Probabilistic Sharpe Ratio

Definition

The Probabilistic Sharpe Ratio (PSR; Bailey-López de Prado 2012) is an uncertainty-adjusted measure of investment skill: the probability that a strategy's true Sharpe ratio exceeds a chosen benchmark $SR^*$, given an estimated Sharpe ratio $\hat{SR}$ computed from a finite, possibly non-Normal return series. It converts the Sharpe ratio from a point estimate into a confidence statement that accounts for track-record length, skewness, and kurtosis.

Key Ideas

• The PSR statistic. $\widehat{PSR}(SR^*) = \Phi\!\left[\dfrac{(\hat{SR}-SR^*)\sqrt{n-1}}{\sqrt{1-\gamma_3\hat{SR}+\frac{\gamma_4-1}{4}\hat{SR}^2}}\right]$, where $n$ is the number of returns, $\gamma_3$ skewness, $\gamma_4$ kurtosis, and $\Phi$ the standard-normal CDF. The denominator is the (non-Normal) standard error of $\hat{SR}$.
• Minimum Track Record Length (MinTRL). Inverting the PSR at a target confidence gives the record length required to declare $\hat{SR}$ significantly above $SR^*$: $\text{MinTRL}=1+\left[1-\gamma_3\hat{SR}+\frac{\gamma_4-1}{4}\hat{SR}^2\right]\left(\frac{Z_\alpha}{\hat{SR}-SR^*}\right)^2$.
• Skew–kurtosis penalty. Negative skewness and positive excess kurtosis increase the SR standard error, so such strategies need longer records for the same confidence — a formal statement of why "picking up pennies in front of a steamroller" returns flatter the naive Sharpe ratio.
• Sampling frequency. Under IID returns, higher-frequency sampling raises $n$ and tightens the estimate; the benefit disappears once autocorrelation breaks IID.
• Sharpe Ratio Efficient Frontier (SEF). Portfolio optimisation that maximises confidence-adjusted performance under non-Normal, leveraged returns instead of the mean–variance Sharpe alone.
• Deflated Sharpe Ratio (DSR; Bailey-López de Prado 2014). Setting the benchmark $SR^*$ to the expected maximum Sharpe ratio attainable across $N$ independent trials turns the PSR into the Deflated Sharpe Ratio. The benchmark is estimated by extreme-value theory: with $N$ trials whose Sharpe ratios have cross-sectional variance $V[\hat{SR}_n]$, the expected maximum is $E[\max SR]\approx\sqrt{V[\hat{SR}_n]}\big[(1-\gamma)\Phi^{-1}(1-\tfrac1N)+\gamma\,\Phi^{-1}(1-\tfrac1{Ne})\big]$ (Gumbel/$\gamma$ = Euler-Mascheroni). DSR thus corrects for two leading sources of performance inflation at once — selection bias under multiple testing and non-Normal returns — separating genuine skill from statistical flukes. The companion Minimum Backtest Length (MinBTL) gives the shortest backtest for which a given number of trials would not, by chance alone, produce a spuriously high in-sample Sharpe ratio.

Why It Matters

• Honest performance claims. It supplies a standard error and a significance test for the Sharpe ratio, exposing track records that are too short or too fat-tailed to support their headline number.
• Foundation for anti-overfitting tools. PSR is the building block of the deflated Sharpe ratio and the probability of backtest overfitting — the finance analogue of the factor zoo's multiple-testing critique.
• Practical portfolio choice. The SEF integrates estimation uncertainty directly into allocation, rather than treating the Sharpe ratio as known.

Open Questions

• IID dependence. The variance formula assumes IID returns; serial correlation (smoothed/illiquid returns) inflates confidence and must be corrected.
• Moment estimation. Skewness and kurtosis are themselves noisily estimated in short samples, feeding estimation error back into the PSR.
• Benchmark choice. Results depend on the chosen $SR^*$; the multiple-testing correction requires knowing (or estimating) the effective number of independent trials.

Related

• Backtest Overfitting
• Factor Zoo
• Bailey-López de Prado (2012) The Sharpe Ratio Efficient Frontier
• Bailey-López de Prado (2014) The Deflated Sharpe Ratio
• David H. Bailey
• Marcos López de Prado