Executive summary
This article examines trading behaviour in seven equity markets[1] with the aim of identifying evidence of possible 'window dressing'. Window dressing is a deliberate strategy of price manipulation where the price of a security is increased significantly at the end of the trading day, in particular at month and quarter ends.
The basic motivation for window dressing, often attributed to fund managers, is to increase the price of securities in which their funds have relatively high exposure, in order to improve the performance of their funds under management. Given that fund performance is often measured using the month-end closing price, we focus here on these days and times. While it is recognised that there are incentives for window dressing on alternative days, such as option and future expiry days, an examination of these days is left to future research.
We examine the period April 2000 to September 2000 and concentrate on the top 150 securities by trading activity. Notwithstanding the exclusion of smaller securities, this represents between 72% (Nasdaq) and 100% (Oslo Stock Exchange) of the market trading activity. The liquidity of the securities examined varies significantly across the different markets. The Nasdaq is the most liquid with an average of 5,918 trades per day for the sample and the Oslo Stock Exchange is the most illiquid with an average of 63 trades per day for the sample.
Our main findings are:
- The Australian Stock Exchange has the highest incidence of window dressing over the period with an average of almost 10 securities being window dressed each month, although it also displays the lowest average percentage change in price for these securities.
- Hong Kong Exchanges exhibits the lowest incidence of window dressing, with an average of three securities being window dressed each month.
- The Oslo Stock Exchange displayed the highest average return for window dressed securities with an average value of 8.65%.
- The Nasdaq showed an equal number of instances of window dressing on month-end and non-month-end days suggesting that further research should consider other incentives for window dressing in this market.
- Window dressing is not typically concentrated in a particular group of securities. In the eight markets examined there are only five securities that were window-dressed on more than two occasions.
- Window dressing is relatively evenly distributed across the securities with different levels of liquidity. However, this may be due to the fact that only the top 150 most liquid stocks in each market have been included in the analysis.
Introduction
This article examines prima facie evidence of window dressing across seven different securities markets: the Australian Stock Exchange (ASX), Hong Kong Exchanges and Clearing (HKEX), the London Stock Exchange (LSE), the Nasdaq Stock Market (Nasdaq), the Oslo Stock Exchange (OSE), the Singapore Exchange (SGX), and the Toronto Stock Exchange (TSE).
There is no legal or generally accepted definition of window dressing. However, the term is commonly used to describe situations where the price of a stock is increased significantly at the end of a trading day, particularly at the end of a month. Window dressing is typically thought to occur as a result of the actions of fund managers attempting to increase the price of securities in which they have a large exposure. The incentive to manipulate comes as a direct consequence of fund and fund manager performance being assessed at this time[2].
We define window dressing as a situation where the increase in the price of a stock in the last 15 minutes of the trading day is greater than the top 1% of price changes for that stock during the previous month.
The high level of concern about window dressing is apparent in recent decisions by a number of exchanges to implement special mechanisms for the determination of closing prices. These mechanisms are typically aimed at making it more difficult for investors to influence the closing price. For example, the ASX and SGX have implemented closing call auctions. An alternative approach is to use a special algorithm to determine the closing price rather than using the last trade price. The HKEX and the Madrid Stock Exchange take this approach[3]. The New Zealand Stock Exchange takes a slightly more novel approach and closes the market randomly between 15:25 and 15:30 each day. The effectiveness of these mechanisms is an interesting source of research[4]. In particular it would be worth studying whether these changes have any effect on the practice of window dressing. However, here we simply attempt to confirm whether the incidence of the practice is apparent.
We examine differences in the frequency of window dressing across various markets, in particular focusing on whether window dressing occurs more frequently on month-end days. We also consider whether particular securities are more susceptible to window dressing. Future research should extend the time period examined in order to consider whether securities are more likely to be manipulated in particular months (i.e. quarter ends). It might also seek to gain access to trust deeds to establish the precise dates and mechanisms used to assess fund and fund manager performance. This would allow a closer examination of the impact of fund managers on particular dates.
Data
Our data (from the period March-September 2000) is obtained from a Reuters database maintained by the Securities Industry Research Centre of Asia-Pacific (SIRCA)[5]. March is used to generate descriptive information about the markets such as average trading activity and typical trading periods. Window dressing is examined during the period April to September.
Descriptive statistics
The analysis is limited to the top 150 securities by trading activity each month, primarily because it is unlikely that holdings outside these securities would be significant enough (if manipulated) to influence the average return on a portfolio[6]. Table 1 provides details of the percentage of market trading activity that this sample represents. The OSE provides the greatest coverage of the market with the whole market being captured. The Nasdaq sample captures approximately 72% of the market[7].
Table 1: Summary of market and sample trading activity during March 2000
An examination of the number of trades per day in each market was undertaken in order to identify any obvious differences in liquidity in the securities used in the analysis[8]. Table 2 summarises these results. This shows that Nasdaq is the most liquid market with an average of 5,918 trades per day. Worthy of note is that the Nasdaq market is quite liquid even in quintile 5. This suggests that these securities are less likely to be subject to window dressing. The OSE has the lowest trading activity with an average of 63 trades per day. The analysis in the OSE may be problematic in quintiles 4 and 5 due to the low numbers of trades per day.
Table 2: Average number of trades per day during March 2000
Intra-day patterns
Prior to investigating window dressing, we looked at intra-day patterns in trading activity in order to develop a visual appreciation of obvious differences in trading activity in the markets being studied. Figure 1 displays the intra-day patterns in relative volume in each market[9]. It shows that all markets exhibit increased trading activity at the end of the day. With the exception of ASX, TSE and Nasdaq, the volume traded is highest during the last 10 minutes of the trading day. It should be noted that the ASX holds a closing call auction between 16:05 and 16:06 each day. On average over 3% of daily activity is traded at the call. Like the other markets, the last 10 minutes of normal trading is the most active on the ASX.
Method
The method of analysis used identifies particular securities and dates on which window dressing occurs. Due to the fact that window dressing does not have any legal or generally accepted definition a number of assumptions must be made about the conditions that need to be met in order for a security to have been classified as 'window dressed'.
In general terms, window dressing occurs when the price of a security increases significantly at the end of the day, in particular at month and quarter ends. In order to identify specific cases of window dressing, it is necessary to specify what constitutes a 'significant increase' in price and to identify the most appropriate period over which to examine these price changes.
In this study, the last 15 minutes of the trading day is used to examine price changes. In order to determine what constitutes a significant price rise, historical trading activity of each security is analysed to identify outliers. A price change is considered to be significant if it is greater than the top 1% of price changes during a benchmark period. In this study, the benchmark period is the trading activity during the previous month. For example, trading in April is compared to a March benchmark. This research design is perhaps best explained through the use of an example.
In order to identify the top 1% of price changes for a security during the benchmark period, returns (price changes) are sampled every 15 minutes during the day. Assuming that there are approximately 20 trading days in a month and twenty-four 15-minute intervals in each trading day (assuming 6 trading hours per day), each month should provide approximately 480 return observations. If these observations are sorted, the largest 4.8 returns (or 1% of the distribution) constitute what we have earlier described as 'significant price changes'. The value of the fifth return is where the threshold for window dressing for that security is set. For example, if the fifth highest return for BHP was 0.97% during March, then the security is deemed to have been subject to window dressing if the return in the last 15 minutes of 30 April was greater than 0.97%.
We compare the number of securities that are window dressed on month-end days compared to the average for non-month-end days. These results are examined to determine whether window dressing is more common in particular quintiles. We also consider whether particular securities are more susceptible to window dressing.
Results
Market comparisons
Table 3 summarises the instances of window dressing on month-end and non-month-end days in each market. This table shows that the ASX has the highest incidence of window dressing on month-end days both in an absolute sense and relative to the number on non-month-end days. An average of almost 10 securities are window dressed at month-end on the ASX. Interestingly, however, the results also show that the average increase in price for window dressed securities is lowest on the ASX.
The LSE exhibits the second highest incidence of window dressing, with an average of almost 9 securities being manipulated on month-end days, compared to approximately 7 on non-month-end days. The results for the Nasdaq, OSE and SGX indicate that approximately 5 securities are window dressed on month-end days. The OSE exhibits the largest price increases on these days with an average increase of 8.65%.
The TSE and HKEX display the lowest incidence of window dressing with an average of approximately 3 securities per month-end day. The average increase in price on these days is higher on the HKEX with an increase of 3.65% compared to only 2.28% on the TSE.
Table 3: Window dressing summary
Security concentration
Table 4 illustrates the concentration of window dressing across different securities. The HKEX and TSE display little concentration. On the TSE there was only one instance of a security being window dressed more than once and on the HKEX there were none. The ASX shows the highest level of concentration with 3 securities being window dressed on three occasions and 5 on two occasions.
Table 4: Concentration of window dressing in particular securities on month-end days
Although there is little evidence of window dressing being concentrated in particular securities at month-end, there is evidence of securities being window dressed repetitively within months. For example, on the ASX, one stock was window dressed four times during September 2000. This evidence suggests that there are many other dates and reasons why window dressing is occurring than we have been able to address. Investigation of this is left to future research.
Security liquidity
To determine whether the incidence of window dressing is influenced by a security's liquidity, the sample of 150 securities is divided into quintiles based on trading activity. Quintile 1 is the most liquid group of securities and quintile 5 is the least liquid. It is anticipated that window dressing will occur more frequently in the less liquid securities as they are easier and cheaper to manipulate. However, fund managers would have to have significant proportions of their portfolio in these securities, and/or the price would have to be moved substantially, to provide suitable incentives. Fund managers are more likely to be responsible for window dressing in index and highly liquid stocks.
Table 5 shows the percentage of window dressing cases that occur in each quintile. Contrary to expectations, there is no evidence of window dressing occurring more frequently in less liquid securities.
Table 5: Window dressing by quintile
For markets with a large number of securities traded, such as the Nasdaq and LSE, this analysis may not cover enough securities to find differences in behaviour. In these markets, all of the securities examined are relatively liquid. Future research should consider whether the frequency of window dressing differs in securities with lower levels of liquidity.
Results by market
Australian Stock Exchange
Table 6 shows details of the incidence of window dressing on the ASX. These results indicate that the incidence of window dressing is particularly high in September, with 21 cases occurring. The average increase in price for window dressed securities is 1.87% compared to an average price increase of 0.11% on month-end days for securities not subject to window dressing.
Table 6: Summary of ASX window dressing
Hong Kong Exchanges and Clearing
Table 7 summarises the incidence of window dressing on the HKEX. HKEX had the lowest incidence of window dressing, with an average of 3 securities per month.
It should be noted that HKEX uses a special algorithm to calculate the closing price. It is calculated using the median of five nominal price values taken during the last minute of trading at fifteen-second intervals (i.e. 15:59:00, 15:59:15, 15:59:30, 15:59:45 and 16:00:00)[10]. As a result, investors hoping to influence the closing price must influence the price throughout the last minute, rather than just the last trade.
The low incidence of window dressing may be due to the effectiveness of this mechanism or it may be due to the failure of our measure to capture the true closing price. Further tests conducted using the quoted closing price should be undertaken to distinguish between these alternative explanations.
Despite this issue, the average increase in price for the window dressed securities is quite high. These securities display an average increase of 3.65% compared to only 0.02% for other securities on month-end days.
Table 7: Summary of HKEX window dressing
London Stock Exchange
The frequency of window dressing on the LSE is shown in Table 8. The high incidence of window dressing is driven by the results for July, August and September. The difference between the average number of window dressed securities on month-end and non-month-end days is quite low. Further investigation is required to determine the incentives for influencing the closing price of a security on days other than month-end. It is also noteworthy that the average price increase on these is significantly higher on the non-month-end days, with an average increase of 15.57% compared to only 2.87% on month-end days. This large increase is driven by outliers in April and September. On two days during these months there was a stock which closed 10,000 and 459% higher than previous prices. These increases were probably driven by broker error (for example, a price of 77,331 was entered instead of 77.331). When these instances are excluded from the sample, the average non-month-end price increase is reduced to 4.96%.
Table 8: Summary of LSE window dressing
Nasdaq Stock Market
A summary of the incidence of window dressing on the Nasdaq is provided in Table 9. This indicates that window dressing is just as likely to occur on non-month-end days as month-end days. Like the LSE, the average price change in window dressed securities is higher on non-month-end days. Once again, further investigation is required to determine incentives for influencing price on non-month-end days. It would be worth investigating trust deeds to see whether there is something different about the mechanisms used to measure fund and fund manager performance.
Table 9: Summary of Nasdaq window dressing
Oslo Stock Exchange
Table 10 displays the frequency of window dressing on the OSE. The highest number of cases of window dressing occurred at the end of September. The average price increase in window dressed securities is 8.65% on month-end days compared to 4.24% on non-month-end days. This is the highest month-end increase across all the markets studied.
Table 10: Summary of OSE window dressing
Singapore Stock Exchange
Table 11 shows the number of cases of window dressing each month on the SGX. Note that the highest number of cases of window dressing took place in May, when 11 securities were window dressed. The increase in prices was also greatest during May, with an average increase of 7.8%. There was also an increase in the number of cases of window dressing on non-month-end days during May.
It is interesting to note that the SGX introduced a closing call auction on August 21, 2000. One of the objectives of the call was to reduce the possibility of closing price manipulation. Notwithstanding, it is noteworthy that the number of cases of window dressing in September is high compared to the other months. Further investigation is required to determine the impact of the introduction of the closing call on window dressing.
Table 11: Summary of SGX window dressing
Toronto Stock Exchange
Table 12 shows that the TSE has one of the lowest instances of window dressing. An average of only 3.5 securities per month are window dressed. September exhibits the highest incidence of window dressing.
Table 12: Summary of TSE window dressing
Conclusions and future research
We have defined and provided prima facie evidence of the frequency of window dressing across markets. The ASX shows the highest rate of window dressing with an average of just under 10 securities per month identified. However, the ASX also displays the lowest percentage price change for window dressed stocks. The lowest rate of window dressing is found on HKEX. However, further attention should be given to HKEX due to the special algorithm used to calculate closing prices.
Against expectations, window dressing occurs relatively uniformly across securities with different levels of liquidity.
Given that only five months of trading activity were examined, it is not possible to comment on whether particular months are more susceptible to window dressing. Further research should extend this analysis to determine whether month and quarter ends are more likely to exhibit window dressing. This would also identify any trends in window dressing over time and how it has changed in response to regulatory action.
This research should also be extended to consider the sensitivity of the results to the definition of window dressing used here. For example, how would the results differ if an absolute price change was used to identify window dressing rather than one based on a security's previous trading activity?
Future research should also examine the impact of the introduction of the closing call auctions on the frequency of window dressing on the SGX. Further, a longer time series of data is required to determine whether particular months are more susceptible to window dressing. This would also identify how patterns in window dressing have changed over time and in response to actions by regulators. Second, research should consider the trust deeds of fund managers to determine their incentives for price manipulation at different times of the month. Third, future research should also consider the impact of changes in the mechanisms used to determine the closing price on the frequency of window dressing.
Professor Michael Aitken is CEO of the Capital Markets Co-operative Research Centre and Chair of Capital Market Technologies at the University of New South Wales.
Dr Carole Comerton-Forde is Research Manager at the Securities Industry Research Centre of Asia Pacific. She is the corresponding author for this article and can be contacted at: Level 1, 275 George Street Sydney NSW 2000 Phone: +61 2 8296 7826 Fax: +61 2 9299 1830 Email: carole@sirca.org.au
Reference
Aitken, M, and Comerton-Forde, C, (2000) Do closing call markets improve market efficiency?, Working Paper, Securities Industry Research Centre of Asia-Pacific.
Notes
[1] The choice of the seven markets was driven by SMARTS Pty Ltd (www.smarts.com.au), the party providing the research funding. Regulators from other markets who would like their market analysed using the same computer routines (over any period) are invited to contact the authors.
[2] Our concentration on fund managers is driven by their significant and growing proportion of the overall market.
[3] The HKEX calculates the closing price using the median of five nominal price values taken during the last minute of trading at fifteen-second intervals. The Madrid Security Exchange uses the value weighted average price of the last 500 shares traded. In addition, if the absolute price change in the last minute before the close exceeds a certain range, the closing price is calculated using the value weighted average price in the last five minutes of trading.
[4] For example, an examination of the introduction of the closing call auction on the ASX indicated a significant increase in liquidity and reduction in volatility at the end of the day (Aitken and Comerton-Forde, 2000).
[5] A comparison of results obtained using the Reuters data and data obtained directly from two exchanges (ASX and OSE) provides generally consistent, but not identical, results. Our results should therefore be considered in this context. We acknowledge and thank Reuters and SIRCA for access to this data.
[6] Given that it is easier to manipulate securities that have lower liquidity, there is a greater likelihood of smaller securities being the subject of manipulation. However, this is not likely to be applicable to fund managers, who shun such securities because of the risks (primarily the transaction costs) of getting in and out of them.
[7] In some markets products other than equities are traded. A number of assumptions have been made about what codes relate to equities. Where these assumptions are incorrect the paper may not have captured the whole market. The number of securities identified by us has been displayed in order to highlight any problems with our assumptions.
[8] We accept that number of trades is only a rough proxy for liquidity.
[9] Relative volume is calculated as the volume traded in each interval divided by the total volume traded during the day. This measure, rather than absolute volume, is used to facilitate comparisons across the markets, which have significantly different trading volumes.
[10] The nominal price is calculated using the following rule. If no trades have occurred then the true price is equal to the midpoint price (i.e. the midpoint of the best bid and ask). Where there is no midpoint price, the nominal price is equal to the best bid or ask. Where there are no bids or asks in the schedule then the nominal price is undefined. If trades have occurred and the last traded price lies between the current best bid and best ask, the nominal price is equal to the last trade price. If the last trade price lies above (below) the current best ask, then the nominal price is equal to the best ask (bid).
