In this Part I of a series, I summarize a sampling of academic studies on the effectiveness of technical trading. In Part II, I'll synthesize the academic studies and add my own thoughts, and in Part III, I'll discuss the findings with regard to the development of CASTrader.
Teresa Lo at PowerSwings has a treasure trove collection of studies finding mixed degrees of utility in technical trading rules (note: PowerSwings has copies of the papers, but alternate links are given) and they are summarized below:
In Risk-adjusted, Ex Ante, Optimal, Technical Trading Rules in Equity Markets (1999), Neely creates a genetic algorithm to evaluate trading rules and finds that the system cannot outperform buy-and-hold on a risk adjusted basis. The authors state that evaluation of risk is central to the evaluation of a trading system, and the paper discusses several ways to examine risk. It's not entirely clear, but the "rules" appear to be functions of moving averages.
In Data-snooping, technical trading rule performance and the bootstrap (1997), the authors re-examine an old previous study and find that the trading rules that had superior performance there lost their superiority in the 10 years since the study of moving averages and trading range breaks. They discuss the hazards of datasnooping in the studies in the financial literature, and discuss a method called White's Reality Check to check for data mining. In this paper, the authors examine a wide range of trading rules, including filter, moving average, support and resistance, channel break-outs, and on-balance-volume averages applied to the Dow on 100 years of data as well as S&P 500 futures data. They find superior rules to the previous paper, but none that stand up to the data mining effect.
In Adaptive Market Hypothesis: Evidence from the Foreign Exchange Market(2006), the authors find that "The excess returns of the 1970s and 1980s were genuine and not just the result of data mining. But these profit opportunities had disappeared by the mid-1990s for filter and moving average (MA) rules. Returns to less-studied rules, such as channel, ARIMA, genetic programming and Markov rules, also have declined, but have probably not completely disappeared....These regularities are consistent with the Adaptive Markets Hypothesis (Lo, 2004), but not with the Efficient Markets Hypothesis."
In Comprehensibility and Overfitting Avoidance in Genetic Programming (2003), the authors add a complexity penalizing factor to the fitness function to avoid overfitting and find rules that outperform buy-and-hold on the S&P 500 after transaction costs from 1990-2002. More here and here.
In Re-examining the Profitability of Technical Analysis with White’s Reality Check, the authors find profitable rules for NASDAQ and the Russell 2000, but not the S&P 500 and DJIA, even after adjusting for data snooping bias. These authors examine a large universe of strategies, larger than (2) above. The authors state "the best rules for DJIA and S&P 500 are, respectively, a momentum strategy in volume and a contrarian rule in the OBV class. Neither of these rules yields statistically signiﬁcant return based on the Reality Check p-values. The same conclusion also holds when the performance measure is Sharpe ratio. On the other hand, the proﬁts of the best rules for NASDAQ Composite and Ruseell 2000 are statistically signiﬁcant at 1% level. For the former, the best rule is the 2-day MA rule with 0.001 multiplicative band, yielding average daily return 0.00152 (or 38.19% annually); for the latter, the best rule is the 2-day simple MA rule that gives average daily return 0.00186 (or 47.1% annually)." In A Stepwise SPA Test and Its Applications on Fund Performance Evaluation (2006), the authors propose alternative data snooping methodologies to White's Reality Check discussed in (2), (3) and (5) above and find that "only eight mutual funds are found to beat the S&P 500 index, and only few hedge funds outperform the risk-free rate. With these two empirical cases, we substantiate the empirical value of our test in fund performance evaluation."
In The Profitability of Technical Trading Rules in US Futures Markets: A Data Snooping Free Test (2005), the authors re-examine a previous study using a wide range of technical indicators and find: "Results indicate that in 12 futures markets technical trading profits have gradually declined over time. Substantial technical trading profits during the 1978-1984 period are no longer available in the 1985-2003 period."
I'll add my own discovery of a study that looks at something completely different from all of the above: candlestick patterns. In The Predictive Power of Price Patterns (1998), the authors look at all S&P 500 stocks from 1992 to 1996 examine the predictive ability of candlestick patterns and state:
Out-of-sample tests indicate statistical significance at the level of 36 standard deviations from the null hypothesis, and indicate a profit of almost 1% during a two-day holding period. An essentially non-parametric test utilizes standard definitions of three-day candlestick patterns and removes conditions on magnitudes. The results provide evidence that traders are influenced by price behavior. To the best of our knowledge, this is the first scientific test to provide strong evidence in favor of any trading rule or pattern on a large unrestricted scale.
Trading Platform in Brazil