A reality check on technical trading rule profits in the u,s, futures markets, страница 8

Table I shows that over 1985–1994 the best technical trading rules identified under the mean net return criterion generate positive annual mean net returns for each of the 17 markets. The mean net returns range from a low of 1.06% for treasury-bills to a high of 20.18% for crude oil and they are statistically significant for all but corn and soybeans at the 10% level when the effect of data snooping is ignored. Even at the 5% significance level, White’s nominal p-values indicate the statistical significance holds for 12 of the 17 markets. Hansen’s nominal p-values provide similar results, showing statistical significance for 13 markets at the 10% level. In contrast, White’s Reality Check bootstrap p-values that account for data snooping biases indicate that technical trading rules generate statistically significant net returns only for 2 of the 17 markets: the Eurodollar (p-value of 0.03) and the yen (p-value of 0.10). For the rest of the markets, the null hypothesis that the mean net return of the best

technical trading rule is not greater than that of the benchmark (in the present case, mean zero profit) cannot be rejected at the 10% significance level, although financial futures have much lower p-values than commodity futures. These results are consistent with the findings of Silber (1994), who documented that moving average rules generated substantial net profits for several major currency futures (e.g. the mark, yen, and Swiss franc) and interest rate futures (e.g. Eurodollar) but negative profits for commodity futures (e.g. silver and gold) over 1980–1991. Hansen’s Consistent SPA p-values also indicate that the best rule generates a statistically significant return only for the Eurodollar (p-value of 0.02). Even considering his Lower SPA p-values that remove the effects of poor-performing trading rules does not alter the result. These results suggest that the conventional statistical test, which does not account for the dependence in performance across technical trading rules, may mislead researchers in interpreting the profitability of technical trading rules. It is interesting to note that Hansen’s test does not necessarily produce lower p-values than those of White’s test. Hansen’s Consistent SPA p-values are equal to or higher than White’s Reality Check p-values for 7 of the 17 markets.

Out-of-sample performance of the best rules chosen over 1985–1994 is disappointing. During the out-of-sample period of 1995–2004, as shown in Table I, the best rules generate positive mean net returns for six markets, with statistically significant returns only for silver and the Eurodollar. For the rest of the 11 markets annual mean net returns are negative. Moreover, the size of positive returns is substantially reduced compared with that of in-sample returns in most cases (e.g., 10.46–2.04% for the mark; 10.90–4.33% for the yen; 1.76–0.53% for the Eurodollar; and 14.15–1.49% for the S&P 500 Index). To investigate whether the poor out-of-sample results were caused by high transaction costs of $100 per round turn, lower transaction costs of $50 are applied to the same best rules.^[5] Such lower transaction costs may be possible because commissions through discount brokers are around $12.50 per round turn (Lukac & Brorsen, 1990; Lukac et al. 1988), and even lower for high volume trades or electronic trades introduced in the early 1990s. Results (not presented) show that annual mean net returns are still negative for 9 of 17 markets and are statistically significant only for three markets (silver, the yen, and the Eurodollar), although they increase by 0.05–6.15% across all the markets. Technical trading rules, therefore, generally fail to produce economically and statistically significant returns in out-of-sample tests.

Table II shows in-sample performance of the best technical trading rules over 1995–2004. During this sample period, annual mean net returns of the