1 Committee member Axel Rudolph recently interviewed Jack Schwager at the Technical Analysis Trading Forum in Washington DC where he was presented with the Traders Hall of Fame Award for his contribution to technical analysis. While researching his books, Schwager interviewed over 50 of the worlds most successful hedge fund managers and found a crucial common thread: there is no place for emotion. Truly great traders retain their objectivity. Thats what sets them apart from most investors, he asserts. Some traders will determine beforehand how a trade should run and will place loss-limiting orders accordingly. Others will exit, or reduce their exposure if they become unsure about it. Either way the emotion is stripped out. As long as you are in a position you cannot think straight, Schwager recalls, as he thinks back to a time when he ran a profit before giving most of it back again. Naturally, I wanted most of all to know from him what the winning methods and common traits of the most successful traders are. There is no single true path, he declared. The method is the least important part. From his roots as a pure fundamentalist Schwager has, over the years, come to favour the technical approach at least in his own trading arena futures. But Schwager is ready to admit that both approaches have the ability to give people an edge. Many of the Market Wizards he has interviewed have made tremendous returns relying solely on fundamental analysis. But Schwager is convinced by hard evidence that pure technical analysis can consistently yield profits so far in excess of losses that there can be no question of a statistical fluke. You can make an empirical, statistical argument that technical analysis works. Most people may lose using technical analysis but the fact that any of them can get such an amazing outweighing of gains to losses would argue that there is something there. Some hedge funds which consistently make annual returns over 20% are right less than 50% of the time. Its not a question of the percentage of times you win. Its a matter of what the expected gain is compared with the expected loss, he points out. But Schwager also cites a contrasting example of one of the worlds most successful S&P stock futures traders, Mark Cook, who has a trade where he will risk three times as much as he stands to gain. However, on that particular trade he will win about 85% of the time. Investors thus need to have some kind of combination of either being right very often or, on average, to make much more than they are risking to consistently make money. To complement a methodology with an edge, investors must also apply a consistent discipline, says Schwager. Of course, this sounds much easier than it is, because of the emotions involved in trading. Schwager believes psychology is a key constituent of successful investing. Anyone that has ever traded knows that nearly everyone makes irrational decisions based on emotions such as fear, greed or hope at some point or other. It is at these times that big losses occur and investors get wiped out. In all three books, Schwager interviewed a number of trading coaches, one of whom is Dr. Ari Kiev, author and expert in trading psychology. I dont need any more reason to include Dr. Kiev in my latest book beyond the fact that Steve Cohen, probably one of the greatest traders that has ever lived, thinks enough of him to have him on his staff. Thats all the endorsement I need, he declares. For outstanding investors, says Schwager, failure is not an option and losses are part of the game. Many of the power traders have indeed lost all of what they had and actually ended up in debt before becoming truly successful at trading. Persistence and a sheer unbreakable conviction that they will succeed is a common thread among not just great traders but anyone who excels. And just one other thing you have to like doing it. You have got to be sure that trading is what you want to do, he warns. Most highly successful traders are workaholics working 12-14 hour days, Saturday and Sundays. The message for aspiring market wizards who have studied Schwagers thoughts is clear: there is no mystique, no magic formula, no hidden secrets. Successful traders eat, sleep and dream about trading. They absolutely thrive on it and enjoy it. Reproduced by permission of Dow Jones Newswires www.dowjonesnews.com COPY DEADLINE FOR THE NEXT ISSUE SEPTEMBER 2001 PUBLICATION OF THE NEXT ISSUE OCTOBER 2001 July 2001 ISSUE No. 41 MARKET TECHNICIAN FOR YOUR DIARY IN THIS ISSUE Wednesday 12th September Monthly Meeting Wednesday 10th October AGM Wednesday 14th November Monthly Meeting N.B. The monthly meetings will take place at the Institute of Marine Engineers, 80 Coleman Street, London EC2 at 6.00 p.m. Exam Results 2 D. Watts Software review 3 A. Goodacre Evidence on the performance of the CRISMA trading system in the US and UK equity markets 4 R. Klink Is the close price really the closing price? 8 Financial Overconfidence In markets 9 Markets Group D. Watts Bytes and pieces 13 M. Smyrk Book review 13 B. Millard Predicting future price movement 14 WHO TO CONTACT ON YOUR COMMITTEE CHAIRMAN Adam Sorab, Deutsche Bank Asset Management, 1 Appold Street, London EC2A 2UU TREASURER Vic Woodhouse. Tel: 020-8810 4500 PROGRAMME ORGANISATION Mark Tennyson dEyncourt. Tel: 020-8995 5998 (eves) LIBRARY AND LIAISON Michael Feeny. Tel: 020-7786 1322 The Barbican Library contains our collection. Michael buys new books for it where appropriate, any suggestions for new books should be made to him. EDUCATION John Cameron. Tel: 01981-510210 Clive Hale. Tel: 01628-471911 George MacLean. Tel: 020-7312 7000 Michael Smyrk. Tel: 01428 643310 EXTERNAL RELATIONS Axel Rudolph. Tel: 020-7842 9494 IFTA Anne Whitby. Tel: 020-7636 6533 MARKETING Simon Warren. Tel: 020-7656 2212 Kevan Conlon. Tel: 020-7329 6333 Tom Nagle. Tel: 020-7337 3787 MEMBERSHIP Simon Warren. Tel: 020-7656 2212 Gerry Celaya. Tel: 020-7730 5316 Barry Tarr. Tel: 020-7522 3626 REGIONAL CHAPTERS Robert Newgrosh. Tel: 0161-428 1069 Murray Gunn. Tel: 0131-245 7885 SECRETARY Mark Tennyson dEyncourt. Tel: 020-8995 5998 (eves) STA JOURNAL Editor, Deborah Owen, 108 Barnsbury Road, London N1 0ES Please keep the articles coming in the success of the Journal depends on its authors, and we would like to thank all those who have supported us with their high standard of work. The aim is to make the Journal a valuable showcase for members research as well as to inform and entertain readers. The Society is not responsible for any material published in The Market Technician and publication of any material or expression of opinions does not necessarily imply that the Society agrees with them. The Society is not authorised to conduct investment business and does not provide investment advice or recommendations. Articles are published without responsibility on the part of the Society, the editor or authors for loss occasioned by any person acting or refraining from action as a result of any view expressed therein. STA DIPLOMA APRIL 2001 PASS LIST L. Anastassiou F. Lien M. Chesterton V. Mahajan E. Chu M. Randall D. Cunnington J. Richie P. Day M. Schmidt B. Jamieson M. Shannon P. Kedia M. Sheridan R. Khandelwal M. Wahl R. Klingenschmidt P. Wong M. Knowles DITA II CANDIDATES T. Kaschel D. Lieber M. Meyer M. Unrath A. A. Waked A. Will 2 MARKET TECHNICIAN Issue 41 July 2001 Networking Exam Results ANY QUERIES For any queries about joining the Society, attending one of the STA courses on technical analysis or taking the diploma examination, please contact: STA Administration Services (Katie Abberton) P.O. Box 2, Skipton BD23 6YH. Tel: 07000 710207 Fax: 07000 710208 www.sta-uk.org For information about advertising in the journal, please contact: Deborah Owen 108 Barnsbury Road, London N1 OES. Tel/Fax: 020-7278 7220 Software available for sale as a result of a prize draw win. Software is unused. Nirvanas Omnitrader end of day stocks edition: high end automation of technical analysis with 120 built-in trading systems. Hot Trader Gann Swing Charting Program. 200 per package/ 350 for both inc. P&P (mainland UK). A 20 reduction per package for people who attend the monthly meetings, and who therefore dont require P&P. Contact: Mr Rahul Mahajan MSc MBA MSTA E-mail: r.mahajan@btinternet.com Issue 41 July 2001 MARKET TECHNICIAN 3 TradingExpert 6 is the latest offering from the AIQ people. AIQ (which stands for Artificial Intelligence Quotient) refers to the Expert System that is at the heart of this suite of programs, but it is also much more, with state of the art market charting and analysis being incorporated. The program modules integrate well together, forming a coherent complementary suite of programs that few others can match. Installation of the suite of programs is straightforward, both with an instructional video provided to ease this process and the programs initial operation. After installation, the programs are found via a tool bar on the desktop, which allows you to access the specific module that you require. What is impressive is the range and depth of the applications on offer. These comprise the charting package with the associated technical indicators and expert system. There is a data manager and communication interface. An excellent addition is Matchmaker a program to correlate groups of stocks and sort them into groups or sectors. Also significant is the real time quote monitor, a system and scanning writing programme called the Expert-Design Studio, and a Portfolio Manager. In short, you have a complete suite of programs to suit all your analysis or trading needs. At the heart of this package is the technical charting package with an array of indicators and the Expert System. The security chart is displayed with a list of indicators to the right. With one click any of these can be added to the chart. The indicator window also incorporates the Expert Rating together with the indicators that form the basis of the system. The Expert System gives a consensus rating of the technical indicators, providing a prediction of the likely future price movement of the security, with a rating from 0 to 100. Further information is provided by clicking upon the ER icon and up pops a box showing the reasoning for the expert rating shown. Does the Expert system perform? Overall I found the expert rating useful with some correlation to future price movement, but this is subjective judgement. In the Charting module, I particularly liked the one click insertion of technical indicators and moving averages into a chart, also the ability to change the defaults by clicking on the indicator displayed. Bar, line and candlestick charts can quickly be displayed via the top toolbar, but point and figure charts were disappointing, being too small to be analysed in the indicator window. Another column of icons provides the drawing and retracement tools, all being very easy to place and change on the chart. The suite also provides a Data Manager, with an extensive collection of pre-defined databases for the SP500, Nasdaq, FTSE100 stocks etc. The US/UK stocks with their tickers and exchanges are defined. All data is kept in the proprietary AIQ format, at which point I wished the package could read the standard Metastock/Computrac data format, as yet another database had to be set-up and maintained. However the Data Manager and communications program makes this easy. The communications program allows you to download data from the Mytrack servers, which I found fast and easy to use. The communication package loads data once you are connected to the Internet, making an Internet connection essential. The Mytrack servers also provide real-time data via the Quotes and Alerts modules, so once online both historical and real-time data are available. One excellent feature is that once a download is completed, the AIQ package also computes a set of reports for the most common technical criteria reports, Relative Strength, new highs or lows, breakouts, high volume days, and P&F breakouts. These are instantly available, making processing thousands of stocks a breeze, and can be immediately printed or stored for later analysis. The reports are available via the reports module, their breadth being the most extensive I have ever seen. Daily, weekly, sector, group and technical stock reports are all available, allowing a top down approach to stock selection to be developed, via a market, then sector, then stock. Furthermore, by clicking upon any security in the report module, up comes the chart for analysis. It is this integration that AIQ provides that allows real productivity gains to be made when analysing thousands of stocks and, importantly, not missing key technical events. The AIQ package scores again with its Expert Design Studio; here you can produce your own reports or back test a system. The program allows you to produce your own system, then scan for when those system criteria are met. Also unique is the ability to reference and hence back-test the Expert System in your analysis. An extensive list of pre-programmed studies are available should you not wish to program your own. Clearly AIQ have not left anything out of this suite of packages, and even more programs are included, such as the account manager and the fundamental database module. So having found your stock you want to monitor it real-time? With one click the real-time quote monitor, gives you a table of quotes with both intra-day and historical charts. I did find one annoyance, in that this did not work for a historical chart of a UK stock, the designation for London having to be changed each time, from -L to @L, whereas with one click US stock charts were instantly available. For real-time UK charts this was not a problem, with charts down to one minute being available in the charting module. How did the feed perform? Using a standard 56k modem, I found that the real-time data feed performed well with timely data, but a good Internet service provider is critical. Matchmaker is a real gem in the program suite. Here, via correlation analysis, it is possible to generate groups and sectors and look for leaders or laggards within a particular correlated group. AIQ uses a correlation approach to defining sectors and the performance of a stock within a sector. Then a variety of groups can be produced; for example, following the Wychoff approach, an index surrogate is possible. The Index surrogate generated could perhaps be of the top ten FTSE100 stocks by capitalisation as a leading indicator? Or you can generate a surrogate using spreads between correlated stocks, buying or selling the spread once it is overbought or oversold. Overall I liked this package for its ease of use and integration with the Internet. It offers a number of often excluded essential facilities for the equity analyst or stock trader. The extensive reports are a real boon for the analyst handling large numbers of stocks. The reports and data handling allow fast processing, while the expert system is a bonus for the less experienced. Further details can be found via the AIQ website at www.aiq.com. Software review AIQ TradingExpert 6 Review By David Watts FTSE100 INDEX 4 MARKET TECHNICIAN Issue 41 July 2001 The work reported here was undertaken jointly (but at different times) with three researchers while they were students on the MSc in Investment Analysis programme at Stirling University. I would like to acknowledge the considerable contributions made by Jacqueline Bosher, Andrew Dove and Tessa Kohn-Speyer. One of the problems with technical analysis is that it is difficult to apply academic rigour to what is essentially an art form. As a result, academics have tended to focus on the more objective processes in technical analysis, and typically, relatively simple trading rules such as filter rules, relative strength and moving averages have been tested. In many ways, this is somewhat unfair to technical analysts since they often implore the use of several indicators to confirm trading decisions. One unusual feature of the CRISMA (Cumulative volume, RelatIve Strength, Moving Average) trading system is that it is based on a multi-component trading rule that is objective and can therefore be subjected to rigorous analysis and testing. Another is that details of the system were first published more than ten years ago (Pruitt and White, 1988, 1989 and Pruitt, Tse and White, 1992). The system was shown to work very successfully in the US equity market. It was tested out-of-sample, adjusted for risk, and with due allowance for transaction costs and yet still provided statistically and economically significant excess returns. Based on data for the period 1976 to 1985 the system predicted annualised excess returns of between 6.1% and 35.6% (depending on the level of transaction costs) and similar excess returns continued for the 1986 to 1990 period. The CRISMA system was also applied to identify the purchase of call options on the underlying shares (Pruitt and White, 1989). Even in the presence of maximum retail costs, the average round-trip profit was 12.1% with an average holding period for each option contract of approximately 25 days. In an efficient market, technical trading systems such as CRISMA should not be able to produce abnormal returns or beat the market, once adjusted for risk and transaction costs. That CRISMA did produce abnormal returns, not only during the initial test period, but using out-of-sample data after the initial publication of CRISMAs success, led Pruitt and White to conclude that staggering market inefficiencies existed. Whats more, the system is used by one of its originators to run real money in a professional portfolio. The systems impressive credentials attracted our attention and so, some time ago, we set out to investigate whether CRISMA continues to work in the US and whether it works in the UK market. The rest of the paper is organised as follows. First, details of the CRISMA system are described, followed by the data and methods we used in back-testing the system. Results from applying the system in the US market are presented and compared with Pruitt et als results, as well as an analysis of the contributions of the system components. A brief summary of the results for the UK market is provided, including application of the system to option purchases, before some final concluding remarks on the usefulness of the system. The CRISMA trading system The CRISMA trading rule operates a filter system for selecting equity targets. The first three filters attempt to triple confirm upward momentum, and use the 50 and 200-day price moving averages, relative strength and volume as indicators. These filters must be satisfied on the same day. The moving average filter is satisfied when a stocks 50-day price moving average intersects the 200-day price moving average from below, when the slope of the latter is zero or positive. This can only occur when the price of the stock is increasing relative to previous periods. This golden cross is used to indicate the start of an upward trend, which is expected to continue upwards. The relative strength filter is satisfied when, over the previous four weeks, relative strength has increased or remained unchanged. This filter ensures that the relative performance of the stock is at least equal to the relative performance of the market as a whole in the previous four weeks. Cumulative volume (also called on-balance volume by some users, for example in Bloombergs) is defined as the cumulative sum of the volume on days on which the price increases, minus the volume on days on which the price falls. The cumulative volume filter is satisfied if its level has increased over the past four weeks, suggesting that there is increasing demand for the stock. Once the first three filters are satisfied, the stock is purchased when it reaches 110% of the intersection point of the 50 and 200 day price moving averages. This final penetration filter attempts to reduce the likelihood of whipsaws, the inadvertent issue of false signals; it does not have to be satisfied on the same date as the first three filters. Stocks are sold when either the price rises to 120% of the moving average crossover point or when it falls below the 200 day price moving average. There are 3 exceptions to the trading system mechanics as detailed below: 1. If the price of the stock does not reach the 110% level of the intersection between the 50 and 200 day moving averages within 5 weeks of the first three filters being satisfied, it is not purchased, 2. If the stock price is already above the 110% filter when the first 3 filters are satisfied, it is not purchased until it has fallen below this level, and 3. If the stock price is already above the 110% level, when the first three filters are satisfied and reaches 120% of the intersection point of the 50 and 200 day price moving averages without falling below the 110% level, it is not purchased. METHODS Pruitt and White (1988) used the Daily Graphs series published weekly by William ONeil and Company, Inc. to aid in the identification of trades. Figure 1 re-creates the graphical selection process for the CRISMA trading rule, using data from the current study for American Airlines. Identifying trades in this manner involves elements of subjectivity and is time consuming, so a more efficient and accurate approach, using a spreadsheet macro, was adopted. The macro imports pre-tidied price and volume data for each company into a template containing the formulae for the moving averages, relative strength and cumulative volume measures. It then searches for days on which CRISMAs three initial filters are satisfied and calculates the 110% and 120% price levels. The dates of the purchase and subsequent sale of the stock are then identified and, providing no exceptions to the rule have been breached, the dates and prices are subsequently copied to a results table. A random sample of 333 companies from the S&P 500 was selected in June 1997 with the requirement that a minimum of 4 years of volume data could be obtained between January 1987 and June 1996. 11 companies were excluded due to insufficient available data giving a final sample of 322. This sample selection process implies a degree of survivorship bias (Kan and Kirikos, 1995) which is discussed later. The prices obtained from Datastream were daily mid-market closing prices, pre-adjusted for bonus and rights issues. The benchmark index used for the return calculations was the S&P 500 Index. Evidence on the performance of the CRISMA trading system in the US and UK equity markets By Alan Goodacre Issue 41 July 2001 MARKET TECHNICIAN 5 To assess how successful CRISMA might have been, several questions were considered. On average: Are the trades profitable? Do the trades beat a market benchmark? Do the trades beat a risk-adjusted benchmark? Is the proportion of successful trades greater than the 50% expected by chance? How are the answers affected by transaction costs? All of the above questions have to be considered in the context that, effectively, we have taken a sample from all of the possible trades (over all companies and all time). Sampling theory suggests that positive answers to the above performance questions can be achieved by chance, (i.e. solely as a result of the sampling process), so the likelihood of the observations being chance events needs to be assessed. This is achieved by carrying out hypothesis tests on the results to identify those that are unlikely to be chance events (described as being statistically significant and noted by an asterisk in the Tables). This provides some confidence that any success might be due to the CRISMA system rather than luck. RESULTS Results in the US market for the period 1 January 1988 through 30 June 1996 From the 322 companies included in the sample, over the 8 1 /2 year test period the CRISMA trading system identified 328 trades in 208 companies whose stocks were assumed held for a total of 14,424 security-days (approximately 2% of the total security-days). This gives an average annual total of 1,697 security-days which implies a portfolio of approximately 6.8 stocks, on average. The mean and median holding periods were 44 and 31 days respectively, with a maximum holding period of 374 days and a minimum of 1 day. Table 1 presents a summary of the raw CRISMA generated profits. Assuming zero transaction costs, the CRISMA trading system would have been profitable with a mean and median profit per trade of 3.1% and 8.9% respectively. After allowing for round-trip transaction costs of up to 2% the system was still profitable (all returns are statistically significantly at 5% or less). The final column of the table shows arithmetically annualised mean returns. These are calculated by working out the average (mean) return per day across all of the 328 trades and multiplying by 250 (the approximate number of potential trading days in a year). These annualised returns ranged from 17.6% to 6.2% depending on the level of transaction costs assumed. Thus, the system would have been profitable over the period, even allowing for transaction costs of up to 2%. However, a more important question is whether the system adds value. Does it achieve better performance than a buy-and-hold policy? To test this, the return on the market (S&P500) was deducted from the return obtained for each day a stock was held to give a return in excess of the market benchmark. Again, this was averaged over all the days stocks were held and then annualised; the results are in column 3 of Table 2. A further question of importance is whether the system returns compensate for the risks taken. Traditional finance theory suggests that the only risk element that should be rewarded is systematic (or beta) risk, since all stock-specific risk can be eliminated by holding a well-diversified portfolio. Of particular importance here is that, in a rising market, high beta stocks should out-perform the market. Therefore, column 4 of Table 2 reports the results from using the so- called market model to allow for the effects of each individual stocks beta risk (consistent with Pruitt et als testing of the system). However, it is worth noting that this model is potentially biased against the trading system. A momentum-based system, such as CRISMA, seeks to identify relatively good performance by a security in the expectation that such performance will continue. The market model will have some recent good performance built into the expectations, for example, via a positive alpha (the mean value for alpha in the current study was 8.4%, on an annualised basis). The market-adjusted return does not suffer from this bias, but unfortunately neither does it account for the relative riskiness of the stocks held. The nature of the CRISMA system is that for each trade it seeks to capture a profit of, or limit the loss to, approximately 9% (i.e. 10/110%). This is quite clearly illustrated in the histogram of unadjusted returns in Figure 2. If, on balance, a higher proportion of profitable trades is identified then the system will be successful. Thus, a key measure of success is the proportion of trades that are profitable. The lower half of Table 2 reports the proportion of profitable trades for each of the benchmarks and transaction cost levels. Transaction Profit per trade (%) Annualised Costs Mean Median mean profit (%) 0.0% 3.1* 8.9 17.6* 0.5% 2.6* 8.4 14.8* 1.0% 2.1* 7.9 11.9* 2.0% 1.1* 6.9 6.2* Table 1: Unadjusted Profits for CRISMA trading system Jan 88 to June 96 Figure 2: Histogram of raw CRISMA returns Jan 88 to June 96 Figure 1: The CRISMA trading system applied to American Airlines 1. The 50-day moving average crosses the 200-day moving average from below at a price of $60.9 on 13 December 1991. The slope of the 200-day moving average over the previous four weeks is positive. The 110% and 120% price levels are $67.0 and $73.1 respectively. 2. The relative strength and cumulative volume graphs are positively sloped over the previous four weeks. 3. The 110% buy level is reached on 23 December and a buy at $68.9 is indicated. 4. The 120% sell level of $73.1 is exceeded on 7 January 1992, when a sell price of $75.0 is indicated. 5. Thus this trade achieves a raw profit (unadjusted for dividends, transaction costs and risk) of 8.9% over a period of 9 trading days when the market (S&P 500) rose by 5.2%. 6 MARKET TECHNICIAN Issue 41 July 2001 As can be seen from Table 2, adjusting for market movements and risk significantly decreases the returns predicted by CRISMA. Assuming zero transaction costs, the annualised mean excess return is just 0.1% when compare with the market benchmark. Taking beta risk into account further reduces the return to -6.8%. For non-zero levels of transaction costs, the excess returns are negative for both benchmarks, statistically significant for transaction costs of 2% (market) and 0.5% and above (risk-adjusted). An alternative way of assessing the success of CRISMA is to test the proportion of recommended trades that were profitable (lower half, Table 2). Generally, more than the 0.5 expected by chance were identified, statistically greater than 0.5 for unadjusted returns and also at low transaction cost levels for market-benchmark returns. Thus, there is some evidence of success at predicting profitable trades but no evidence of the system generating excess returns. It is not easy to identify the likely effect of the element of survivorship bias implicit in the sample selection procedure. Companies that have been forced out of the S&P 500 through failure and those that have been acquired, or merged, will be under-represented in the sample; i.e. good performance companies are likely to be over-represented, suggesting a possible positive bias in reported returns. Kan and Kirikos (1995) report a significant overstatement of the performance of a momentum based trading strategy resulting from a similar selection process in the Toronto stock exchange. If there is a positive bias in the current study, this would imply that the CRISMA system performs even less well than reported here. In summary, back-testing the CRISMA system in the US market over the 1988-96 period produces disappointing results. While the system identifies profitable trades, even in the absence of transaction costs it does not add value by providing significant returns in excess of a buy-and-hold strategy. Once transaction costs are incorporated, the system actually under-performs a buy-and-hold strategy. Comparison with Pruitt et al studies While much of the current study period is subsequent to the Pruitt et al test periods, there is an overlapping period of 2 1 /2 years. This allows a direct comparison between the graphical approach of Pruitt et al and the spreadsheet-automated approach adopted here. Results from this comparison and a summary of the original Pruitt et al studies are presented in Table 3 (nr means that the measures were not reported in the Pruitt papers). The contrast between the positive excess returns in both Pruitt studies and the negative returns here is very striking; similarly, there is a marked difference between the proportions of profitable trades. Particularly disconcerting are the differences in the results over the identical overlap period June 88 through December 1990. Our method finds more trades, resulting in more days held but identifies less than 0.5 profitable trades. This leads to significant negative excess returns of 16.5% against the market benchmark and 18.3% on a risk-adjusted basis. By contrast, Pruitt et al found 0.76 of profitable trades (at zero transaction costs) giving a significant positive annualised excess return of 29.1% on a risk-adjusted basis. Possible explanations for this different performance were investigated relating to three areas: sample selection, method of application and returns measurement. For this exercise zero transaction costs are assumed throughout. Pruitt et al. included all stocks jointly included in the CRSP daily data tapes and the Daily Graphs series in contrast to our random selection of approximately two-thirds of the S&P 500. One possible explanation might be that these two samples have different distributions of company size. We found evidence in the current study that CRISMA performed better on the largest companies in the sample, so if the Pruitt sample was biased towards larger companies this might have contributed to the observed differences. Application of the CRISMA system using the objective spreadsheet approach rules out any subjective input to trade identification. For example, one issue concerns the potential false identification of a sell signal when a stock price falls when it goes ex-dividend; such false signals could be avoided in the graph-based approach by subjective intervention. This issue might have a negative impact on performance in our computer-based approach through the induced sale of profitable trades before the profit was achieved or through the sale of loss-making trades at a lower price. The effect of this was tested by re-running the full sample through the system with trades identified on the basis of a smoothed returns price index for each stock. This index adds back the dividend amount to the price series thus avoiding a step-down effect on the ex-dividend day and is available in the Datastream database. The results from this approach indicate that excess returns were slightly improved to +1.0% (market) -4.2% (risk-adjusted). Thus, a subjective input to trade identification of this nature might have improved the Pruitt results slightly. Pruitt et al. did not define the specific measure of relative strength that they used, so the sensitivity of the current results to an alternative definition of relative strength was tested. A larger number of trades (394) was identified but the effect on returns was quite small and negative (e.g. excess returns were -0.9% against the market benchmark). Perseverance is less of a difficulty in our objective computer-based identification of trades. The mean and maximum stock holding Return in excess of Transaction Unadjusted Market Risk-adjusted costs return benchmark benchmark Annualised mean excess return (%) 0.0% 17.6* 0.1 -6.8 0.5% 14.8* -2.8 -9.6* 1.0% 11.9* -5.6 -12.5* 2.0% 6.2* -11.3* -18.2* Proportion of trades profitable 0.0% 0.61* 0.56* 0.52 0.5% 0.61* 0.55* 0.51 1.0% 0.61* 0.53 0.50 2.0% 0.60* 0.51 0.47 Table 2: Tests of CRISMA trading system performance Jan 88 to June 96 Return in excess of Study Time No of Days Trans Market Risk-adj period trades held costs (%) benchmark benchmark Annualised mean excess return (%) Current 01/88-06/96 328 14424 0 0.1 -6.8 2 -11.3* -18.2* Current 06/88-12/90 81 3084 0 -16.5* -18.3* 2 -29.6* -31.4* Pruitt (1992) 06/88-12/90 68 1718 0 nr 29.1* 2 nr 9.3 Pruitt (1992) 01/86-12/90 148 3477 0 nr 26.5* 2 nr 5.2 Pruitt (1988) 01/76-12/85 204 4970 0 28.5* 27.0 2 13.1* 11.6 Proportion of trades profitable Current 01/88-06/96 328 14424 0 0.56* 0.52 2 0.51 0.47 Current 06/88-12/90 81 3084 0 0.44* 0.43* 2 0.41* 0.40* Pruitt (1992) 06/88-12/90 68 1718 0 nr 0.76* 2 nr 0.66* Pruitt (1992) 01/86-12/90 148 3477 0 nr 0.75* 2 nr 0.64* Pruitt (1988) 01/76-12/85 204 4970 0 0.73* 0.71* 2 0.66* 0.61* Table 3: Comparison with original Pruitt et al results Issue 41 July 2001 MARKET TECHNICIAN 7 periods in our study were 44 and 374 days respectively compared with 24 and 102 in the later Pruitt study. We had 17 trades with long holding periods (i.e. longer than their maximum period). The total number of security-days for these trades was 3,254, which represented 23% of the overall total of security-days, so it is possible that these trades could significantly influence the overall picture. Whilst the unadjusted returns for the long trades showed that they were profitable, on average they achieved lower annualised returns (10.8%) than the remaining normal holding period trades (19.6%). However, the removal of long trades would leave small (but insignificant) positive excess returns of +0.7% (market) or slightly less negative returns of -5.9% (risk-adjusted). Overall, the fundamental picture remains unchanged. CRISMA fails to identify trades that give significant positive returns when adjusted for market movements and risk, even in the absence of transaction costs. Importance of the individual components of the CRISMA trading rule A weakness of the original Pruitt study is that it did not fully disclose how the CRISMA system was developed; for example, the rationale for the specific choices of moving average lengths, filter sizes, relative strength periods and so on. The papers compensating strengths are that the returns were adjusted for risk, and testing (certainly in the second paper) was fully out-of-sample and even subsequent to the market having knowledge of the system. Ultimately, the continued success of the system itself can be argued to be sufficient justification. However, the current study provides an opportunity to investigate the relative importance of the different components of the system. First, we identify all potential trades that meet the basic moving average golden cross component. Then, we observe the proportion of such trades that are eliminated as a result of the three exceptions to the systems mechanics (e.g. the price does not rise to the purchase point within five weeks of the moving average crossover). Last, we identify the proportion of the remaining trades that are ruled out by the relative strength (RS) and cumulative volume (CVOL) components and measure the returns which these excluded trades would have generated. A number of interesting observations can be made. First, the number of potential trades eliminated as a result of the exceptions to the systems mechanics is quite large, representing over 50% of the potential trades, indicating that these exceptions are not quite as rare as suggested in Pruitt et al. (end note 4, 1988). Second, the proportion of trades that are excluded because of the failure of the trades to meet the relative strength and cumulative volume components of the system is fairly small. The RS rule excludes approximately 15%, and the CVOL rule approximately 26% of potential trades that meet the basic golden cross condition; jointly, these two rules exclude approximately 31% of potential trades. Third, and more important, the profitability of these excluded trades is positive and actually superior to the trades that are included! Overall, the effect of excluding trades because they failed to meet either the RS or CVOL conditions is to reduce the annualised excess return by 1.2% against the market benchmark. Thus, the benefits of the multi-component nature of the CRISMA system over a simple moving average rule are not proven. The proportion of trades that fail the extra two conditions is relatively small, and the exclusion of the trades actually reduces the profitability of the system. Results in the UK market for the period 1 January 1988 through 30 June 1996 Applying the CRISMA system to a sample of 254 FTSE 350 index constituents produced very similar results to those presented above for the US market. It identified 176 trades in 113 companies whose stocks were assumed held for a total of 7,067 security-days. The mean and median holding periods were 40 and 32 days respectively, with a maximum holding period of 152 days and a minimum of 2 days. Assuming zero transaction costs the CRISMA trading system would have been significantly profitable with a mean profit per trade of 3.1% and median profit per trade of 7.9%. After allowing for round- trip transaction costs of up to 2% the system was still profitable, though not significantly so at the 2% level of costs. Arithmetically annualised returns (based on a 250 business day year) ranged from 19.3% to 6.9% depending on the level of transaction costs assumed; for comparison, the mean annual return on the FTSE All Share Index over the same time period was 14.0%. Assuming zero transaction costs, annualised excess returns of 2.2% (market) and -2.9% (risk-adjusted) were observed. For non-zero levels of transaction costs the excess returns were negative for both market and risk-adjusted benchmarks (statistically significant for 2% and 1% transaction costs, respectively). For unadjusted returns, the proportion of profitable trades was 0.60 for all four transaction cost levels. For market benchmark returns, the proportions of profitable trades were 0.57, 0.56, 0.54 and 0.51 for transaction costs of 0%, 0.5%, 1% and 2% respectively, statistically significantly greater than 0.50 for the two lower cost levels. For risk adjusted returns only 0.51 of trades were profitable at zero transaction costs, but less than 0.5 for the other three levels of costs. Thus, for trades based on UK equity securities, the results are similarly disappointing to those observed in the US market. CRISMA applied to UK exchange traded options Pruitt and White (1989) argued that few market traders actually purchase equity securities when applying strategies similar to the CRISMA system. Rather, they seek the higher leverage and profit potential represented by exchange-traded options. Consideration of the performance of such an option strategy was assessed for the UK market, but was limited by the relatively small number of options traded at LIFFE. Only 64 trades from the original 176 could be tested because either there were no options listed on the particular securities or there were no in-the-money call options available at the time of the recommendation. The mean holding period for options was 41 days with maximum and minimum holding periods of 116 and 5 days. Assuming zero transaction costs, the mean return per round-trip option trade was 27.7% with 61% of all trades profitable (statistically greater than 0.50). Even in the presence of maximum retail costs (estimated at approximately 50 per contract, or 11.5% of the mean contract price), the CRISMA system generated a mean return of 10.2% per option trade, although with only 55% of trades profitable (not statistically significantly greater than 0.50). Nevertheless, the CRISMA system does appear to be able to produce high returns when used to select options, since a mean return of 10.2% in the presence of maximum retail transaction costs for an average holding period of 41 days could not be considered undesirable. The returns to CRISMA-generated option trades are very similar to those reported by Pruitt and White (1989, 1992) but the US mean holding period per trade of 25 days is shorter. The proportion of profitable trades for the UK is lower than the 68% (64% for maximum retail costs) reported by Pruitt and White (1992) for the 1986-1990 period. One important feature of the option-based strategy is that, as with equities, success depends on achieving a larger proportion of profitable than unprofitable trades. However, the balance is between positive and negative returns of about 75% rather than the 9% for equities. In other words it is a much riskier strategy, so much higher returns are needed to justify the additional risk. CONCLUSIONS The present study back-tested a spreadsheet-based version of the CRISMA trading system in both the US and the UK markets over the period 1988 to 1996. In general, the results were somewhat disappointing. Whilst trades were profitable, on average, once these were adjusted for market movements and risk they ceased to be so and transaction costs merely served to worsen the negative excess returns. The leverage attribute of exchange-traded options tested for the UK market improved the picture and trades were profitable, on average, after due allowance for maximum retail transaction costs. However, this needs to be balanced against the increased risk of adopting the option strategy. continues on page 8 8 MARKET TECHNICIAN Issue 41 July 2001 Nothing is more critical to technical analysts in their trading than accurate price information. Inaccurate historical data can result in invalid back testing results, leading to the development and implementation of faulty trading systems. Building a solid and profitable trading system is hard enough for most of us without this handicap! One of the underlying themes in the determination of data quality is whether the quality issues are ones of system error or by way of definition. As we will see, there are many and varied ways of calculating a close price. All methods are valid and their accuracy may depend on the end users perspective and intended use of the data. Whilst I will be primarily focusing on technical analysis in this article, the effects of data quality extend deep into the finance industry. For example, Portfolio Valuations, Indexing and Benchmarking rely heavily on accurate and consistent close prices. Close prices provide a great starting point in terms of reviewing data quality issues and consistency questions. As a provider of financial data globally, we are constantly challenged with the different variations on close prices defined by global stock exchanges. These include terms such as end-of-day, last traded, indicative price, derived price and settlement price. In an attempt to better define and document the processes used by global exchanges to derive their close processes, the Financial Information Services Division (FISD) of the Software and Information Industry Association has weighed in with their Closing Price Project. For more details on the project, visit http://www.fisd.net/price The purpose of this project is to gain transparency and full disclosure of how close prices are created and calculated by equity exchanges and processed and displayed by end-of-day price vendors globally. The focus of this project is not one of dictating the method of calculation by each exchange and vendor, rather, to define and document their process so that users of the data will be better able to understand the origins of the data they are dealing with and any associated derivations, modifications or assumptions applied to the same. The team is currently in the process of surveying each exchange and data vendor to determine this information. In terms of the exchanges we deal with globally today, many have published end-of-day price files available for the data vendor redistribution. These files include, at minimum, Open, High, Low, Close and Volume values. Some examples of these exchanges include Australia (www.asx.com.au), New York Stock Exchange (www.nysedata.com) and Singapore (www.ses.com.sg). By publishing their own, defined files daily, data consistency levels across data vendors is high. Other exchanges, including the London Stock Exchange (www.londonstockexchange.com) supply data vendors with a real time data feed. The data vendors then on-process this real time feed to strip an end-of-day file to supply to end users. This can lead to differing end-of-day prices being distributed by each data vendor depending on the methodology used to strip the data from the feed. For example, how does a vendor define the Open price? The first trade, or by the definition some exchanges use, an average of the first few minutes trading. The LSE is currently estimating the release of their published end-of-day data service by year end. Also, how do vendors allow for spurious or invalid trades? Some automatically reject bad or doubtful ticks, based on cleansing algorithms. In addition, some real time data providers have a team of data analysts continuously monitoring the data to remove bad ticks that would throw out end users charting applications for the rest of the day. It is the thoroughness and speed of this cleansing process that really separates data vendors in terms of data quality. Another key determinant of data quality to look out for is the application of the appropriate dilution factors (splits and adjustments) to raw data pricing. The spikes sometimes present in non back- adjusted data can play havoc with technical analysis indicators. For example, moving averages and rate of change indicators can and will provide spurious calculation results that will invalidate back-testing and market scanning exercises. In summary, the good news for Investors and Traders globally is that efforts by organizations such as FISD are great steps forward in driving accurate and consistent pricing globally. By raising the issue with the global exchanges and working with them to better define pricing methodologies, transparency will occur. Rick Klink is the CEO of Paritech continued from page 7 Further investigation of the multi-component nature of CRISMA revealed that it is, in large part, a moving average trading rule. The impact of the cumulative volume and relative strength components on observed returns is negative; CRISMA would have produced slightly improved returns had these components of the system been ignored! Over a common time period, application of the objective spreadsheet version of the CRISMA provided disappointing performance compared with that obtained using the graphical approach by Pruitt et al. The sensitivity of the results to various aspects of the model was investigated but no major individual impacts were identified. However, it is possible that a set of minor differences in the application of the model, taken together, could account for the observed differential performance. Overall, the results suggest that, at best, the success of the trading system is extremely sensitive to the exact method of application. Alan Goodacre is Senior Lecturer in Accounting and Finance at the University of Stirling and can be contacted by e-mail: Alan.Goodacre@stir.ac.uk Further details of the research can be found in the two academic papers in Applied Financial Economics listed in the references. Additional information about Applied Financial Economics is available on the journal publishers web site: http://www.tandf.co.uk REFERENCES Goodacre, A., Bosher, J., and Dove, A. (1999), Testing the CRISMA Trading System: Evidence from the UK Market, Applied Financial Economics, 9, 455-468. Goodacre, A. and Kohn-Speyer, T.K. (2001), CRISMA revisited, Applied Financial Economics, 11, 221-230. Kan, R. and Kirikos, G. (1995), Biases in Evaluating Trading Strategies, Working Paper, University of Toronto. Pruitt, S.W. and White, R.E. (1988), The CRISMA Trading System: Who Says Technical Analysis Cant Beat the Market?, Journal of Portfolio Management, 14, Spring, 55-58. Pruitt, S.W. and White, R.E. (1989), Exchange-Traded Options and CRISMA Trading, Journal of Portfolio Management, 15, Summer, 55-56. Pruitt, S.W., Tse, K.S.M. and White, R.E. (1992), The CRISMA Trading System: The Next Five Years, Journal of Portfolio Management, 18, Spring, 22-25. Is the close price really the closing price? By Rick Klink Issue 41 July 2001 MARKET TECHNICIAN 9 Last year stock markets reached unprecedented levels, with extremely high Price-Earnings multiples and high price volatilities. In the technology sector share prices of firms that had never posted any earnings reached extraordinarily high levels. The rise of the New Economy and the emergence of the biotechnology sector raised fundamental questions regarding the proper valuation of companies operating in these sectors. The levels of volatility last year threw into doubt conventional valuation methodologies. It remained to be seen whether market valuations were being driven by expectations of future earnings or by irrationality and overconfidence. In November, the Financial Markets Group of The London School of Economics, with the support of UBS Warburg, organised a conference on Market Rationality and the valuation of Technology Stocks to analyse key issues, including valuation of firms that exhibit no earnings and have limited history, aspects of trading overconfidence in competitive securities markets and market efficiency in the new environment. The first session of the Conference was chaired by Hyun Song Shin (FMG/CEP/LSE) and dealt with Market Efficiency in the New Environment. The conference opened with Sunil Sharma (International Monetary Fund) who presented the paper entitled Herd Behaviour in Financial Markets: A Review co-authored with Sushil Bhikchandani (Anderson Graduate School of Management, UCLA). Motivated by the growing literature in this area, Sharma looked at the causes for and empirical evidence of herding with particular emphasis on the Lakonishok-Shleifer-Vishny (LSV) measure and the so-called learning among analysts. He surveyed the literature corresponding to the three main causes of herding: (a) information- based, (b) reputation-based, and (c) compensation-based. Information-based herding is the belief that analysts are uncertain about the costs and benefits of their actions and so rely on the observable behaviour of other analysts, and not their own information set. The information inferred from actions of predecessors overwhelms the individuals private set of information. Reputation-based herding models focus on the idea that fund managers, unsure of their own abilities and those of their competitors and concerned about the loss of reputation that may result as their performances are compared to their peers, will end up imitating each other. Finally compensation-based herding relies on the idea that investment managers are compensated on the basis of their performance relative to a chosen benchmark. If investment managers are very risk averse, then they will have a tendency to imitate the benchmark. Sharma claimed that the empirical evidence of these issues is weak given that the literature does not examine or test particular models of herding. Most papers use purely statistical approaches, based on the idea that the analysis may be picking up common responses to publicly available information. The most commonly used measure of herding is the LSV measure, which defines herding as the average tendency of a group of money managers to buy (sell) particular stocks at the same time relative to what could be expected if they traded independently. Using this measure there is not much evidence of herding in developed markets but it tends to be greater in small and growth firms and is correlated with managers use of momentum strategies. The analysis for emerging markets, although it tends to be less econometrically robust, indicates that herding behaviour forms much more easily compared with stocks in larger and more developed markets. Increased herding is also associated with greater return volatility. Evidence on herding amongst investment analysts, collected by Graham (1999), indicates that the likelihood that an analyst follows a herding strategy is decreasing in his ability, increasing in the reputation of the analysts newsletter, the strength with which prior information is held and the correlation of informative signals. Other literature finds that conditional on performance, inexperienced analysts are more likely to herd than their older colleagues. Welch (1999) shows that herding tends to be in the direction of the prevailing consensus and the two most recent revisions. Such trading behaviour is stronger in market up-turns than down-turns. Sharma concluded that it is important to distinguish between endogenous interactions, contextual decisions and correlated effects because of the implications for public policy. There is a need for better characterizations of interactions in terms of preferences, expectations, constraints and equilibrium. Finally there is tremendous need for better and more robust data to facilitate credible inference. The discussant of this paper, Leonardo Felli (LSE/FMG), focused on the question of how powerful is career-concern herding. He indicated that comparing the behaviour of direct investors and delegated manager investment might shed more light on the identification of reputation-based herding. The second paper of the session was presented by Kent Daniel (Kellogg School of Management, Northwestern University). The paper, co-authored with Sheridan Titman (College of Business Administration, University of Texas), was entitled Market Efficiency in an Irrational World. Daniel explained the predictability of prices by examining the over- reaction effect in the long term based on the distinction between tangible and intangible information. Tangible information refers to accounting measures that are indicative of the firms growth obtained from its fundamentals; whereas intangible information is accountable for the part of a securitys past returns that cannot be linked directly to accounting numbers, but which may reflect expectations about the future cash flows. The predictability is broken down into three possible sources: (a) extrapolation bias (over-reaction to tangible information, as measured by the book-to-market ration); (b) over-reaction to intangible information about future growth, and (c) pure noise (price movements unrelated to future cash flows). While finding no evidence to support the first effect, Daniel found that over-reaction to intangible information is a significant factor in explaining predictability. This is related to the observation that people tend to be over confident about vague, non-specific information. The paper was discussed by George Bulkley (University of Exeter), who questioned the strength of the results. He claimed that, although the decomposition is interesting, there is difficulty in determining whether there is support for any particular model. He claimed that the results presented in the paper were consistent with any model which generates mispricing measured by the book-to- market ratio. However, he admitted that there is substantial econometric difficulty in testing these models given the continuously overlapping layers of information in financial data and the lack of discriminating tests. Countering Bulkleys comments, Daniel argued that the aim of the paper is to explain long run returns and he did not want to contaminate the results with effects from the post-earnings announcement drift. The second session, covering Overconfidence in the Stock Market, was chaired by Sudipto Bhattacharya (FMG/LSE) and started with the presentation of a paper entitled Learning to be Over Confident by Simon Gervais (Wharton School, University of Pennsylvania). Gervais developed a multi-period market model describing both the process by which traders learn about the ability and how a bias in this learning can create overconfident traders. The principal goal of this work was to demonstrate that a simple and a prevalent bias in evaluating a traders personal performance is sufficient to create markets in which investors are, on average, overconfident. Overconfidence in markets 10 MARKET TECHNICIAN Issue 41 July 2001 In the model, traders initially do not know their ability. They learn about their ability through experience. Traders who successfully forecast net period dividends, improperly update their beliefs; they overweight the possibility that their success was due to superior ability. In so doing they become overconfident. A traders level of overconfidence changes dynamically with his successes and failures. A trader is not overconfident when he begins to trade. Ex-ante, his expected overconfidence increases over his first trading periods and then declines. Thus, a trader is likely to have greater overconfidence in the early part of his career. After this stage of his career, he tends to develop a progressively more realistic assessment of his abilities as he ages. Overconfidence does not make traders wealthy, but the process of becoming wealthy can make traders overconfident. Gervais concentrated on the dynamics by which self-serving attribution bias engenders over- confidence in traders, and not on the joint distribution of trader ability and the risky securitys final payoff. In this model, overconfidence is determined endogenously and changes dynamically over a traders life. This enables predictions about when a trader is most likely to be overconfident (when he is inexperienced and successful) and how overconfidence will change during a traders life (it will, on average, rise early in traders career and then gradually fall). The model predicts that overconfident traders will increase their trading volume and thereby lower their expected profits. To the extent that trading is motivated by overconfidence, higher trading will correlate with lower profits. Volatility increases with a traders number of past successes. Both volume and volatility increase with the degree of a traders learning bias. The model has implications for changing market conditions. For example, most equity market participants have long positions and benefit from upward price movements. Therefore it is expected that aggregate overconfidence will be higher after market gains and lower after market losses. Since greater overconfidence leads to greater trading volume, this suggests that trading volume will be greater after market gains and lower after market losses. David de Meza (Exeter University/Interdisciplinary Institute of Management, LSE), the discussant for the paper, commented that he found the results useful and interesting. He pointed out that the paper makes a distinction between illusion of validity and unrealistic optimism (understood as forecast error). He also emphasized the mechanics of overconfidence by which insiders overestimate their forecasting ability in the light of accurate forecast. De Meza finished by agreeing that overconfidence leads to an increase of trading volume. However he pointed out that it is not clear why market confidence should fluctuate over time. He mentioned, in addition, that correlated signals might also mean that errors are correlated so there was a need for aggregated tests. The session concluded with the presentation of a paper by Avanidhar Subrahmanyam (UCLA), entitled Covariance Risk, Mispricing and the Cross Section of Security Returns. The focus of the presentation was a model in which asset prices reflect both covariance risk and misperceptions of firms prospects, and in which arbitrageurs trade against mispricing. Subrahmanyam examined how the cross-section of expected security returns is related to risk and investor mis-valuation. He provided a model of equilibrium asset pricing in which risk averse investors use their information incorrectly in selecting their portfolio. The model was applied to address the ability of risk measures versus mis-pricing measures to predict security returns; the design trade-offs among alternative proxies for mis-pricing; the relation of volume to subsequent volatility; and whether the mis-pricing equilibrium can withstand the activities of smart arbitrageurs. The presenter stated that may empirical studies show that the cross section of stock returns can be forecast using not only standard risk measures, such as beta, but also market value or fundamental/price ratios such as dividend yields or book-to-market ratios. The interpretation of these forecasting regressions is controversial, because these price-containing variables can be interpreted as proxies for either risk or mis-valuation. Subrahmanyam also pointed out that, so far, this debate has been pursued without an explicit theoretical model. This paper presents an analysis of how well, in this situation, beta and fundamental/price ratios predict the cross-section of security returns. The model implies that even when covariance risk is priced, fundamental-scaled price measures can be better forecasters of future returns than covariance risk measures such as beta. Intuitively, the reason that fundamental/price ratios have incremental power to predict returns is that a high fundamental/price ratio (e.g. high book- to-market ratios) can arise from high risk and/or overreaction to an adverse signal. If price is low due to a risk premium, on average it rises back to the unconditional unexpected terminal cash flow. If there is an overreaction to genuine adverse information, then the price will on average recover only in part. Since high book-to-market reflects both mispricing and risk, whereas beta reflects only risk, book-to-market can be a better predictor of returns. Subrahmanyam concluded with an examination of the profitability of trading by arbitrageurs and overconfident individuals. He further mentioned that it would be interesting to extend the approach to address the issues of market segmentation and closed-end fund discounts. Sudipto Bhattacharya, (FMG/LSE) the discussant for this paper, pointed out that the model presented in the paper introduces investor psychology to theory of capital asset pricing by integrating risk measures (CAPM) with overconfidence (mispricing). He asked whether the misvaluation effects identified in the model could persist in the long run and hypothesised that investors may become overconfident about a theory of how the stock market functions rather than about the realization of a signal. Bhattacharya further commented that it would also be interesting to investigate the correlation between signals and future returns as well as the implication of univariate equilibrium conditions. He was excited about the further research suggested by Subrahmanyam, especially regarding how different groups of investors fasten their overconfidence upon one analysis versus another, and how the social process of theory adoption influences security prices. The third session dealt with Analysts Forecasts and Pricing in the New Economy and was chaired by Bob Nobay (FMG/LSE). Peter David Wysocki (University of Michigan) presented a joint paper with Scott A Richardson (also University of Michigan Business School) and Siew Hong Teoh (Ohio State University) on The Walkdown to Beatable Analyst Forecasts: The Role of Equity Issuance and Insider Trading Incentives. Motivated by the observation that in the last couple of years, earnings announcements that exceed analysts forecasts and subsequent jumps in share prices have become more frequent, Wysocki asked whether this fact can be statistically explained, and whether there is a plausible story to be told for why this should happen. In this context, he concentrated on the question of whether firms play an earnings-guidance game where managers release information selectively to bolster the share price, thereby producing some benefit either directly for themselves or for the future equity issuance programmes of the firm. Comparing US companies in the 1980s to the 1990s, he observed that there has been a trend towards stock-based compensation that should give managers incentives to influence the stock price. In addition, he alluded to mainly firm-initiated rules on insider trading, restricting it to time periods immediately after earnings announcements; the argument here apparently being that in these periods, there would be less insider information to trade on. Furthermore, changes in US rules on insider trading initiated by regulators in May 1991 imply that managers now have a more precise target date for when to exercise their stock options. Given these structural changes in the 1990s, Wysocki argued that there are now higher incentives to achieve beatable forecasts. One would also expect a higher likelihood of observing pessimistic forcasts for firms with greater needs to sell stock after an earnings announcement: firms that want to issue stock in the future have an incentive to generate a greater number of positive earnings surprises. Issue 41 July 2001 MARKET TECHNICIAN 11 The paper calculates ex-post forecast errors from the Institutional Brokers Estimate System (I/B/E/S), which contains individual analysts, monthly forecasts, from 1983 to 1998. The short-term errors (e.g. one month prior to the earnings announcement) are then regressed on sub-period dummies, controlling for various other variables, to determine whether there has been a shift in pessimism. The results suggest that there has indeed been a shift towards small negative forecast errors in the 1990s. Further, from a cross-section regression, Wysocki found that the effect is indeed stronger for firms that issue equity after the earnings announcement, and/or where insiders sell the announcement. From this evidence, he concluded that institutional changes have made managers more likely to behave opportunistically and guide analysts expectations through earnings announcements in order to facilitate favourable insider trades. In his discussion of the presentation Pascal Frantz (LSE) reviewed the main features of the paper, and suggested that there may be several other reasons why forecasts are systematically pessimistic. He also proposed examining whether the magnitudes of differences of observed share price movements were economically significant, and whether they actually allow non-negligible profits to be made. Next Enrico Perotti (University of Amsterdam) presented joint work with Sylvia Rossetto (University of Amsterdam), entitled The Pricing of Internet Stock Portals as Platforms of Strategic Entry Options. The main motivation behind the presentation was to show how one could calculate valuations for companies that have a significant proportion of their value tied up in growth opportunities. Perotti presented a model based on the concepts of strategic real options and platform investment. Platform investment is the creation of an innovative distribution and production infrastructure, which increases access to customers and as a result reduces entry costs in related products. Examples include the development of a software operating system or an internet portal web page. Having a platform technology confers strategic pre-emptive advantages, creating a set of entry options into uncertain markets. In the model, two firms compete as parallel monopolies in a market with two differentiated products, and contemplate cross-entry into the other firms market. Demand in the market evolves stochastically over time and follows a geometric brownian motion. In the absence of strategic considerations, the standard argument applies there is a positive value to waiting due to the fact that profits are convex in demand. Here, however, strategic interaction creates a first-mover advantage. In the model, there is a tension between these two effects. Asymmetry is introduced by giving one firm the platform technology, essentially lower costs of cross-entry, and the other one conventional technology. Depending on parameter values, there are three different possible outcomes: (a) there is no cross entry, (b) the platform firm cross- enters, but the conventional firm does not follow, producing a de facto monopolist, or (c) the platform firm leads and the conventional firm follows. In the latter two cases, the platform firm will derive an advantage from its technology. The central result that emerges is one where, with higher uncertainty, the platform strategic advantage increases and firms tend to behave more as a de facto monopolist. Perottis presentation was discussed by Pierre Mella-Barral (London Business School), who reviewed the model and suggested some simplifications in the exposition of the model, as well as proposing the introduction of a choice between investing in conventional or platform technology as an additional stage in the game. The second day of the Conference opened with a session in which the Valuation of Technology Securities was discussed. The session, chaired by Bob Nobay, opened with a paper presented by Meir Statman (Leavey School of Business, Santa Clara University), which was a joint work with Kenneth L. Fisher (Fisher Investment Inc.) and entitled Cognitive Biases in Market Forecasts. Statman discussed several cognitive biases that people are likely to have in making market forecasts, presented examples related to the importance of Price-Earnings ratios for forecasting stock prices, and drew implications for good investment management practice. The five types of bias they discuss are (a) overconfidence, (b) confirmation, (c) representativeness, (d) anchoring and (e) hindsight. Overconfidence refers to those forecasts made with a precision that is too high, a mistake that psychological studies show people make very often people tend to attach too much significance to imprecise information they receive. Confirmation refers to the case when people choose to look exclusively at data that supports their preferred hypothesis. Representativeness alludes to forecasters reaching conclusions by comparing one security or group of securities to another, but in an inappropriate way. There is a bias towards highlighting similarities and ignoring differences across different time periods. The anchoring bias involves the tendency to anchor estimates to salient numbers even if these numbers have little or no relevance to the estimates. For example, people might have expectations of the level of the market related to the 1929 crash, even though future levels of the market might not be related to the level of the crash at all. Another cognitive bias arises when we know the reason for market events ex post, and use these to overestimate the quality of forecasts that happen to be right. Of course, with hindsight it is easy know which was the best forecast, but is ex ante accuracy will in general be lower than its ex post accuracy. As remedies, Statman stressed the importance of checks and balances through an appropriate organisation of decision making, and the role of colleagues and committees in making sense out of forecasts. Competing views can moderate one another, which might serve to suppress cognitive biases. The presenter also highlighted the importance of quantitative tools in providing a guide, although they come with their own pitfalls and limitations. Summing up, Statman stressed the need to be humble about ones own forecasting abilities, and the need to recognise the true nature of the financial services industry, that its not about making excess returns, its about selling hope. Gregory Connor (FMG/LSE), gave a brief synopsis of the paper, and extended the idea of the need to be aware of cognitive biases from the investment management discipline to the production of scientific research. In trying to find academic explanations for asset markets behaviour, he argued, we must also be aware of not falling prey to the same biases. The next presentation saw Josef Lakonishok (University of Illinois Urbana-Champaign) present a joint paper with Lous Chan and Theodore Sougiannis (all at the College of Commerce, University of Illinois Urbana-Champaign) on The Stock Market Valuation of Research and Development Expenditures. The problem that the Lakonishok tackled is how to value intangible assets, taking the concrete example of Research and Development Expensiture (R&D). In the US, accounting standards require that R&D be expensed rather than capitalised. However, if todays R&D provides benefits in the future then, in principle, this suggests that such an item should be capitalised this in general would lead to initially higher earnings and, of course, bigger balance sheets. The question then is: do capital markets take account of R&D in the correct way, and are R&D stocks fairly priced? A consequence of expensing R&D for financial analysis would be that Price Earning ratios or Price to Book ratios might not be reliable, and stocks could be underpriced. On the other hand, investors might be too optimistic about R&D intensive stocks, leading to the opposite mistake being made. To answer these questions, Lakonishok examined Compustat datasets, and calculated returns of portfolios of stocks with different R&D intensity, measured either as R&D spending to sales, or as R&D spending to market value. He found that, for the first measure, stocks with a high R&D intensity tend to show the same performance as stocks of companies that do not perform R&D. For the second measure, however, R&D intensive companies exhibit an excess average annual 3-year return in the range of about 6 per cent. Lakonishok reported that when he sub-divided this last portfolio into companies with high sales and companies with small sales, they find that companies with high sales tend to outperform. The explanation that he provided is that high sales/high R&D companies are relatively large and developed companies (rather than high-tech start-ups) that have underperformed prior to the observation period, possibly due to the fact that analysts do not account for R&D properly. Arguably, it is the fact that analysts do no 12 MARKET TECHNICIAN Issue 41 July 2001 capitalize R&D that is causing the mis-pricing of these stocks, leading to the outperformance later on. He also cast the analysis in a more rigorous Fama and French three-factor regression model, and arrived at similar conclusions. In his discussion, Ian Tonks (University of Bristol/FMG) stated that even though, in principle, R&D should be capitalised, there are important reasons not to allow this, since this might create the wrong kind of incentives in terms of the appropriate financial reporting. He agreed, though, that analysts must impute capitalisation if they want to arrive at reasonable values. He suggested that future research should consider a comparison with the UK where the capitalisation of software development is allowed. David Webb (LSE/FMG) chaired the fifth session which dealt with Valuation of Internet Securities. The first paper presented by Stephen Cooper (UBS Warburg) was entitled Navigating the i-Valuation Jungle. It focused on the challenges encountered for valuation and financial analysis of internet based companies. Cooper highlighted the fact that traditional methods and techniques for valuation, traditionally based on accounting numbers, are not suitable for the analysis of internet companies. The major difficulty is that accounting profits are poor indicators of future performance for companies in this sector. Despite the fact that uncertainty is present in valuation of old economy companies, the situation becomes somewhat more complicated in the internet sector. For these companies, the history of past results is not long enough to allow any inferences and more importantly, the real value drivers have not been well established. The paper contained a model for absolute valuation based on the traditional discounted cash flow model and target multiples. The model is based on the input of key data (revenue growth rate, cost of capital, business margin, etc) which generates target multiples for valuation which are compared with the current patterns for those companies. The author also commented on the real option valuation framework, criticising its empirical validity. The next presenter was John Hand (Kenan-Flagler Business School, University of North Carolina-Chapel Hill) with the paper Profits, Losses and the Non-Linear Pricing of Internet Stocks. His central theme was how accounting information can help in the valuation process of internet-based companies. The major hypothesis was that accounting should not be relevant to explain the behaviour of internet-based stocks, because traditional accounting rules (US-GAAP on the paper) focus on the evaluation of tangible assets. He criticised traditional accounting research which assumes a linear and stable relationship among the variables. Those assumptions, Hand argued, are not realistic due to characteristics of the accounting system such as conservatism (biased recognition). The paper evaluates the value relevance of profits, losses, intangibles and the linearity assumptions. There is evidence that the pricing of accounting information for internet stocks is non-linear. First, losses are priced negatively though they may be caused by huge investments in intangibles and R&D and need not be associated with poor performance. Second, intangibles are priced as expenses when income is positive and priced as assets when income is negative. The authors explanation for this phenomenon is that the stock market only attributes future benefits to intangibles when they play a significant role in the firms strategy. This is exactly what happens for loss-making firms that spend more than twice as much of their revenues on intangibles as do profit-making firms. Finally the paper reports that when income is positive, operating gross margin is more positively priced than when income is negative. These results show that the traditional assumptions of accounting- based valuation (linearity and constancy of the parameters over time) are not appropriate for internet firms. Eli Talmor (London Business School), discussing the paper, commented on the originality of the paper and the need for caution in interpreting results from research that is contemporaneous to the data. He also pointed out that internet firms provide a good laboratory to test established theories and that the econometric design of these studies must be improved. The third presenter of the day was Baruch Lev (Stern School of Business, New York University) with the paper A Rude Awakening: Internet Value Drivers in 2000. His paper aimed at identifying value drivers for business-to-consumer (B2C) internet companies both before and after the internet bubble in the spring of 2000. The paper uses a relative valuation approach (cross-section) considering web traffic performance, strategic alliances, cash and traditional financial indicators. Lev pointed out the intrinsic difficulties in the absolute valuation approach, considering it an exercise of educated guess and excessively complicated. The paper presents evidence or correlation between monthly stock returns and lagged web traffic metrics and also discriminates between the most relevant dimensions of web traffic performance: reach, stickiness and customer loyalty. Lev also showed that investors apparently capitalise R&D and marketing expenses during the bubble period and that this is no longer taking place after the shakeout in the spring of 2000. Finally, his empirical model seemed to be able to identify over-valued internet stocks over the sample considered in the study. The discussant, David Blake (Birkbeck College) commented on the accuracy and reliability of accounting data for internet companies and suggested improvements related to other performance measures and longer datasets. The last section of the conference was a Round Table chaired by David Webb (LSE/FMG) with the title Valuation Models for Technology Stocks: Value Driven or Going With The Wind?. The participants of the Round table were Ron Anderson (LSE/FMG), Walter Kemmsies (UBS Warburg), Josef Lakonishok (University of Illinois at Urbana-Champaign), Peter Martin (Financial Times), Miguel Ortiz (Boston Consulting Group) and Kent Daniel (Kellogg School of Management, Northwestern University). Each speaker began with some brief comments, which were followed by a lively discussion. Daniel started by commenting on the anomalies found in recent studies and how these are inconsistent with the rational expectations model. He pointed out that overconfidence has a strong role in market behaviour and that in general people are more overconfident when the situations are fuzzy; this point has not been established in the existing models. However, this kind of behaviour does not necessarily create opportunities for arbitrage. Lakonishok noted that there is no evidence of companies providing the high returns over a long period of years implicit in the high market prices we have seen for some internet companies. He observed that growth rates in this sector are almost unpredictable, and that they are excessively dependent on technological developments that are not captured by traditional valuation models. Kemmsies emphasised the importance of empirical considerations and commented that many of the challenges presented by internet-based companies, such as high growth rates and valuation of growth options, are being felt in other sectors as well, such as the telecom companies. Anderson observed that the high volatility of those markets makes it almost impossible to discriminate between alternative hypotheses. In such a scenario it is hard to extract relevant information from the data. He went on to support the argument that technological events make the forecasts of growth rates very much a hopeless task. He also questioned what really differentiates losers from winners in those markets and if analysts evaluations were really time-consistent. Ortiz focused on the importance of the New Economy for the Old Economy pointing out that, as a consultant, he has to focus on strategic issues related to the creation of value. He highlighted that the value drivers that had been discussed in some of the papers are of extreme relevance for managers. He also discussed the limited scope for valuation based on market values and questioned the role of the fundamentals for the firms involved in the New Economy. Martin presented his view that traditional investment theory is not well suited for the new type of technology firms that are being formed. He analysed the enormous impact of internet based technologies on real activity and the consequent transfer of wealth among agents which may be creating an environment for overinvestment. continues on page 13 Issue 41 July 2001 MARKET TECHNICIAN 13 After these first comments the general discussion, with questions from the floor, focused mainly on the valuation of internet firms and the possible reconciliation with existing valuation models. The valuations of technology securities over the recent past seem to suggest that every single firm in this sector will be a winner when there is evidence that suggests that, in every sector, there are winners and losers. The distinction between firm types has been blurred for technology firms, mainly through highly exaggerated growth rates. The nature of accounting information and its use by analysts was a major issue addressed. Other comments focused on the incentive of investment bankers, brokers and analysts, mainly referring to the fact that the number of buy recommendations have been, and continue to be, much larger than hold and sell recommendations. It was suggested that such recommendations are likely to lead to overconfidence in valuations. After a round of comments and questions from the audience, David Webb brought an end to the two-day Conference, and thanked the presenters and participants for a very lively and stimulating conference, which helped not only encapsulate the current thinking but also clearly laid the themes which are a priority for further work. This article first appeared in the Financial Markets Review. FMG Review April 2001 pp 1-5, 14-15, Number 50 The conference papers are available at http://fmg.lse.ac.uk under events, past conferences, market rationality and the valuation of technology stocks. I will bind a copy and place it in the STA Library (M. Feeny) The Way To Trade Discover Your Successful Trading Personality By John Piper FT/Prentice Hall (Pearson Education); 1999; 282pp; hardback John Piper is a long-time newsletter writer and money manager, who has strong views on particular aspects of technical analysis, particularly indicators, Market Profile, and behavioural finance. The first he (generally) despises, the second two he strongly favours. Much of the book (as the sub-title suggests) has to do with the third, the psychological aspect of trading. The title is valid in another respect this is a primer for traders, rather than investors (even though longer-term traders would also doubtless benefit from its lessons) since much of the content relates to trading in equity futures and options, mostly in the UK markets. John has no shame in admitting his mistakes, but equally he takes pride in where these mistakes have taken him to a secure position as one of the UKs leading newsletter writers, and a money manager well on the way to earning his CTA (which takes a five-year audited track record). It is educational to see how he has got there. It is also interesting to see who has had particular influence on his thinking. One Appendix (out of five) is a reprint of a Tony Plummer article on the brain. The introduction is by Alexander Elder (psychologist, trader, and writer of Trading for a Living, a splendid book for those seeking self-knowledge). Acknowledgements are made to Adrienne Laris Toghraie, a well-known US traders coach and writer and coach to John Piper himself. But the main influence has been Johns own trading, and it is fascinating to discover what he has learnt. Each chapter of the book (which itself is split into two main sections and two smaller ones) ends with a summary of the topics discussed, and these are excellent reminders of what might need to be re-read when returning to the work. A quick riffle through the pages is not going to be productive; there is a lot to be learnt here, as the trader is taken through the various stages of how best to hone his skills. This book is written in a down-to-earth and practical style. You may not agree with all he says (particularly regarding indicators!), but you will be stimulated into thinking about why you might disagree with him, and that thinking process may to help you to become a better trader. Aggravating? Yes. Stimulating? Yes. Worth the read? definitely. M. Smyrk TraderMade TraderMade has developed a range of desktop applications and web-based services. TraderMade Workstation is the top of the range and has been designed for banks and high-end technical analysis users. It is available as Standard, Intermediate or Advanced versions to cater for all levels of sophistication and budget, and data can be delivered via leased line, satellite or internet. TraderMade Web has been designed as a browser version for total portability, with users accessing a central database for their preset charts, while the desktop version offers speed and ease of use. Other web services include TraderMade Weblink, which allows real-time charts (containing clients analysis) to be linked directly in real time to any clients website, and TraderMade Charting is a plug-&-play applet (via internet or intranet) for customers and employees to do their own technical analysis direct from the client's web-site. Prices are calculated per licence or on an enterprise licence basis, and all products can be contended between a number of different users to spread the costs. Flexible Global Contracts are encouraged by means of an attractive global sliding price scale. For more details, visit us at www.tradermade.com * * * * Virtual Trading Seminars on CD Clickevents.co.uk distribute a series of recorded training seminars on CD, the Microtrader seminars work out at under 20 and under a current special offer have a past seminar CD for under 5.00. These seem great value for money to me, so you may wish to check it out. www.clickevents.co.uk * * * * Metastock Portfolio level Testing The ability to carry out portfolio level testing has been lacking from many technical trading packages, Metastock included. But now with the release of MSBT (Multiple Security Back Testing) those with Metastock can achieve this elusive goal. Details can be found at http://www.holygrailsoftware.com/msbt.shtml. Cost $45. The site also offers a variety of other Metastock tools. * * * * Gobal Stock Selection and filtering The Henley Investor Suite of stock selection investor tools enables you to select and filter the most promising stocks for analysis. Unlike many vendors Henley offers a variety of both UK and overseas data packages, allowing gobal stock charts to be compared. Also included in charting package. www.henleyinvestor.co.uk Tel: (44) 020 8332 1094 * * * * Trading and Tradestation Support in the UK? Scalper UK offer trading support and free seminars in Covent Garden promoting their methods of trading the Emini futures contract using Tradestation via an indicator package. Pity there was no defined system tests, but their seminar covered the several aspects of setting up a trading desk. You can register for their free seminar via their website www.Scalper.co.uk Scalper also offers Tradestation Support and Sales. Bytes and Pieces David Watts Book Review 14 MARKET TECHNICIAN Issue 41 July 2001 Introduction This talk will discuss three models of future price movement, one which is to all intents and purposes totally random and directionless, one which has a modest element of non-random (or directional) behaviour, and one in which there is a higher element of non- random behaviour (directional) behaviour. Techniques will be given which can take advantage of the characteristics of these models to generate increasingly large profits. Model 1 Totally random An underlying assumption in the valuation of options is that past daily changes in the price of the underlying security followed the normal distribution. A second assumption is that future prices, at least in the short term, will follow this same distribution. The distribution is easily calculated from a knowledge of the volatility of the security. From this the probability of occurrence of various price levels for the security at some point in the future, for example in 30 days time, can be determined. Figure 1 shows a plot of Marks and Spencer up to 8th June 2001 (price = 252.5p) with the probability distribution for its price 30 days onwards. From this curve probabilities such as the price being greater or less than a certain level, or staying within a particular price band, or being an exact value on the day 30 days into the future can be estimated. Such techniques are used extensively in options strategies, for example in the buying or writing of straddles. A straddle involves the buying or selling of both a call and a put at the same exercise price. The upper and lower break-even points lie above and below the current price and are readily calculated from the premiums. The buyer of the straddle is expecting the price to move outside either of the two break-even points and doesnt particularly care in which direction the large move takes place, while the writer is expecting the price to stay between the two points. Thus the buyer wants the future volatility to be high while the writer wants it to be low. The probability curve can be used to calculate the probabilities of making a profit for either buyer or seller. In view of the fact that it is easily demonstrated that share price movement does not follow the normal distribution, it is difficult to understand why this method is still followed so widely. When the daily price changes of a share are plotted as an histogram, it becomes obvious that there are price movements which are much larger than expected (in statistical speak the tails of the distribution are too long). Figure 2 shows the frequency of various daily price changes for Marks and Spencer over the last 1000 days. The largest one day fall was 53.75p and the largest one day rise was 40p. At the time these occurred they represented changes of 13.6% and 18.1%. A number of other large changes can be seen, for example falls of 45p and 30p and rises of 40p, 37p and 35p. On the basis of a normal distribution of daily price changes, the probability of these large changes occurring is vanishingly small. The fact that they occur therefore negates the use of the normal distribution of daily price changes as a way of calculating future prices. Thus the conclusion is that although this model assumes a totally random price movement, the distribution upon which this price movement relies is something other than the normal distribution. The approach I have developed to determine future price levels assuming a random model is to use the actual price changes themselves, such as those shown for Marks and Spencer. These changes should be calculated for a period back in time of at least four times the period into the future for which the price levels are required. Thus, taking the 1000 daily changes for Marks and Spencer, it is possible to look ahead for 250 days. One simulation of the future price movement is then obtained by, as it were, putting all of these changes into a pot and taking them out one at a time until 250 have been taken. The first change taken out is added to the current price to produce the estimated price for tomorrow. By this means, the price can be rolled forward a day at a time as each successive price change is taken out of the pot and added to the latest price. This process is known as bootstrapping. By itself, this one simulation is meaningless. In order to build up probabilities for the future it is necessary to run the simulation say 1000 times. None of these simulated future price movements will be identical with the actual price movement that occurs, but the purpose of the exercise is to see what proportion of these simulated price movements take the price above or below specified levels in the look ahead period. Some of the simulations will violate both price levels. In the case of options where a straddle is envisaged, the price levels would be the upper and lower break even points. In Figure 3 are shown the results for Marks and Spencer. The levels chosen here are 220p and 300p with the latest price standing at 252.5p.Three time scales are used, representing the near, middle and far expiry dates. The results are presented as probabilities that the price will be outside these limits. The possibilities are to be outside one limit, or outside both limits, and the position at expiry is also specified. It can be seen that, for example, in the 1000 simulations, the price moved outside of one or other of the limits 818 times (81.8%) during the 40 day period, 993 times (99.3%) during the 131 day period and 1000 times (100%) during the 222 day period. An option trader with a straddle that required the price to move either above 330p or below 220p during the next six or nine months would have a very high probability of success. In checking this method for a large number of situations over the last five years, all those instances in which the estimated probability has been over 99% has resulted in the price moving outside of one or other of the two Predicting future price movement This article is a summary of the presentation given by Brian Millard for the STA meeting in April 2001 Figure 1 Figure 1: The probability distribution for the Marks and Spencer price 30 days on from 8th June 2001. Figure 2 Figure 2: The frequency of daily price changes in Marks and Spencer over the last 1000 days. Issue 41 July 2001 MARKET TECHNICIAN 15 limits. Thus, quite clearly, it is possible to generate consistent profits even if it is assumed that price movement is virtually totally random. Model 2 Partly random It can be demonstrated easily that share price movement is not totally random, but has at least a modest directional element which can be subjected to predictive techniques with a high degree of success. The directional element is due to the presence of cycles in the data, and as an example, the algebraic sum of the cycles with a nominal wavelength of between 21 and 41 weeks are shown for Halifax shares in Figure 4(upper panel).The lower panel is the plot of the weekly closing values of Halifax over the same period of time, and it can be seen quite clearly that major peaks and troughs in the data coincide with peaks and troughs in the cycle. The regularity of the cycle is quite impressive, with modest changes in amplitude and wavelength across the time frame which are caused by random influences. The method of isolating narrow bands of cycles is to select the upper wavelength of interest as an odd number (41 in the case of the Halifax example) and calculate a centred average using the same span of 41. Then take the nearest odd number to one half of this span, i.e. 21 in the current example and calculate a triangularly weighted centred average. The difference between these two averages when plotted gives the narrow band of cycles as shown in Figure 4. The plot will end half of the higher span back in time since the calculation is based on centred averages. The path of the cycles since that time can be estimated by various mathematical techniques, but it must always be remembered that the latter part of the plot is indeed an estimate. The predictive value of these cycles is enhanced by calculating a number of them separated fairly widely in wavelength. Thus in Figure 5 is shown the band of cycles with nominal wavelength between 15 and 31 weeks. Both Figures 4 and 5 show the position as at September 2000, and it is estimated that the 15/31 week cycles have just passed their bottom, while the 21/41 week cycles have a little way to go before bottoming out. Thus the interpretation of these results is that in a few more weeks the cycles will be rising in unison and should take the share price with them. The vertical sweep of the cycles is in pence and can be used to estimated the contribution which they will make to the share price. Thus the 15/31 week cycles should give a rise of around 80p and the 21/41 cycles a rise of 100p. The arithmetical sum of these two is 180p, and this is the expectation of the rise which should follow if these two cycles are still in evidence. That this method can be extremely accurate is shown by the subsequent movement of Halifax shares from September 2000 to February 2001 (Figure 6). A substantial upwards movement occurred taking the price up by nearly 200p. Thus, not only does the analysis of cycles present in share price data give valuable information about the start of a trend, but also of by how much it can be expected to cause the price to change. Model 3 Highly directional Moving averages, especially triangularly weighted ones, represent the net result of all of those cycles with wavelengths more than 1.5 times the span of the average. Since a trend can be considered to be one half of a cycle (either the rising or falling half), then the average is a good representation of all of those trends which have a time scale of about three-quarters of the span of the average. This is true, however, only if the average is plotted as a centred average. Taking Boots as an example, a 201 day centred, triangularly weighted average is shown superimposed on the daily share price up to 6th April 2001 in Figure 7. The fact that various sections of the plot of the average follow faithfully the overall direction of the price illustrates the relationship between trend and centred average quite clearly. What is also obvious is that excursions of the price away from the position of the centred average are quite limited in extent. The largest excursions can be contained between two boundaries set equidistant above and below the position of the average. These boundaries are therefore exact duplicates of the centred average itself as far as shape is concerned. These two boundaries form a constant depth channel within which the price movement is Figure 3 Figure 3: The results from 1000 simulations of the Marks and Spencer share price over the next 222 days using as a source the daily price changes from the last 1000 days. Figure 5 Figure 5: Upper panel: Cycles of wavelengths between 15 and 31 weeks in the Halifax share price. Lower panel: Halifax share price as weekly data. Figure 6 Figure 6: The subsequent movement of Halifax share price to February 2001. Figure 4 Figure 4: Upper panel: Cycles of wavelengths between 21 and 41 weeks in the Halifax share price. Lower panel: Halifax share price as weekly data. 16 MARKET TECHNICIAN Issue 41 July 2001 contained. The drawing and estimation of the current and future positions of these boundaries constitute the technique of channel analysis. Since the share price normally reverses direction once it arrives at the channel boundary, the description of this model as highly directional can be understood. In Figure 8 the boundaries have been drawn so as to enclose some 97% of the price data points. It is permissible to allow a small tolerance in placing the boundaries so that small penetrations of either boundary can occur if this brings a large number of price peaks and troughs to the boundary positions. It is remarkable at how many places the price reaches a boundary and then reverses direction. Note that since the channel is based on a centred average, it terminates half a span of the average back in time from the present. There are mathematical techniques which can predict the path of the channel from this cut-off point up to the present time and beyond. The channel boundaries represent places where there is a low probability of the price remaining for more than a day or two. There is a high probability that the price will reverse direction, and a lower probability that the price will continue in the same direction. Thus the high probability is that the channel boundaries represent areas where a new trend will probably start. Not only that, but since the channel contains the price within it, the general channel direction is also of importance since it gives an indication of the longer term trend. In the case of Boots this long term trend can be estimated as rising, and it can be expected that the price will eventually (in the next few months on from April) meet the lower boundary of this rising channel and be converted into a rising short term trend. The rate at which the short term trend causes the price to rise will then be augmented by the rising outer channel. Rentokil in 1999 is an example where the price did not reverse direction but continued outwards from the channel boundary. Thus in Figure 9 is shown the position on the 7th May 1999. It would be anticipated that the price would bounce up from the lower boundary, which is at about 360p, to a level somewhere around 480-500p. However, as can be seen from Figure 10, the price made a disastrous fall through the estimated lower boundary to the 240p level. When looked at in retrospect, it can be seen that the channel made a rapid downward hook at this point, taking the price down with it. In practice, the ratio of occasions in which the price retreats from the boundary to those occasions where the price continues through the boundary is about five to one. A stop loss is essential to protect the investor against such possible adverse moves. Options traders can of course take a position which will take advantage of this five to one ratio. Brian Millard is the author of five books on technical analysis published by Wiley in the UK and a sixth Channels and Cycles published by Traders Press Inc in the US. He owns Qudos Publications Ltd which carries out independent research on mathematical methods of analysing stock markets. Figure 8 Figure 8: The average from Figure 7 has been used as a template to produce a constant depth channel. The channel has been extrapolated to the present time. Figure 9 Figure 9: The position for Rentokil in May 1999 suggests a rise from the lower boundary should occur. Figure 10 Figure 10: The subsequent movement of Rentokil. Figure 7 Figure 7: A 201 day triangularly weighted centred average superimposed on the Boots share price. Spreadsheets and Markets More expertise in market analysis than in Excel spreadsheet design? More expertise in Excel spreadsheet design than in market analysis? Why not trade your expertise by joining the Excel Traders List, for free, at http://groups.yahoo.com/group/xltraders or e-mail: xltraders-subscribe@yahoogroups.com Queries? Dr Mike Wignall MSTA: wigmail@blueyonder.co.uk