Detailed test period performance:
Each rule is better than random guessing but worse than on the training data, as expected: Rule 0: 42 firings, 55% wins, zscore 1.1 Rule 1: 57 firings, 56% wins, zscore 1.9 Rule 2: 63 firings, 60% wins, zscore 1.0 Unexpectedly, profit increases on
Since the strategy is not optimized for predictive accuracy, we expect overfitting to be less on accuracy than on mean return. It is not clear whether this hypothesis is verified.