Identifying the best technical trading rule: a .632 bootstrap approach.

In this paper, we estimate the out-of-sample predictive ability of a set of trading rules. Usually, this ability is estimated using a rolling-window sample-splitting scheme, true out-of-sample data being rarely available. We argue that this method makes a poor use of the available information and creates data mining possibilities. Instead, we introduce an alternative bootstrap approach, based on the .632 bootstrap principle. This method enables to build in-sample and out-of-sample bootstrap data sets that do not overlap and exhibit the same time dependencies. We illustrate our methodology on IBM and Microsoft daily stock prices, where we compare 11 trading rules specifications. For the data sets considered, two different filter rule specifications have the highest out-of-sample mean excess returns. However, all tested rules cannot beat a simple buy-and-hold strategy when trading at a daily frequency.