correction for risk. If transaction costs increase, then the number of data series for which a significantly positive estimate of α is found declines. This can especially be observed in the results for the US stock market in Chapter 3 for which the fraction of data series for which a significantly positive estimate of α is found declines to one quarter if 1% transaction costs are implemented. However, in the case of 1% transaction costs, for approximately half of the Dutch stock market data in Chapter 4 and for approximately half of the stock market indices in Chapter 5, the estimate of α is still significantly positive. If the in-sample CAPM estimation results are compared with the out-of-sample CAPM estimation results, then the results in favour of technical trading of the latter tests are obviously worse than the results of the former tests. However, if transaction costs are zero, then in each chapter a group of data series can be found for which technical trading shows significant out-of-sample forecasting power, after correction for risk. As transaction costs increase, this group becomes smaller and smaller.
White's (2000) RC and Hansen's (2001) SPA-test are utilized to correct for data snooping. If little costs are implemented, then for the US stock market data in Chapter 3, the RC does not reject the null of no superior forecasting ability of the best selected technical trading rule over the buy-and-hold benchmark for all data series for both selection criteria. For the Dutch stock market data in Chapter 4 the same conclusion can be made, although the results in favour of technical trading are stronger, if the Sharpe ratio criterion is used. For a group of stock market indices in Chapter 5, in the case of zero transaction costs, it is found that the null hypothesis of no superior forecasting ability is rejected, especially if the Sharpe ratio criterion is used. However, if transaction costs increase to 1%, then for almost all data series the null hypothesis is not rejected anymore. The SPA-test corrects for the inclusion of poor and irrelevant strategies. Differences between the RC and SPA-test can especially be seen in Chapters 4 and 5, if the Sharpe ratio selection criterion is used. Then, for both the Dutch stock market data and the local main stock market indices, if 1% transaction costs are implemented, for more than one quarter of the data series the null hypothesis of no superior forecasting ability is rejected. Thus the biased RC leads in numerous cases to the wrong inferences.
If no transaction costs are implemented, then technical trading shows economically and statistically significant forecasting power for a group of data series, in all three chapters. In that case, generally, the results of the Sharpe ratio selection criterion are slightly better than the results of the mean return selection criterion. However, if transaction costs increase, then in Chapters 4 and 5 the Sharpe ratio selection criterion performs better in selecting the best technical trading strategy. If the Sharpe ratio criterion is used in selecting the best strategy, then for transaction costs up to 0.25%, technical trading