Out-of-Sample Validation

Everything before this node can be tuned until the backtest shines, which is why out-of-sample validation is the final gate. Walk-forward testing rolls the evaluation window through time, re-estimating on each trailing window and scoring on the unseen months that follow, mimicking live decisions without peeking. Even then multiple testing corrupts results: try enough variants and the best backtest is mostly luck, which is what the deflated Sharpe ratio of Bailey and López de Prado corrects for, discounting measured performance by the number of strategies examined and the autocorrelation of returns. Only a strategy that survives this gauntlet with costs included deserves capital.