In this paper, we provide a novel way to estimate the out-of-sample predictive ability of a trading rule. Usually, this ability is estimated using a sample-splitting scheme, true out-of-sample data being rarely available. We argue that this method makes poor use of the available data and creates...