005), FXAll Press Release. http://www.fxall.com/about/pressrelease58.html.
[38] Geithner, T. J. (2004), Hedge Funds and their Implications
for the Financial System. Federal Reserve Bank of New York.
http://www.newyorkfed.org/newsevents/speeches/2004/gei041117.html.
[39] Greenspan, A. (2004), Euro in Wider Circles. Re-
marks to the European Banking Congress, Frankfurt.
http://www.federalreserve.gov/boarddocs/speeches/2004/200041119/default.htm.
[40] Hersey, B. and J. Minnick (2000), Currency Overlay. Global Pensions.
February.
[41] HSBC (2003), FX Flow Dynamics. April. HSBC.
[42] Ito, T., R.K. Lyons and M.T. Melvin (1998), Is There Private Information
in the FX Market? The Tokyo Experiment. The Journal of Finance, June,
1111-1130.
[43] Jansen, D.J. and J. de Haan (2003), Statements of ECB Officials and their
Effect on the Level and Volatility of the Euro-Dollar Exchange Rate. CESifo
Working Paper # 927.
[44] Kearns, J. and P. Manners (2004), The Profitability of Speculators in
Currency Futures Markets. Research Discussion Paper # 2004-07. Reserve
Bank of Australia.

22
[45] Knott, D. (2002), A Guide to Currency Overlay Management. Deutsche
Bank. Global Market Research.
[46] Kyle, A.S. (1985), Continuous Auctions and Insider Trading. Econometrica,
Vol. 53, 6, 1315-1336.
[47] Layard-Liesching, R. (2002), Adding Value Through Currency Manage-
ment. In A Guide to Currency Overlay Management, D. Knott (ed.)
Deutsche Bank. Global Market Research.
[48] Lewis, K.K. (1994), Puzzles in International Finance Markets. NBER
Working Paper #4951.
[49] Lyons, R. K. (1995), Tests of Microstructural Hypotheses in the Foreign
Exchange Market. Journal of Financial Economics, 39, 321-351.
[50] Lyons, R.K. (2001), The Microstructure Approach to Exchange Rates. MIT
Press.
[51] Lyons, R.K. (2002a), Foreign Exchange: Macro Puzzles, Micro Tools. Fed-
eral Reserve Bank of San Francisco Annual Review.
[52] Lyons, R.K. (2002b), Theoretical Perspective on Euro Liquidity. Economic
Policy. Vol. 17, 35, 572-597.
[53] Lyons, R.K. (2003), Explaining and Forecasting Exchange Rates with Or-
der Flows. 3RD Bank of France Economic Forum. http://www.banque-
france.fr/gb/fondatio/telechar/R_k_lyons.pdf.
[54] Morris, D.M. (2004), Presentation to Currency Roundtable. Putnam In-
vestments mimeo.
[55] NYSE (2005), New York Stock Exchange.
http://www.nyse.com/home.html.
[56] Osler, C.L. (2002), Stop-Loss Orders and Price Cascades in Currency Mar-
kets. Federal Reserve Bank of New York Staff Report # 150.
[57] Osler, C.L. (2003), Currency Orders and Exchange Rate Dynamics: An
Explanation for the Predictive Success of Technical Analysis. Journal of
Finance, 58, 5, pp.1791-1820.
[58] Parker Global (2004), Parker FX Index. www.parkerglobal.com.
[59] Portes, R. (2002), Comment on Hau, H., W. Killeen and M. Moore, How
has the Euro Changed the Foreign Exchange Market? Economic Policy,
vol. 17, 34, 149-191.
[60] Rime, D. (2001), Private or Public Information in Foreign Exchange Mar-
kets? An Empirical Analysis. www.sifr.org.

23
[61] Rogoff, K. (2002), Discussion, in Theoretical Perspective on Euro Liquidity,
R. K. Lyons, Economic Policy, Vol. 17, 35, 572-597.
[62] Sager, M.J. and M.P. Taylor (2005), Order Flow and Exchange Rate Fore-
casting. University of Warwick mimeo.
[63] Sarno, L. and M. P. Taylor (2002), The Economics of Exchange Rates.
Cambridge: Cambridge University Press.
[64] Taylor, M.P. (1995), The Economics of Exchange Rates. Journal of Eco-
nomic Literature, 33, pp.13-47.
[65] Taylor, M.P. (1988), What do Investment Managers Know? An Empirical
Study of Practitioner’s Predictions. Economica, 55, 218, 185-202.
[66] Taylor, M.P. and H. Allen (1992), The Use of Technical Analysis in the
Foreign Exchange Market. Journal of International Money and Finance,
11, 304-314.
[67] Taylor, M.P., D.A. Peel and L. Sarno (2001), Non-Linear Mean Reversion in
Real Exchange Rates: Towards a Solution to the Purchasing Power Parity
Puzzles. International Economic Review, 42, 1015-42.

24