A reality check on technical trading rule profits in the u,s, futures markets, страница 2

A fairly blatant form of data snooping in the technical trading literature is an ex post and in-sample search for profitable trading rules. More subtle forms of data snooping occur when a set of data is repeatedly used to search for profitable “families” of trading systems, markets, in-sample estimation periods, out-of-sample periods, and trading model assumptions including performance criteria and transaction costs. In the presence of such data snooping, researchers may be misled because of exaggerated trading returns and significance levels of conventional statistical tests (Cooper & Gulen, 2006; Denton, 1985; Lo & MacKinlay, 1990).

To deal with data snooping problems, White (2000) proposed a rigorous statistical procedure, the Bootstrap Reality Check, that can directly quantify the effect of data snooping by evaluating the performance of the best model in the context of the full universe of models.^[1] In White’s approach, the best rule is identified by applying a performance statistic to the full set of trading rules, and then a desired p-value can be obtained from comparing the performance of the best trading rule with approximations to the asymptotic distribution of the performance statistic across all the trading rules. Hansen (2005) argues that White’s test may reduce rejection probabilities of the test under the null hypothesis by the inclusion of poor and irrelevant alternative models because it does not satisfy a relevant similarity condition that is necessary for a test to be unbiased. Hansen shows that his new test, the Superior Predictive Ability (SPA) test, can improve the power of hypothesis tests by adopting a studentized test statistic and a data-dependent null distribution.

Sullivan et al. (1999), and Sullivan, Timmermann, and White (2003) applied White’s test to the DJIA and S&P 500 futures index and found that Brock et al.’s (1992) successful findings were robust to data snooping biases, although they found that technical trading rules did not continue to generate economically and statistically significant returns for both data sets in the subsequent 10-year period, 1987–1996. Hsu and Kuan (2005) applied both White’s and Hansen’s tests to four main stock indexes, DJIA, S&P 500, NASDAQ Composite, and Russell 2000, over 1989–2002. Their in- and out-of-sample results indicated that technical trading rules were profitable in relatively young markets (NASDAQ Composite and Russell 2000) but not in mature markets (DJIA and S&P 500) after transaction costs. Using White’s test, Qi and Wu (2006) investigated seven foreign exchange rates over 1973–1998 and found substantial trading profits for five of the seven exchange rates even after adjustment for transaction costs and systematic risk.

This article examines whether technical trading rules have been profitable in the U.S. futures markets over 1985–2004 after explicitly accounting for the effect of data snooping and transaction costs. We measure the statistical significance of technical trading profits using both White’s Bootstrap Reality Check test and Hansen’s SPA test. We also expand the number of technical trading systems and parameters in Lukac et al. (1988), and construct a universe of technical trading rules with more than 9,000 rules drawn from 14 trading systems. This is a large universe of trading rules, and some of the trading systems were used by actual futures market traders before the beginning of the sample period (Lukac et al., 1988).

In addition to the 12 futures markets analyzed by Lukac et al. (1988), five highly traded futures contracts are investigated to ensure a more general test of the profitability of technical trading rules. The 17 markets represent each major group of futures contracts, i.e., grains, meats, softs, metals, energies, currencies, interest rates, and equity indices. Hence, this study improves upon previous studies of technical analysis in futures markets by incorporating more trading rules and markets and conducting superior statistical tests. Bootstrap simulation results show that there are virtually no significant technical trading profits in the U.S. futures markets during both in- and out-of-sample periods on a data-snooping-adjusted basis, presenting a striking contrast to the results of Sullivan et al. (1999) and Qi and Wu (2006) who found significant profits during the in-sample period using both unadjusted (standard) test statistics and data-snooping-adjusted test statistics.