Probability of backtest overfitting (PBO)

PBO (Bailey, Borwein, López de Prado & Zhu, 2017) measures how likely it is that your selection process — not your strategy — produced the performance. Using combinatorially symmetric cross-validation (CSCV), the sample is split into many train/test partitions; in each, the configuration that wins in-sample is checked out-of-sample. PBO is the fraction of partitions where the in-sample winner falls below the out-of-sample median.

A PBO near 50% says your selection is a coin flip: whatever ranked configurations in the past carries no information about the future. Low single-digit PBO is what a robust selection process looks like. It is one of the few overfitting measures that evaluates the process rather than a single equity curve.