TURN-OF-THE-MONTH: WINDOW DRESSING BEHAVIOR 1

TURN-OF-THE-MONTH: WINDOW DRESSING BEHAVIOR

Linus Nilsson

Abstract

We study the impact of trading during the end of the month cycle. We find that there is a

meaningful negative expected return from owning equities in the last trading hour of the month.

The effect is large and potentially exploitable by investors that are not tied to monthly reporting

cycles. The return pattern is different from other days in equity markets and has persisted over

more than a decade. The effect is generally larger for US small cap indices than large cap

indices.

The reasons for the effect may be related to window dressing by fund managers, risk control,

lottery-ticket behavior or less likely market manipulation. The effect applies to broad indices

making it less likely that it is driven by a few traders but rather by the behavior of a large group

of traders responding to the same incentives.

Keywords: Turn-of-the-month, window dressing, market anomalies

TURN-OF-THE-MONTH: WINDOW DRESSING BEHAVIOR

The Turn-of-the-month (TOM) has been documented in a number of different research

papers and articles. A number of calendar driven anomalies were identified in (Josef & Smidt,

1988) that found number of persistent (or semi-persistent) anomalies.

Other researchers have found (Lakonishok, Shleifer, Thaler, & Vishny, 1991) that there is

persistent pressure on fund managers to only display “good stocks” in (mainly quarterly) client

reports. According to the authors, this leads to selling pressure into the month-end, effectively

cleaning up the book from any bad bets.

Fund managers have a strong incentive to show that they are actively managing clients’

money. Sitting on poorly performing stocks may lead to difficult client explanations. As

identified by earlier researchers (Lakonishok, Shleifer, Thaler, & Vishny, 1991) there is a

connection between quarterly transparency reports and window dressing behavior. While

transparency might differ between funds, funds typically have at least quarterly reporting. While

pension funds are generally contrarian, that is, they buy poorly performing stocks and sell

strongly performing stocks they also tend to clean up “mistakes” at reporting points.

Other authors (Carhart, Kaniel, Musto, & Reed, 2002) have found that there seems to be

a degree of gaming behavior around quarter-end and year-end leading to inflated prices around

those points in time. They find that less liquid stocks increase more in price than larger more

liquid shares. However, for both groups, the effect is significant, but may also relate to

aggressive marking of smaller positions. They found that the average mutual fund outperformed

the index on the last day of the month and under performed on the first day of the month. That is,

some evidence that popular positions are being aggressively traded / marked.
TURN-OF-THE-MONTH: WINDOW DRESSING BEHAVIOR 3

Empirical study

We will execute a study, examining the behavior of the last trading day, compared to an

average trading day to see if the pattern is different on that day or not. The last trading day would

be the day that is the most relevant for window dressing activities.

We will be using two indices. The S&P 500 and the Russel 2000 to evaluate if there is a

difference in behavior for a large cap index compared to a small cap index. Theoretically, a small

cap index should be easier to “push” in a certain direction, especially so during days when there

are many small incentives, across a large variety of fund managers, to increase the price of

certain securities.

To be noted is that we are not going to explore individual stocks, but rather act on the

aggregate. We will be using continuously traded futures for the indices. The benefit of these is

that we can have a continuous proxy for the price of the underlying indices, even when the

regular cash market is closed. This is potentially important to estimate overnight effects on the

indices. All returns in this paper are expressed as excess returns as Futures are self-financed.

Average Intraday Behavior

For the study to make any sense, we need to find out the average behavior of the stock

market on an average day. Here we do not correct for any possible window dressing only seeking

a baseline from which we can conduct further studies.

We use so called continuous contract when dealing with futures and total return series

when dealing with the indices. The benefit of using continuous contracts is that we do not have

to manually adjust for the finite duration of a future contract. The downside is that we have to

rely on the provider of the contract that it is done correctly. Futures include a funding component
TURN-OF-THE-MONTH: WINDOW DRESSING BEHAVIOR 4

(the prevailing risk free rate and adjustments for dividends and are thus instruments that are

“self-funded”).

One of the factors that have been documented historically (Cooper, Cliff, & Gulen, 2008)

is that equity returns are mostly generated from the market close to the market open. The size of

the effect depends if an actual cash security is used, compared to a continuously traded future.

Investors should note that the opening price is volatile and may not be tradeable.

As a brief note, there are a few issues with data, as markets have changed over time. For

instance, the S&P 500 E-mini future has changed trading hours at several points in time. The first

(open) trade has also changed definitions over time, skewing the results in certain at certain

points in time. The index trading session has remained largely unchanged with official trading

hours taking place between 9:30 and 16:00. Over the last decade, pre-open and post-close trading

has become more prominent.. The data became (semi-)continuous in mid-2001 when trading

only paused for 15 minutes after the close and for 45 minutes between 17:15 and 18:00. We will

therefore use data from 2002. There has been a gradual increase in length of the tradeable hours

since then, but the changes are relatively minor. One of the recent changes was to extend the

trading hours on Fridays.

Settlement, which has an important impact on the portfolio results (accounting purposes)

is an important point in time. Since it carries economic impact, it is likely more sensitive to price

pressures. The opening price carries less economic value from this perspective.
TURN-OF-THE-MONTH: WINDOW DRESSING BEHAVIOR 5

Various existing anomalies

Overnight vs Session behavior

We study the behavior of the futures when the market is open or closed. The open price

itself, is usually subject to more volatility than for instance the close prices. The open price, does

not carry any real economic significance for settlement or clearing activates. It does carry value

for some trading applications. As an empirical observation, trading “on the open” is subject to

larger slippage as the price gyrates back and forth for the first minute before finding a stable

level. Prior research (Cooper, Cliff, & Gulen, 2008) documented excess return during the

overnight session but not during the day. We are unable to find returns of the same magnitude

(Figure 1). The magnitude is much less than previously reported. If not adjusting for dividends

and similar effect, the results would be lower in prior studies but would still point in the same

direction.

𝑟𝑓 ∙ 𝑑𝑎𝑦 𝑡𝑜 𝑒𝑥𝑝
𝑃𝑟𝑖𝑐𝑒 = 𝑆𝑝𝑜𝑡 (1 + ) − 𝐷𝑖𝑣
360

Fair value for an ETF, is given above, including the divided yield and risk free interest

rate. An investor closing out a position at the end of the day, is theoretically able to place the

cash in an overnight deposit and alternatively in a dividend swap.

We believe that much of the excess returns is due to how the ETF is accumulating interest

rate return. The cost of doing trade might be small, but may also not allow for an arbitrage trade.

This could partially explain the difference between the active and non-active sessions. Overnight

returns are still larger than the returns from the day session.
TURN-OF-THE-MONTH: WINDOW DRESSING BEHAVIOR 6

Weekend Effect

Since investors are carrying risk over the weekend, without being able to hedge or adjust

exposure there ought to be a risk compensation for this. In addition, there are two additional

earning days which would imply that Mondays would have a high realized return than Friday

due to the risk compensation and additional earnings accumulation over the weekend.

Isolating the effect enables us to adjust the potential issue. For a more thorough report on

well-known anomalies with sound criticism about the actual effects see (Schwert, 2002). Among

other things, the author claims that the weekend effects have lost its predictive power.

There weekend behavior changed around the great recession in 2008, at least using the

data we have access to (Figure 2). We measure this as the last trade (which have taken place at a

slightly different point in time throughout the sample) to the first 15-minutes after the

commencement of the new trading week (adjusted for holidays and market closures). If there is

a weekend effect, this would be cleanest way to extract the excess returns as macro news are

rarely released over a weekend (especially US economic data).

We note that effect is small, approximately one negative basis point (bp) per week, over

the whole sample and subject to a regime change that seems to have happened in 2008. The size

and changing sign of the effect makes this difficult to trade upon.

The change in the sign of the effect seems to have started in 2007 already (and accelerate

in 2008). One way to explain this is the amount of negative news increased over the weekends

throughout that period. As a potential driver, during the crises, the Federal Deposit Insurance

Corporation (FDIC) takes control of failed banks over weekends. The FDIC maintains a list of

failed banks and 95% of the banks was closed over a weekend (Failed Bank List, n.d.). Given the
TURN-OF-THE-MONTH: WINDOW DRESSING BEHAVIOR 7

behavior of the regulator, this could have an effect on the behavior of people seeking to avoid

risk over weekend. Furthermore, we also have anecdotal evidence that proprietary traders shifted

their risk management strategies, reducing position over night and over weekends.

Other effects

There are turn-of-the-month effects, weekday effects (Josef & Smidt, 1988), option

expiry effects and as documented in a recent paper (Nilsson, 2015) there is a strong effect around

the FOMC meeting.

The effects will influence the average behavior and the reader needs to be aware that

there may be other significant effects in the data we are analyzing.

Baseline behavior

After adjusting for all of the previously mentioned anomalies, we can calculate the

average intraday behavior by removing previously identified anomalies. We use 15-minute

intraday data, based on the E-mini S&P 500, one of the most liquid securities and arrive at the

results in Figure 3. Equities have not generated large excess return over the period and the daily

average is a few positive basis points. Compared to the daily volatility, the results are small.

Nevertheless, the results mirror the overnight return pattern, with most of the return

generated from close to open. We note that returns generally accrue when the US market is

“silent”, i.e. no news, from midnight to approximately 7:30 in the morning. We observe a decline

until 11-ish, a gain into the close and a pullback of equal size incorporating additional earning

data that is usually released after the equity close (16:15). The average active session after is
TURN-OF-THE-MONTH: WINDOW DRESSING BEHAVIOR 8

rather volatile. More importantly, the behavior can mostly be tied to economic (open / close in

different time zones).

Remarkable is that there is no average move post-macro data release, usually taking place

at 8:30. Potentially there is a slight decrease ahead of those event. There are some yearly

differences that are striking, in declining equity market years (2002, and 2008 for this sample).

We find that, although the size of the effect is still very small, there is a slight change in patterns.

Anomaly isolation

Having removed all of the previously known anomalies, we can now move on to the main

observation in this paper. We identify the last the day of the month. A day when there is

substantial trading activity, reporting requirements and transparency insights into most equity

market portfolios. There is a fair amount of pressure to get the portfolio in line, adjust beta-

exposures, jettison losers and add to winners. All of this, can be described as window dressing

and if executed with intent, market manipulation. Researchers (Carhart, Kaniel, Musto, & Reed,

2002) have identified a price pressures for the last day of the month, generating excess returns

over that period. The author noted especially high returns so small-cap equities.

We break out the last day in the month and analyze the average pattern around the event.

The last day of the month, generates negative returns, but the returns are driven by the last hour

of trading, rather than from a slow drift during the day (Figure 5). From 15:00 (EST) there is a

sharp change in the average market behavior.

On average, equity market declines by approximately 20 bp per month-end. This should

be contrasted to a small gain on any non-month-end day as previously observed. It should also be

judged against the speed of the decline. The decline is relatively rapid. We should remember that
TURN-OF-THE-MONTH: WINDOW DRESSING BEHAVIOR 9

this is the average behavior of the market and that there is significant deviation around the

average.

The potential profit is a tradeable effect and can be exploited even for institutions with

poor execution infrastructure. The cumulate effect is displayed in Figure 6 and shows that the

effect is rather persistent but does not depend on a single event.

The effect has been persistent over more than a decade and with a positive relationship

compared to equity market volatility. The effect does not seem to exhibit any structural changes

nor short term dependency on prior equity market direction.

Small cap comparison

Repeating the same calculations, but with the Russell 2000, to capture a potential small-

cap bias. We find that the intraday behavior is approximately similar, which is as expected given

that there is a high correlation between the two indices.

Since the Russell 2000 has higher volatility compared to the S&P 500, the effect is more

pronounced for the former index (Figure 7).

Discussion

We have uncovered a previously undiscovered end-of-month effect, where stocks have a

comparatively large and persistent negative return in the last hour of trading. The size of the loss

in the last hour is significant, exceeds transaction cost estimates and is economically meaningful.

As always, there are an infinite number of explanations for the phenomena. Suspected

window dressing, but probably not conducted by a single entity but rather by a large swarm of

asset managers behaving in the same way. This behavior would fall in line with prior research,
TURN-OF-THE-MONTH: WINDOW DRESSING BEHAVIOR 10

stating that there is a tendency for fund managers to eject losers and add to winners, that is,

window dressing.

If this is true, this could imply that winners are added first (the pleasant task) and fund

managers are only taking care of the unpleasant tasks (selling losers) when any hope of a

rebound is gone. This may suggest that fund managers are believing that losers have a “lottery-

ticket” characteristic and that it is worth holding onto until they are forced to sell the shares. The

reasons for why they are forced to sell shares are many, chiefly career risk and client reporting

issue. The coordinated selling at the end of the day, drives down the index as there are fewer

portfolios willing to assume the risk of holding losers over the month-end.

A portfolio manager that is reporting a losing portfolio is likely to face career risk as

clients would demand more and more explanations about the realized underperformance. The

underperformance from the losing shares, has already happened. Seeing a cleaned-up portfolio of

winners might sway the client in the direction that things have actually changed for the better and

the portfolio is back on track.

The behavior may also relate to over and under leverage. If a particular fund is leveraged

or has a beta-risk that is not compatible with risk guidelines, it is forced to sell or buy shares,

getting the portfolio back in line with risk. There is anecdotal evidence that equity hedge-funds

have more beta risk intra-month than reported at the month-end. In line with the lottery-ticket

assumption, fund managers would only sell the option of a rebound as late as possible.

Other leveraged instrument, such as leveraged ETFs may also add fuel to the effect as

noted in prior research (Bai, Bond, & Hatch, 2012).

The realized behavior implies that fund managers have the wrong assumption or asses the

optionality value in losing positions too high. A fund manager should thus start with the
TURN-OF-THE-MONTH: WINDOW DRESSING BEHAVIOR 11

unpleasant task of reducing losers only adding to winners as late as possible throughout the

trading session. The current behavior seems to be to delay unpleasant activities.

Alternatively, explanations include that risk control is stricter at reporting periods and it is

important to be in-line with guidelines and restrictions. If this is true, we would be able to detect

additional price pressure into periods that coincide with internal and external transparency.

References

Bai, Q., Bond, S. A., & Hatch, B. (2012). The Impact of Leveraged and Inverse ETFs on

Underlying Stock Returns. University of Cincinnati: Department of Finance.

Carhart, M., Kaniel, R., Musto, D., & Reed, R. V. (2002, April). Leaning for the Tape: Evidence

of Gaming Behavior in Equity Mutual Funds. Journal of Finance, pp. 661-693.

Cooper, M., Cliff, M. T., & Gulen, H. (2008). Return Differences between Trading and Non-

Trading Hours: Like Night and Day. SSRN. Retrieved from

http://ssrn.com/abstract=1004081

Failed Bank List. (n.d.). Retrieved from FDIC:

https://www.fdic.gov/bank/individual/failed/banklist.html

Josef, L., & Smidt, S. (1988, Winter). Are Seasonal Anomalies Real? A Ninety-Year Perspective.

The Review of Financial Studies, 403-425.

Lakonishok, J., Shleifer, A., Thaler, R., & Vishny, R. (1991). Window Dressing by Pension Fund

Managers. NBER. Retrieved from http://www.nber.org/papers/w3617.pdf

Nilsson, L. (2015). The Pre-FOMC Drift Explored. SSRN. Retrieved from

http://ssrn.com/abstract=2640477
TURN-OF-THE-MONTH: WINDOW DRESSING BEHAVIOR 12

Schwert, W. G. (2002, October). Anomalies and Market Efficiency. University of Rochester -

Simon Business School ; National Bureau of Economic Research (NBER). Retrieved

from http://papers.ssrn.com/sol3/papers.cfm?abstract_id=338080
TURN-OF-THE-MONTH: WINDOW DRESSING BEHAVIOR 13

Figures

Figure 1 Indexed returns from the non-active and active trading session. Active is defined as

9:30 to 16:15 Eastern Standard Time (EST). Source: CME, Author’s calculations. For all

calculation and charts throughout this report, we will be using returns from the S&P 500 E-mini

future from inception in 1997 to mid-2015. The S&P 500 Index, using available providers does

not provide a good open price. The open price is too similar to the close price and does not

mirror the concurrent price action available from equivalent futures. This prevents longer time

series analysis and we will have to do with the data from 1997.
TURN-OF-THE-MONTH: WINDOW DRESSING BEHAVIOR 14

Figure 2 Indexed weekend returns, measured from last trade to the first 15 minutes of the new

week. Note regime changes that seems to have occurred in 2007/2008. The effect is too small

trade, even with advanced knowledge about the sign. Source: CME, Author’s calculations
TURN-OF-THE-MONTH: WINDOW DRESSING BEHAVIOR 15

Figure 3 Average Intraday behavior. Source: CME, Author’s calculations

Figure 4 Intraday Behavior for negative and positive equity years. Source: CME, Author’s

calculations
TURN-OF-THE-MONTH: WINDOW DRESSING BEHAVIOR 17

Figure 5The last hour before the month-end has significant negative returns. Source: CME,

Author’s calculations
TURN-OF-THE-MONTH: WINDOW DRESSING BEHAVIOR 18

Figure 6 Cumulated returns from being short the last hour at month-ends. Source: CME,

Author’s calculations.
TURN-OF-THE-MONTH: WINDOW DRESSING BEHAVIOR 19

Figure 7 Russell 2000 compared to S&P 500 on the last day of the month. Source: CME,

Author’s calculations.