|
G.A.W. Griffioen, "Assessing simple trend-following strategies in the stock markets", the Technical Analyst, July/August 2004
|
Subject Matters
|
Assessing Simple Trend-Following Strategies in the Stock Markets
by Gerwin Griffioen
|
In his mammoth 322 page PhD thesis "Technical Analysis in Financial Markets", Gerwin Griffioen assesses the success of 5350 trend-following trading rules for cocoa futures and 787 trend following trading rules for stocks and stock indices from around the globe, constituting one of the most comprehensive surveys of its kind. Below, he summarises its main findings in relation to stocks and stock indices. His research suggests that, unlike for FX markets, simple trend-following strategies provide insufficient returns to compensate for risk and that other techniques may offer superior results.
|
In this article some popular trend-following forecasting techniques are applied to a large number of financial data series. These forecasting rules are back-tested out-of-sample to see if they yield any profit.
Research design
The forecasting rules investigated are based on three very popular and simple classes of technical forecasting rules. The first class is based on moving averages. A buy signal is generated if the price crosses the moving average upwards, while a sell signal is generated if the price crosses the moving average downwards. The second class is based on Alexander's (1961) filters . A buy signal is generated if the price increases x% from the most recent through. A sell signal is generated if the price declines x% from the most recent top. The final class of forecasting rules is based on support-and-resistance, where if the price breaks through a support (resistance) this is considered an important signal to sell (buy). The forecasting rules described above are refined with a number of extensions. For example, one possible refinement is the time-delay filter. That is, a buy or sell trade will only be executed if a buy or sell signal holds for a certain number of periods. Other possible refinements apply commonly used risk management techniques, such as the use of stop-loss levels. By extending the three basic classes of forecasting rules with many kinds of refinements and by varying the parameter val-
|
ues, I construct a set of 787 technical forecasting rules. Of the 787 rules, 425 rules are based on moving averages, 170 on support and resistance, and 192 on Alexander's filters.
Inferences about the effectiveness of these trading rules can only be made by testing whether the rules that performed well in the past continue to perform well in the future. To do this, at the beginning of each month the technical forecasting rule that performed best in the preceding half year is found (selection period) and this forecasting rule is selected to generate trading signals during the month (testing period). In this article, I choose different time frames for the selection and testing periods. The results of the best selection and testing period combination are presented.
Data sets
The set of 787 technical forecasting rules is applied to the stock prices of all firms that were quoted in the Dow Jones Industrial Average (DJIA) in the period 1973-2002, to the stock prices of all firms that were quoted in the Amsterdam Exchanges (AEX) Index in the period 1983-2003, and to fifty local stock market indices of exchanges located in Africa, the Americas, Asia, the Middle-East, Europe and Australasia in the period 1981-2001 (see Table 1 for an overview). The results for the fifty local stock market indices are computed in dollars. The results for the firms quoted in the AEX Index are computed in euros. All
|
|
results are corrected for transaction costs. These include research costs for finding a good forecasting rule, the costs of setting up a trading system, price slippage, and the costs of trading. I vary transaction costs from 10 basis points to 100 basis points per trade, capturing a broad range of trader types from floor traders to home investors.
Results
Table 1 shows, for each financial series, the profits arising from the (recursively selected) trading rule with the best out-of-sample performance. These profits are measured in percentage points per year. The results in the table are after correction of 50 basis points transaction costs per trade. According to the Sharpe-Lintner capital asset pricing model , , any excess profits resulting from holding a certain financial asset in portfolio is just the reward for bearing the risk of holding the
|
|
Jensen's alpha
|
|
Jensen's measure is calculated by estimating the coefficient a in the Sharpe-Lintner capital asset pricing model: rt - rtf = a + ß (rtM - rtf) + et . Here rt is the return of the recursively optimising and testing procedure, rtf is interest, rtM is the return of a well diversified and passively held portfolio, and et is noise at day t. The coefficient ß measures the riskiness of the active strategy relatively to the passive strategy. If ß is not significantly different from one, then it is said that the strategy has equal risk to buying and holding the well-diversified portfolio. If ß >1 (ß <1), then it is said that the strategy is more risky (less risky) than buying and holding the well diversified portfolio, and that it should therefore yield larger (smaller) returns. The coefficient a measures the return of the strategy after correction for bearing risk. If it is not possible to beat a well-diversified portfolio after correction for risk (in which was technical trading rule profits are just the reward for bearing risk) then a should not be significantly different from zero.
|
|
|
July/August 2004 THE TECHNICAL ANALYST 39
|
|
Subject Matters
|
|
asset in portfolio. Therefore, I also calculate Jensen's alpha. This is a measure to test whether the profits that accrue from making investment decisions are significantly positive after a correction is made for bearing risk. If Jensen's alpha is not significantly larger than zero, then the investment decisions do not add to passively holding a well-diversified portfolio. In Table 1, an asterix denotes whether Jensen's alpha is significantly positive.
It appears that the simple technical forecasting rules I test are not so effective when applied to the stock prices of firms that were quoted in the DJIA. For none of these financial series is Jensen's alpha significantly positive. On average the profits are even negative. On the contrary, the simple technical forecasting rules do show profits when applied to the stock prices of firms that were quoted in the AEX Index. On average, the mean yearly return of a long/short trading strategy is equal to 15%. The forecasting rules show exceptionally good performance for firms in the communications and internet technology industries. However, Jensen's alpha is significantly positive for only five stocks, none of these belonging to the same industry. It is interesting to compare our results to Fama (1965) and Theil and Leenders (1965) . Theil and Leenders (1965) found that the proportion of securities advancing and declining today on the Amsterdam Stock Exchange could help in predicting the proportion of securities advancing and declining tomorrow. However, Fama (1965) in contrast found this was not true for the New York Stock Exchange. The results presented here suggest these differences in the forecastability of the Amsterdam and New York stock exchanges persist into the 1980s and 1990s.
The forecasting rules also perform well for the local stock market indices. On average, the mean yearly return of a long/short trading strategy is equal to 12.4%. It is noteworthy that the forecasting rules do not show any profits for North American and West European stock market indexes, but do show some remarkably good results for stock market indices of exchanges located in Asia,
|
| Time series (in Dollars) | Time series (in Euro) | Time series (in Dollars) |
| DJIA | 2.87 | AEX | -1.07 | World MSCI | -0.68 |
| Alcoa | -5.91 | ABN | 13.83 | Argentina Merval | 12.61 |
| American Express | -2.29 | AMRO | 12.64 | Brazil Bovespa | 22.78 |
| AT & T | -5.81 | ABN AMRO | 5.95 | Canada TSX Composite | 3.43 |
| Bethlehem Steel | 13.9 | Aegon | 13.15* | Chile IPSA | 30.34 |
| Boeing | -3.8 | Ahold | 12.18* | Mexico IPC | 30.9 |
| Caterpillar | -5.67 | Akzo Nobel | 1.07 | Peru Lima General | 41.18* |
| Chevron-Texaco | -2.5 | ASML | 52.79 | US S&P500 | -8.42 |
| Citigroup | -1.76 | Baan | 25.78 | US DJIA | -4.15 |
| Coca-Cola | -2.69 | Buhrmann | 7.37 | US Nasdaq100 | -11.64 |
| E.I. Du Pont de Nemours | -7.99 | Ceteco | 36.01 | US NYSE Composite | -4.47 |
| Eastman Kodak | -3.86 | CMG | 70.14 | US Russel2000 | 3.33 |
| Exxon Mobil | -0.87 | Corus | 81.83 | US Wilshire5000 | -8.46 |
| General Electric | -4.07 | CSM | -5.77 | Venezuela Industrial | 23.84 |
| General Motors | 3.47 | DAF | 15.95* | Australia ASX All Ordinaries | 5.08 |
| Goodyear Tire | 0.62 | DSM | 1.34 | China Shanghai Composite | 9.18 |
| Hewlett-Packard | 0.78 | Reed Elsevier | -8.15 | Hong Kong Hang Seng | 5.54 |
| Home Depot | 4.15 | Fokker | -2.34 | India BSE30 | 13.47 |
| Honeywell intl. | 1.7 | Fortis | -4.69 | Indonesia Jakarta Composite | 66.49* |
| Intel | 2.53 | Getronics | 14.44 | Japan Nikkei225 | -3.61* |
| Intl. Business Machines | -0.85 | Gist Brocadis | -0.51 | Malaysia KLSE Composite | 28.18* |
| International Paper | -4.39 | Gucci | 44.64* | New Zealand NZSE30 | -1.84 |
| J.P. Morgan Chase & Co. | 6.15 | Hagemeyer | 3.53 | Pakistan Karachi100 | 26.02 |
| Johnson & Johnson | -6.78 | Heineken | -6.85 | Philippines PSE Composite | 43.9* |
| McDonalds | -3.83 | Hoogovens | 17.63 | Singapore Straits Times | 11.06 |
| Merck | -3.98 | ING | -1.42 | South-Corea Kospi200 | 30.71 |
| Microsoft | 6.58 | KLM | 6.3 | SriLanka CSE All Share | 38.79* |
| Minnesota Mng. & Mnfg. | -4 | Kon. PTT. Ned. | 9.43 | Thailand SET | 29.13 |
| Philip Morris | -2.46 | KPN | 84.45 | Taiwan TSE Composite | 6.9 |
| Procter & Gamble | -5.43 | KPNQWEST | 39.2 | Austria ATX | 0.67 |
| SBC Communications | -3.79 | Van der Moolen | 18.32* | Belgium Bel20 | -6.22 |
| Sears, Roebuck & Co. | -1.65 | Nat. Nederlanden | 0.48 | Czech Republic PX50 | 21.61 |
| United Technologies | -6.32 | Nedlloyd | 5.31 | Denmark KFX | -7.28 |
| Wal-Mart Stores | 2.84 | NMB Postbank | 10.44 | Finland HEX General | 5.56 |
| Walt Disney | 1.65 | Numico | 5.61 | France CAC40 | -4.02 |
| | Oce | 16.84 | Germany DAX30 | -7.37 |
| | Pakhoed | 3.41 | Greece ASE General | 23.01 |
| | Philips | -1.49 | Italy MIB30 | 0.89 |
| | Polygram | -14.26 | Netherlands AEX | -5.32 |
| | Robeco | 0.4 | Norway OSE All Share | 17.07 |
| | Royal Dutch | -2.25 | Portugal PSI General | 14.13 |
| | Stork | -4.4 | Russia Moscow Times | 83.13* |
| | TPG | 6.07 | Slovakia SAX16 | 11.79 |
| | Unilever | -10.83 | Spain IGBM | -1.39 |
| | UPC | 105.38 | Sweden OMX | -2.31 |
| | Vedior | -22.18 | Switzerland SMI | -0.69 |
| | Vendex KBB | 4.27 | Turkey ISE100 | 27.57 |
| | Versatel | 121.03 | UK FTSE100 | -4.66 |
| | VNU | 2.98 | Ireland ISEQ | -0.85 |
| | Wessanen | -3.64 | Egypt CMA | 26.45* |
| | Wolters Kluwer | -12.54 | Israel TA100 | 0.88 |
| Average | -1.24 | Average | 15.05 | Average | 12.4 |
Table 1.
|
|
South-America, the Middle East, Eastern Europe and Russia.
When looking at the results for the S&P 500 and Nikkei 225, Jensen's alpha appears to be significantly negative, strongly indicating that passively holding a well-diversified portfolio is
|
preferred to using technical forecasting rules. Only for the stock market indices of Egypt, Indonesia, Malaysia, Peru, Philippines, Russia and Sri-Lanka, is Jensen's alpha significantly positive. However, these markets are less liquid than the other markets investigated in this article and transaction costs of 50 basis points
|
|
40 THE TECHNICAL ANALYST July/August 2004
|
|
Subject Matters
|
per trade could be too conservative. If transaction costs are raised to 100 basis points, then Jensen's alpha is significantly positive for the Egyptian CMA Index only.
Conclusion
Simple trend-following technical forecasting rules do not seem to be effective in forecasting the prices of North American stocks. The same forecasting rules, however, do seem effective in forecasting the prices of stocks quoted at the Amsterdam Exchanges. They also yield profits when applied to stock market indices of exchanges located in emerging markets. However, after correction for transaction costs and risk, the forecasting rules are not statistically effective for most data series. Hence, based on the statistical results outlined in this article, I cannot reject the weak form of
|
the efficient markets hypothesis. I do not claim that stock prices or indices are not predictable, but if there is some dependence in daily price changes, it is so complicated that it cannot be detected by applying the simplest technical forecasting rules.
Gerwin Griffioen is an analyst with Insinger de Beaufort Asset Management in The Netherlands. In 2003, he received a PhD in Economic Sciences from the University of Amsterdam for his thesis on technical analysis in the financial markets, upon which this article is based.
|
References
Alexander, S.S. Price movements in speculative markets: trends or random walks, Industrial Management Review, 1961, p. 7-26.
Fama, E.F. Tomorrow on the New York Stock Exchange, Journal of Business, 1965, p. 285-299.
Lintner, J. The valuation of risky assets and the selection of risky investments in stock portfolios and capital budgets, Review of Economics and Statistics, 1965, p. 13-37.
Sharpe, W. Capital asset prices: a theory of market equilibrium, Journal of Finance, 1964, p. 425-442.
Theil H. and Leenders C.T. Tomorrow on the Amsterdam Stock Exchange, Journal of Business, 1965, p. 277-284.
|
|