Haircut Sharpe Ratio
A deflated version of a backtest's Sharpe ratio that accounts for how many strategy variants were tried before landing on the reported one — the more trials, the bigger the haircut needed to keep the result honest.
Prerequisites: Effective Number of Independent Trials
A single backtested Sharpe ratio, reported on its own, hides a crucial piece of context: how many strategies did the researcher try before arriving at this one? A Sharpe ratio of 1.5 from the only strategy ever tested is a meaningfully different result from a Sharpe ratio of 1.5 that happened to be the best out of 200 variants tried. The second number is exactly what you'd expect to find by chance somewhere in a large enough search, even with zero genuine skill. The haircut Sharpe ratio takes the reported number and shaves it down to account for the size of the search that produced it, giving a more honest estimate of what to expect going forward.
Where the haircut comes from
The core idea borrows from multiple-testing correction in statistics: if you run many trials and only report the best, the expected best-of-many result is inflated purely by selection, even under the null hypothesis of no real effect. The haircut Sharpe ratio (closely related to the deflated Sharpe ratio) formalizes this by asking, given the number of effective independent trials attempted and the length of the backtest, what Sharpe ratio would appear "significant" purely by chance — and then reducing the reported Sharpe to reflect that a large search makes an impressive-looking number less surprising. More trials, or a shorter backtest history (which makes any one Sharpe estimate noisier), both push the required haircut larger.
Worked example: 1.5 becomes 0.6
A researcher reports a backtested Sharpe ratio of 1.5, achieved after testing roughly 80 effectively independent strategy variants over a five-year backtest. Using a standard haircut methodology, the number of trials and the backtest length combine to produce a substantially reduced, "haircut" Sharpe ratio — in a case like this, the deflation can easily take a headline 1.5 down to somewhere around 0.5–0.7 once the search size is accounted for. That haircut number, not the original 1.5, is the more realistic estimate of what the strategy might deliver going forward, because it prices in how much of the 1.5 was likely a product of trying many things and reporting the best one, rather than a persistent edge.
What this means in practice
Reporting a haircut Sharpe ratio alongside the raw one is increasingly standard practice among allocators evaluating quant strategies, precisely because raw backtest Sharpe ratios are so easily inflated by an unreported search. The catch is that the haircut can only be computed honestly if the researcher accurately discloses (or the search itself is logged) how many variants were actually tried — a haircut computed on a dishonestly small trial count gives false comfort. This is why some research groups now log every backtest run automatically, so the trial count used in any later haircut calculation isn't left to memory or self-report.
A backtested Sharpe ratio should be discounted based on how many strategy variants were searched to find it. The haircut Sharpe ratio formalizes this discount, turning an impressive but search-inflated number into a more honest estimate of forward performance.
The haircut calculation is only as trustworthy as the trial count fed into it. Understating how many variants were actually tried — even unintentionally, by forgetting early exploratory runs — produces a haircut that still looks reassuring but doesn't correct for the real size of the search.
Further reading
- Bailey and López de Prado, 'The Deflated Sharpe Ratio'