Predictive power is the degree to which a model, indicator, or strategy’s outputs reliably correlate with subsequent market outcomes — quantifying how useful the signal is for forecasting future prices, returns, or risk events. Higher predictive power means a stronger, more consistent relationship between the signal and future outcomes. The concept is central to evaluating whether an investment approach has a genuine edge or merely appears to through data mining and the benefit of hindsight.

Measuring Predictive Power

Predictive power in finance is measured through multiple quantitative frameworks. Information coefficient (IC) — the correlation between predicted and actual returns — is widely used in quantitative investing. The t-statistic on a strategy’s alpha measures whether excess returns are statistically significant above what random chance would produce. R-squared measures the fraction of return variance explained by the predictive model. And economic significance — whether the predictive relationship is large enough to generate returns exceeding costs and risks — is ultimately more important than statistical significance alone.

In-Sample vs. Out-of-Sample Predictive Power

The crucial distinction in evaluating predictive power is between in-sample performance (measured on the same data used to develop the model) and out-of-sample performance (measured on data the model has never seen). In-sample predictive power is nearly meaningless — with sufficient optimization, virtually any model can achieve high R-squared on historical data. Out-of-sample predictive power — and particularly walk-forward performance in live trading — is the true test of whether the model has captured genuine, persistent market relationships or merely overfitted to historical noise.

---