MARKET STRUCTURE BRIEFING NOTES August 31, 2010

Transcription

1 Volume Profiles: A Curve By Any Other Name Matthew Black Kyle Burgess MARKET STRUCTURE BRIEFING NOTES August 3, 2 The most important factor in achieving the VWAP benchmark is matching the volume profile - if we could perfectly predict the volume profile, we could significantly narrow slippages. The question is how likely are we to accurately forecast the volume profile? Is it even feasible to predict the volume profile accurately by looking at historical trading patterns? Our analysis suggests it may be necessary to accept that one cannot accurately forecast the volume profile based on historical profiles. Specifically, although the average profile may be quite stable, there is a significant amount of noise around this average profile that explains the vast majority of the miss from the VWAP benchmark during execution. Consequently, we find it unnecessary to go through the extra effort of creating stockspecific volume profiles, since a general market-wide U-shape pattern will give results that are statistically indistinguishable. METHODOLOGY To test, we compared the performance of stock-specific, sector-specific, and marketwide volume profiles. Stock-specific profiles were assembled using the average volume profile for the stock over the last 6 days. Sector-specific profiles were the volume-weighted average across all stocks in the sector over the last 6 days. The market-wide profile was the volume-weighted average across a universe of about 2, stocks over the last 6 days. ρ i days volume j= = totalvolume EXAMPLE j totalvolume = int ervals days ( i= j= volume ij ) The diagram below compares the stockspecific profile for CM against the sectorspecific profile for Financials and the marketwide volume profile. Notice the stock-specific profile is more noisy and steep, as opposed to the smooth and shallow market-wide profile.

2 EVALUATION We tested against the stocks with the highest average daily value traded over the last 6 days, because this most liquid set of stocks is most likely to exhibit predictable stock patterns. Less liquid names will be even less predictable. Results were gathered across,23 total observations. Specifically, for each volume profile we: ) assume we execute at the VWAP within each interval, 2) calculated realized average price over the trading day using the schedule, 3) take the average absolute miss of the realized price above from the actual VWAP for the day. specific profile is better and results in a lower average absolute VWAP miss - Accept the hypothesis that the stock specific profile is no better than the sector profile, and reject the alternative hypothesis that the stock specific profile is better and results in a lower average absolute VWAP miss RESULTS Across the stocks, the average absolute VWAP misses for the different profiles, in basis points, were: Profile VWAP Miss Standard in bps Deviation Stock Specific.293 ±5.73 Sector Specific.567 ±5.9 Market Wide.335 ±5.87 The standard deviations are large relative to differences in average performance against VWAP, and we can, with over 95% confidence: - Accept the hypothesis that the stock specific profile is no better than the market-wide profile, and reject the alternative hypothesis that the stock specific profile is better and results in a lower average absolute VWAP miss - Accept the hypothesis that the sector specific profile is no better than the market-wide profile, and reject the alternative hypothesis that the sector These tests allow us to conclude that no single profile outperforms any of the others with any statistical significance. The more complex stock and sector-specific profiles do not outperform the simpler market-wide profile 2. The stock-specific profiles come close to improving on the sector-specific profiles - but not quite; they are essentially equal to the market-wide profile. What if the observations for average VWAP miss were exact with no deviation? What if we rejected the hypotheses and concluded that the stock-specific profiles did outperform the sector-specific profiles by an average of.27bps? This still only represents an improvement of 2.4% over the average sector profile miss of.57bps - miniscule compared to the large standard deviations. Likewise, if we accepted that the stockspecific profile gave a.42bps improvement over the U-shaped market profile, this only represents an insignificant.37% improvement. Assuming VWAP is achieved in each interval implies our micro-order placement is no better or worse than the average trader. This focuses our analysis on the shortfall from VWAP that can be attributed to errors in forecasting the trading schedule = 29bps 23 We need a.29% difference to accept hypothesis with 95% confidence that one method is better than the other 2

3 ANALYSIS From the above profiles we can see that the sector and stock-specific profiles are very close, while the market-wide profile is basically a smooth U-shape. We propose that the use of a general U-shape will perform just as well as stock-specific profiles, and that the vast majority of the observed miss relative to VWAP is the result of unpredictable deviations about the expected profile at least when previous day s trading patterns are the basis of prediction. As a specific example we looked at the volume profiles for CM. The ACFs (autocorrelation functions) below describe the historical realized deviations from the stockspecific, sector-specific, and market-wide volume profiles (realized profiles after the respective historical profile has been subtracted). This is done to investigate whether there are further patterns in historical trading data to be exploited. The stock-specific profile, when subtracted from the historical realized volume profiles leaves essentially a white noise pattern, implying that only unpredictable deviations remain. Therefore, the stock-specific profile makes the best use of previous days trading patterns, leaving essentially nothing but unpredictable white noise (which obviously cannot be used in a trading strategy). Note, however, that significant lags do exist in the ACF intraday, i.e. lags less than 78 (see Further Research). The sector-specific and market-wide profiles leave a non-white noise pattern with a noticeable seasonal pattern, implying there are predictable deviations that remain Stock-Specific ACF This seasonality pattern, however, does not seem to harm performance - in terms of average absolute deviation or standard deviation of errors - in any significant way. This is because the unpredictable deviations about these patterns account for the vast majority of the realized shortfall from VWAP Sector Specific ACF Market-wide ACF The stock-specific ACF has only 5 significant lags between interval 78 (one trading day) and interval 2 - no more than would be expected in a random sample (i.e. 5% or of every 2 lags will have a significant correlation in a random sample). This implies that at a 95% confidence level, there are no significant lags beyond 78 (previous day) that could be used to improve our forecast of today s volume profiles. In this case the stock-specific volume profile had an average absolute VWAP miss of around 3.2bps, +/- 63.6bps. By comparison, the sector-specific and market-wide profiles seem to still have some seasonal pattern existing that could be extracted, however, their statistical performance suffers very little, if at all, as a result. The performance of the sector-specific and market-wide profiles in this case are 3.bps +/- 66.4bps and.5bps +/- 6.9bps. Given the above average deviations and their standard deviations, for one method to definitively improve performance it must 3

4 improve execution by almost basis points in order to be statistically significant against the 44 daily observations used in this analysis 3. None of the profiles vary from the others significantly enough for this to be the case. Once again, even assuming the estimates above are correct, the expected improvement in execution performance of any one method over another is small in percentage terms and miniscule compared to the standard deviations. CONCLUSIONS From this we conclude that a general U- shaped market-wide profile - although seemingly suboptimal - performs just as well as the more involved stock-specific profile, and that random intraday deviation from a general U-shaped pattern is the biggest contributor to negative performance versus the VWAP. Clearly, there is only so much information to gain from using previous days trading patterns to predict today s trading patterns. Using a basic U-shape (a market-wide profile) is very simple and provides the same performance as sector or even stock-specific profiles. FURTHER RESEARCH In the above ACFs, the same-day lags (i.e less than 78 5-minute bars) could have been used to in an attempt to better predict the current day s volume profile. Therefore, monitoring and updating the profile intraday and adapting to the unpredictable deviations from the expected profile is the best way to gain meaningful improvement in execution performance. This leaves us with three main avenues to explore in order to improve VWAP execution: ) How to update volume profiles intraday in order to execute closer to the actual volume profile? 2) How much improvement can we achieve by having better order placement logic? i.e. beating the VWAP in each interval? 3) Are there other explanatory variables that can explain deviations from the general U- shape pattern? From our analysis of ACFs and actual performance numbers we can see that the majority of the miss from the actual VWAP using any of the above profiles is caused by unpredictable intraday variations in the expected trading profile. For example, even though the market-wide and sector-specific profiles left significant lags in the ACF, this did not hinder their performance. THOMAS KALAFATIS Executive Director - Business Strategy Development (46) thomas.kalafatis@cibc.ca TIM COVEYDUCK Product Director - Client Interface (46) tim.coveyduck@cibc.ca LUKA MARJANOVIC Product Director Liquidity (46) luka.marjanovic@cibc.ca KYLE BURGESS Product Director Execution Management (46) kyle.burgess@cibc.ca ALI LOTIA Viper Development (46) ali.lotia@cibc.ca PHIL SHAM Algorithm Development (46) phil.sham@cibc.ca MATT BLACK Quantitative Analysis (46) matthew.black@cibc.ca = 9.8bps 44 We need.% difference to accept hypothesis with 95% confidence that one method is better than the other 4

5 Legal Disclaimer: This publication is distributed by (i) in Canada, CIBC World Markets Inc., a member of the Investment Industry Regulatory Organization of Canada, the Toronto Stock Exchange, the TSX Venture Exchange and CIPF, (ii) in the European Union, CIBC World Markets plc, which is regulated in the United Kingdom by the Financial Services Authority, (iii) by CIBC World Markets affiliates in other jurisdictions in compliance with local rules and regulations (collectively, "CIBC World Markets"). This publication is provided for general informational purposes only to institutional and professional investors and does not constitute an offer or solicitation to buy or sell any specific securities or other investments discussed herein. This publication should not be construed as a research report. Any products or services mentioned herein are made available only in accordance with local legal and regulatory requirements. The information contained in this publication is not intended to constitute a "report, statement or opinion of an expert" for purposes of the securities legislation of any province or territory of Canada nor does it constitute a research report as defined in the rules of the Investment Industry Regulatory Organization of Canada nor is any of the information contained in this publication intended to be and it should not be viewed as advice or a recommendation. The information, opinions and statistical data contained herein has been obtained from sources (internal and/or external) that CIBC World Markets believes to be reliable (without having conducted any independent investigation). CIBC World Markets assumes no obligation to update any information, opinions, statistical data or forward-looking statements contained herein for any reason, including if CIBC World Markets subsequently learns that such information is inaccurate, incomplete or otherwise in error or to notify any person in respect of thereof. CIBC World Markets does not represent or warrant the completeness, accuracy or currency of this publication or any information, opinions and statistical data contained herein, including the future performance of any security, investment or strategy mentioned in this publication, nor is it intended to be a complete statement or summary of the securities, markets or developments discussed herein and the information contained herein should not be relied upon as such. Any opinions, estimates and projections contained herein are subject to change without notice, and are provided by CIBC World Markets in good faith but with no legal responsibility or liability whatsoever. Changes in assumptions may have a material impact on any returns detailed. The securities and other investments listed reflect indicative terms only and may change in price and yield and market conditions. The securities, strategies, and investments mentioned in this publication may not be suitable for all types of investors. This publication does not take into account the investment objectives, financial situation or specific needs of any recipient or particular investor. Recipients should consider this publication as only a single factor in making an investment decision and should not rely on investment recommendations contained herein (if any) as a substitution for the exercise of independent judgment of the merits and risks of investments. Nothing in this publication constitutes legal, accounting or tax advice, and recipients should independently evaluate the financial, market, legal, regulatory, credit, tax and accounting risks and consequences involved, and specific to the recipient's situation, before entering into a transaction. Past performance is not a guarantee of future results, and no representation or warranty, express or implied, is made regarding the future performance of any security or investment mentioned in this publication. The price of the securities and other investments mentioned in this publication and the income they produce may fluctuate and/or be adversely affected by exchange rates, and investors may realize losses on investments in such securities, including the loss of investment principal. CIBC World Markets accepts no liability for any loss arising from the use of information contained in this publication, except to the extent that liability may arise under specific statutes or regulations applicable to CIBC World Markets. CIBC World Markets and its affiliates may have an investment banking or other relationship with some or all of the issuers mentioned herein and may trade in any of the securities and investments mentioned herein either for their own account or the accounts of their customers. CIBC World Markets and its affiliates may at any time deal as principal in or have a long or short position in any security or investment mentioned herein, or any option, future or other derivative based thereon. Any products or services mentioned herein are made available only in accordance with local legal and regulatory requirements. As a result, the securities and investments discussed herein may not be eligible for sale in some jurisdictions. CIBC World Markets Inc., and CIBC World Markets plc are wholly owned subsidiaries of Canadian Imperial Bank of Commerce, ("CIBC"), but each is solely responsible for its contractual obligations and commitments. Any securities products offered or recommended to or purchased or sold in any client accounts (i) will not be insured by the Federal Deposit Insurance Corporation, the Canada Deposit Insurance Corporation or other similar deposit insurance, (ii) will not be deposits or other obligations of CIBC, (iii) will not be endorsed or guaranteed by CIBC, and (iv) will be subject to investment risks, including possible loss of the principal invested. The CIBC trademark is used under license. This report may provide addresses of, or contain hyperlinks to, Internet web sites. CIBC World Markets has not reviewed the linked Internet web site of any third party and takes no responsibility for the contents thereof. Each such address or hyperlink is provided solely for the recipient's convenience and information, and the content of linked third party web sites is not in any way incorporated into this document. Recipients who choose to access such third-party web sites or follow such hyperlinks do so at their own risk. 2 CIBC World Markets. All rights reserved. Unauthorized use, distribution, duplication or disclosure without the prior written permission of CIBC World Markets is prohibited by law and may result in prosecution. 5