What Is Probabilistic Sharpe ratio?

The Probabilistic Sharpe Ratio (PSR), introduced by Marcos López de Prado in 2012, computes the probability that a strategy’s true Sharpe ratio exceeds a given benchmark threshold, accounting for the uncertainty inherent in finite samples. A raw Sharpe of 1.5 means nothing if it was measured over 20 trades with heavy-tailed returns — PSR quantifies exactly how confident you should be that the observed Sharpe is not a statistical fluke. This makes it an essential diagnostic for anyone evaluating backtest results or comparing trading strategy candidates.

The key insight is that the standard error of the Sharpe ratio depends not only on the sample size but also on the skewness and kurtosis of returns. Strategies with negative skew (occasional large losses) or high kurtosis (fat tails) require much longer track records before their Sharpe estimates become reliable. PSR incorporates all four moments of the return distribution into a single probability statement. A related concept is the Minimum Track Record Length (MinTRL) — the smallest number of observations needed for the measured Sharpe to be statistically significant at a given confidence level. López de Prado showed that for typical hedge fund return distributions, MinTRL can easily exceed several years of monthly data.

Building on PSR, Bailey and López de Prado introduced the Deflated Sharpe Ratio (DSR), which further corrects for multiple-testing bias. When a researcher tries hundreds of parameter combinations and reports only the best one, the winning Sharpe is inflated by selection bias. DSR adjusts the PSR threshold upward based on the number of independent trials, providing a more honest assessment of whether a strategy genuinely outperforms or merely won a data-mining lottery. Together, PSR and DSR form a rigorous statistical framework that belongs in every quant’s overfitting detection toolkit.

Formula

PSR(SR*) = Φ( (SR̂ - SR*) / σ̂(SR̂) )

where σ̂(SR̂) = sqrt( (1 - γ₃·SR̂ + (γ₄-1)/4·SR̂²) / (N-1) )

SR̂ is the estimated Sharpe, SR* is the benchmark Sharpe, Φ is the standard normal CDF, γ₃ is skewness, γ₄ is kurtosis, and N is the sample size.

Pros

  • Provides a probability, not just a point estimate, making Sharpe comparisons statistically rigorous

  • Accounts for non-normality (skewness and kurtosis) that inflates naive Sharpe estimates

  • The Deflated Sharpe Ratio extension corrects for multiple-testing / data-mining bias

  • Defines Minimum Track Record Length — how long you must observe before trusting a Sharpe

Cons

  • Requires estimates of skewness and kurtosis, which are themselves noisy in small samples

  • Does not replace domain judgment — a high PSR on a fundamentally flawed strategy is still meaningless

  • The correction assumes independent trials, which may not hold when parameter sets are correlated

  • Adds analytical complexity that can be a barrier for practitioners unfamiliar with the framework

López de Prado — The Sharpe Ratio Efficient Frontier (SSRN).

Bailey & López de Prado — The Deflated Sharpe Ratio (SSRN).

QuantConnect — Probabilistic Sharpe Ratio.

Portfolio Optimizer — PSR and Minimum Track Record Length.

See also