سودآوری تجزیه و تحلیل فنی در بازار و تحلیلهای مالی و کالا - بررسی واقعیت

Based on the SPA test (test for superior predictive ability), Sortino and reversed Sortino ratios, we examined the profitability of a universe of 8061 technical trading rules in ten futures markets including five financial and five commodity underlying assets. We tested whether the best performing rule really beats its buy-and-hold benchmark strategy in bullish and bearish markets, respectively, during the in-sample testing period. The best rules' performance relative to the benchmark is also tested during the one-year out-of-sample period for all ten sets of data. A novel set of multi-indicator rules, MFI–RSI, and four popular categories of single-indicator rules, filter rules, moving averages, on-balance volume averages and momentum strategy in volume, were employed to form our universe of trading rules. The results on the SPA test suggest market efficiency in nine of the ten futures markets, while the results on the Sortino and reversed Sortino ratios reveal persistent outperformance of the best ‘downside’ and ‘upside’ rules relative to the buy-and-hold benchmark across time in four and three futures markets, respectively.

Technical analysis has been one of the popular trading techniques among others in the futures markets for years. An excerpt from a report in BusinessWeek on March 19, 2001 [11] emphasizes the prevalence of technical analysis in futures markets: ‘…Most of the futures managers trade on the basis of technical, rather than fundamental, analysis, looking at such measures as price movements and changes in trading volume. They have developed analytical models based on the behavior of different futures markets over the years. “It's a systematic approach, and the systems are designed to profit when the futures products move through certain designated levels,” says Sol Waksman, president of Barclay Trading Group, which researches and tracks futures funds….’ Despite its importance in the futures markets, the profitability of technical analysis might be subject to the so-called data-snooping bias, a stylized fact that is common to research on repetitively discovering the best model to explain an economic or financial time series. When the same set of data is tested using a chunk of models, we will always find that one or two of them are able to explain the data to a satisfactory extent, but only by chance rather than by any specific ability of the models themselves. Such a search for the best model from an enormous union constitutes the so-called data-snooping bias. An implication of the data-snooping bias is that any previous evidence of the profitability of technical analysis might be subject to questions. Without taking into account the data-snooping bias, such a conclusion might be fairly due to the mere luck, which is not robust to variability in sample periods or in variables. Sullivan, Timmermann and White [23] tested five categories1 of simple trading rules each based on one single technical indicator, amounting to 7846 rules in total, for the S&P 500 index futures. STW's2 sample period is from 1984 through 1996 for the S&P 500 futures. Using White's [25] BRC to control for the data-snooping bias, STW found that, though some of the trading rules are able to beat the benchmark model (i.e. holding cash and staying out of the market) during the sample period above, the best rule of them is not able to show statistically significant outperformance relative to the benchmark with a possibility of 90.8%. STW concluded that profitability of technical analysis in the S&P 500 futures market might be due to the mere luck. In this article we aimed to extend STW's work by using more futures contracts, a more powerful test than the BRC to alleviate the data-snooping bias, adding a more sophisticated type of trading rules as well as improving on the way to calculate daily trading returns. In particular, we examined ten futures contracts consisting of five financials and five commodities. To alleviate the significant reduction in the rejection rate of the null hypothesis under the BRC due to too many poorly performing trading rules, we followed Hsu and Kuan [14] and employed the more powerful SPA test introduced by Hansen [13]. All five categories of trading rules in STW were based on a single indicator of different parameter sets, while we modified this simple universe by including a category of technical analysis based on a multi-indicator strategy, the MFI–RSI rules, in our universe of trading rules. Adding more complex trading rules to the set of rules tested would reduce the gap between previous academic studies and what the futures market practitioners have been gradually and really doing. STW and previous studies all ignored the issue of calculating the rate of return for futures trading in a more practical manner. On the contrary, we computed the return for a trading rule using both the futures margin account and the risk-free current account. Without taking into account the market practices in calculating returns for trading futures, the profits or losses will be underestimated which then leads to an underestimated standard deviation of the daily returns. The results of the SPA test, based on the studentized average daily return in our context, will also be biased. Moreover, STW and previous studies did not separate the trading rules' performance during downside markets from during upside markets. We responded to this issue by applying the SPA test to bull and bear markets separately and comparing the (reversed) Sortino ratio of the trading rules with their buy-and-hold benchmark counterpart. Last but not the least, we replaced the benchmark of ‘null’ system (i.e. always staying out of the market) in STW by the buy-and-hold strategy. 3 The ten futures contracts are CME Euro/USD FX futures (EUR/$ FX, hereafter), LIFFE FTSE 100 Index futures (FT-100, hereafter), EUREX DJ Euro Stoxx 50 Index futures (Stoxx-50, hereafter), SIMEX MSCI Taiwan Index futures (TW, hereafter), CBOT/CME US 30-year T-Bond futures (T-Bond, hereafter), CME live cattle futures (Cattle, hereafter), COMEX gold futures (Gold, hereafter), NYMEX/CME New York light sweet crude oil futures (Crude oil, hereafter), NYBOT/ICE coffee futures (Coffee, hereafter) and CBOT/CME soybean futures (Soybean, hereafter).4 The five categories of technical trading rules are, firstly, a hybrid of the money flow index and the relative strength index (hereafter MFI–RSI), followed by the filter rules (hereafter FR), the moving averages (hereafter MA), the on-balance volume averages (hereafter OBV) and the momentum strategies in volume (hereafter MSV). The MFI–RSI category is formed by coupling the money flow index with the relative strength index, a novel application in the academic literature. The MFI tracks the flow of money into and out of a market, which is often used to warn of trend weakness and likely reversal points. The RSI measures price strength by comparing upward and downward close-to-close movements, which is often used to indicate whether a security has been overbought or oversold and thus a likely reversal. Both indicators are formulated to fluctuate between 0 and 100, enabling prespecified overbought or oversold levels. Mixing these two indicators will efficiently reduce the number of noisy trading signals and increase the percentage of successful trades. Based on the SPA test to control for the data-snooping bias, we found that, the MFI–RSI and MSV types of technical trading rules tend to be identified as the best rule which significantly outperforms the benchmark in as many as seven futures markets, four financial and three commodity contracts. Nine of the fifteen sub-testing periods leading to a significant, best rule are in a bearish market, suggesting that the elites from the universe of our 8061 rules tend to perform better in a downside market than in an upside market. However, these ‘real’ outperformers selected from the most recent in-sample testing period generally fail to consistently provide a relatively good performance during the out-of-sample period. What is worse is that most of them generate a negative cumulative excess return, an evidence of no performance persistence among these in-sample ‘real’ outperformers. The only exception appears in the case of live cattle futures, in which the best trading rule selected from the last testing period, an MSV rule, still significantly beats the benchmark and generates a relatively large cumulative excess return over the one-year out-of-sample period. These results suggest that the remaining nine futures markets conform to the weak form of efficient market hypothesis, a finding in accord with what STW found with the S&P 500 index futures market. Our results based on the Sortino and reversed Sortino ratios calculated for the entire in-sample period tend to suggest performance superiority of the best ‘downside’ rule over the best ‘upside’ rule in that the Sortino ratio of the former is larger than the reversed Sortino ratio of the latter across all ten contracts. Unlike the results on the SPA test, the best ‘downside’ rule consistently outperforms the benchmark in more than half the eight futures markets where the best rule generated trading signals during the one-year out-of-sample period. In contrast, the best ‘upside’ rule keeps beating the benchmark in less than half the ten futures markets where the best rule generated trading signals in the out-of-sample period. However, most of these persistent ‘downside’ and ‘upside’ outperformers tend to give a cumulative excess return larger than what their benchmark brings about. Our results based on the Sortino and reversed Sortino ratios indicate an evidence of better profitability relative to the benchmark from using the best technical trading rule out of the universe of our 8061 rules in the Eur/$ FX, Stoxx-50, T-Bond and Crude oil futures markets. The rest of the paper is organized as follows. Section 2 presents the material and methods on the reality check of technical analysis' profitability, involving the trading rules, the SPA test, the (reversed) Sortino ratio and the data used. Section 3 details the practical issues of transaction cost and return calculation. Section 4 discusses the results, while Section 5 concludes.

Our SPA test results tend to suggest market efficiency in that the universe of our 8061 trading rules is generally unable to beat the buy-and-hold strategy although they were selected from among the most-profitable types of technical trading rules in previous studies. Only one of the nine ‘real’ (i.e. statistically significant) outperformers identified from the most recent testing period, i.e. an MSV rule in the live cattle futures market, consistently beats the benchmark over the one-year out-of-sample period. However, the results based on the Sortino and reversed Sortino ratios are not completely in accord with what we found with the SPA test. The best ‘downside’ rule and the best ‘upside’ rule selected from the entire in-sample testing period are both able to generate a rate of return exceeding the benchmark over the one-year out-of-sample period for a few contracts. Despite the inconsistency between the two sets of results on performance persistence of the trading rules, we found that the best rules screened by both methods come from the MFI–RSI and MSV categories, both using the volume information. Although our empirical results tend to deny the profitability of our 8061 rules in many of the ten futures markets, it might be overreaching to imply the deficiency of technical analysis in the futures markets. As technical analysis continues to develop rapidly, new trading rules and strategies employed by futures market practitioners might have been much more complicated than those used in this study. It is worthwhile to reexamine this issue when more techniques are revealed to the public. Moreover, the SPA test used to control for the data-snooping bias can be improved to examine not only the most-profitable rule but also all others which significantly outperform the benchmark. Testing the performance persistence of all outperformers in the out-of-sample period will give a more general picture of the profitability of technical analysis in the futures markets. Two recently developed statistical techniques, the stepwise reality check (SRC) [20] and the stepwise test for superior predictive ability (SSPA test) [15], might contribute to this issue.