Free tool · runs entirely in your browser · nothing uploaded

Deflated Sharpe Ratio Calculator

Is your Sharpe ratio real — or just the luckiest of N tries? This computes the Probabilistic Sharpe Ratio and the Deflated Sharpe Ratio from the Bailey–López de Prado framework, correcting your backtest for track length, non-normal returns, and the size of the search that produced it — plus the approximate minimum backtest length your trial count demands.

Your backtest

Probabilistic Sharpe RatioP(true Sharpe > 0), given track length, skew & kurtosis — before any correction for the search
Expected max Sharpe of N unskilled trialsthe bar your search width sets — beat this or you may just be the luckiest
annualized
Deflated Sharpe RatioP(true Sharpe > expected max of your N trials). Significance bar: 0.95
Minimum backtest lengthapprox. years needed before a Sharpe like yours outruns the luck of N trials
years (approx.)

The deflation curve

Annualized Sharpe required for DSR = 0.95 as the number of trials grows, at your track length, skew and kurtosis. The dot is you.

This calculator has a back door. So does every one like it.

Every number above is only as honest as N — and N is self-reported, here and everywhere. Run a thousand backtests, publish the best, type N=10, and the correction for luck quietly evaporates. A thousand honest backtests with only the winner published is a lie that contains no lie.

MIZAN closes that door: every trial is committed to a Merkle tree before evaluation, N becomes the tree's leaf count — not a number anyone types — and the deflation is recomputed inside a zero-knowledge proof, without revealing the strategies. The engine's first published verdict was against its own maker's flagship: DSR 0.68, not significant, published as-is.

Read the paper → How it works → Prove yours →

The mathematics, exactly

PSR (Bailey & López de Prado, 2012): the probability that the true Sharpe exceeds a benchmark SR*, using the per-period Sharpe, n observations, and the skewness γ₃ and kurtosis γ₄ of returns: PSR = Φ[ (SR − SR*)·√(n−1) / √(1 − γ₃·SR + ((γ₄−1)/4)·SR²) ].

DSR (Bailey & López de Prado, 2014): the PSR evaluated at SR* = the expected maximum Sharpe of N independent unskilled trials, E[max] ≈ σ(SR)·[ (1−γ)·Φ⁻¹(1−1/N) + γ·Φ⁻¹(1−1/(N·e)) ], with γ ≈ 0.5772 (Euler–Mascheroni) and σ(SR) the standard deviation of Sharpe across the trials.

Minimum backtest length (Bailey, Borwein, López de Prado & Zhu, 2014): the approximate track length in years below which the expected maximum Sharpe of N unskilled trials exceeds your annualized Sharpe — reported here from the same E[max] expression; the paper's simple upper bound is 2·ln(N)/SR².

Bailey, D.H. & López de Prado, M., "The Sharpe Ratio Efficient Frontier," Journal of Risk 15:2 (2012) · Bailey, D.H. & López de Prado, M., "The Deflated Sharpe Ratio," Journal of Portfolio Management 40:5 (2014) · Bailey, Borwein, López de Prado & Zhu, "Pseudo-Mathematics and Financial Charlatanism," Notices of the AMS 61:5 (2014). Full lineage: the twenty authors.

Attribution, not endorsement: the cited authors have not reviewed or endorsed this tool. Educational use; not investment advice.

Questions this page answers

What is a good Sharpe ratio?

The uncomfortable truth: there is no threshold that means anything by itself. A Sharpe of 2.0 from one pre-registered trial on ten years of data can be highly significant; the same 2.0 selected from a thousand tries on three years may be pure luck. The only honest answer is the deflated one — which is why this page exists.

Why does the number of trials matter so much?

Because the expected maximum of N random strategies grows with ln(N). Try enough things and one of them will clear any bar by chance. The deflation subtracts exactly that expected luck.

Can I trust a deflated Sharpe someone reports?

Only as far as you trust their N — which is the open wound in the entire framework, and the thing that can be made structurally impossible to misreport. That is the engine's job, not a calculator's.