Backtest Overfitting Score: PBO & Deflated Sharpe

PBO is the probability that the strategy with the best in-sample Sharpe ranks below median out-of-sample. Bailey, Borwein, Lopez de Prado, and Zhu (2017) introduced the metric. A PBO above 0.5 means the in-sample winner is more likely to underperform than outperform in production — i.e., the backtest is more curve-fit than predictive.

Combinatorially Symmetric Cross-Validation (CSCV): split the returns matrix into S equal subsets and form in-sample / out-of-sample partitions. The number of partitions grows fast, so the tool samples up to 500 of them, measures how often the in-sample top performer falls below the out-of-sample median, and divides by the number sampled. It reports the resulting PBO, not the per-partition ranks.

Lopez de Prado's adjustment to standard Sharpe that accounts for skew, kurtosis, and the number of trials run. A DSR above 0.95 (95% confidence) is the rough threshold for 'this Sharpe is unlikely to be from random search'. The tool reports both raw and deflated.

At least 16 for stable estimates; ideally 32+. The tool still computes PBO with fewer, but it does not emit a low-count warning — so treat a small-variant estimate as noisy yourself. If you only ran 2-3 variants, your PBO is more noise than signal.

No. PBO measures relative overfitting across the variants you tested, not absolute predictive power. A strategy can have low PBO (your best variant is genuinely better than the average variant) and still lose money live if all variants are weak. Combine PBO with deflated Sharpe and out-of-sample equity-curve inspection.