Probabilistic Sharpe Ratio Calculator

Why the regular Sharpe ratio lies, and what to use instead

The Sharpe ratio is the most-cited number in quant finance, and it is also one of the most misleading. The standard formula (mean excess return divided by standard deviation) gives you a single point estimate that says nothing about how confident you should be in it. Two things break it: small sample sizes (the law of large numbers has not kicked in) and non-normal returns (skew and kurtosis distort the assumption that volatility captures all the risk). The probabilistic Sharpe ratio (PSR), introduced by Marcos López de Prado in 2012, addresses both.

The intuition behind PSR

PSR asks: "given my observed Sharpe, my sample size, and the higher moments of my returns, what is the probability that my TRUE long-run Sharpe exceeds some benchmark?" The benchmark is usually 0 (i.e. "is there any edge at all?") but you can also test against the S&P 500's long-run Sharpe of ~0.4 to ask "is my strategy better than passive equity exposure?".

The output is a probability between 0 and 1. By convention, PSR > 0.95 is the threshold for "publication-grade significance", equivalent to the standard 5% statistical significance test in regression analysis.

The formula

PSR is the cumulative normal distribution evaluated at a z-score that adjusts the raw Sharpe for sample size, skewness, and excess kurtosis:

z = (Sharpe − Benchmark Sharpe) × √(N − 1) / √(1 − γ₃·Sharpe + ((γ₄ − 1)/4)·Sharpe²)

Where γ₃ is skewness and γ₄ is kurtosis (not excess kurtosis). Then PSR = Φ(z) where Φ is the standard normal CDF. The denominator inflates when skew is negative or kurtosis is high, which lowers the z-score and therefore lowers PSR — exactly the desired behavior.

What MTRL tells you

The minimum track record length (MTRL) is one of the most useful outputs of PSR analysis. It tells you the smallest number of observations needed for the strategy's observed Sharpe to plausibly clear the 95% significance threshold. If MTRL is 800 daily returns and you only have 200, you simply do not have enough data to claim edge — no matter how impressive the point Sharpe looks. Strategies with low Sharpes or fat-tailed return distributions need much longer track records to prove themselves.

Why this matters for capital allocation

Imagine two strategies you are considering for a $10M allocation:

• Strategy A: Sharpe 1.5 over 6 months (125 observations), with negative skew (-1.5) and high kurtosis (8). PSR ≈ 0.75. Looks great by Sharpe alone, but PSR warns you it is mostly small wins punctuated by occasional large losses — and you haven't seen the next loss yet.
• Strategy B: Sharpe 0.9 over 5 years (1,250 observations), with mild positive skew (0.3) and near-normal kurtosis. PSR ≈ 0.99. Lower point Sharpe but dramatically more reliable evidence of true edge.

Sharpe alone says A is better. PSR correctly identifies B as the safer allocation. This is exactly the kind of decision PSR was designed to inform.

Strategies that get penalized by PSR (deservedly)

• Selling out-of-the-money options — heavy negative skew. Looks great until the tail event hits.
• Carry trades — "picking up nickels in front of a steamroller", high Sharpe with severe negative skew.
• Short-vol strategies — same pattern: many small wins, occasional catastrophic losses.
• Recently launched strategies — small N alone hammers the PSR even with normal returns.

Strategies PSR rewards

• Trend-following — positive skew (rare big wins, frequent small losses) actually inflates PSR above raw Sharpe.
• Long-vol strategies — same favorable skew profile.
• Long-track-record strategies — large N narrows the confidence interval around the Sharpe.

Related calculators

For the underlying point-estimate Sharpe with a 95% confidence interval, see the Sharpe ratio calculator. For other return-distribution diagnostics, see the Hurst exponent calculator and the value-at-risk calculator. For position sizing once you have established edge, see the Kelly criterion calculator.