Skip to main content
aifinhub
Backtesting & Validation Guide

How to Compute a Deflated Sharpe Ratio

A raw Sharpe ratio tells you how good a backtest looks. The deflated Sharpe ratio tells you whether that number means anything once you account for how many strategies you tried and how non-normal the returns are. It is the single most useful guard against fooling yourself with a backtest, and it is computable from five inputs you should already have. Each of the five inputs is covered below, along with how to interpret the result.

By AI Fin Hub Research · AI Fin Hub Team

On This Page

Before You Start

Set up the inputs that make the next steps easier

An observed (annualized or per-period) Sharpe ratio from a backtest with costs already subtracted.
The number of return observations the Sharpe was estimated from.
The skewness and excess kurtosis of the return series, or the raw returns to compute them.
An honest count of how many strategy configurations were tried before this one was selected.

Guide Steps

Move through it in order

Each step focuses on one decision so you can keep momentum without losing the thread.

  1. 1

    Gather the five inputs

    The deflated Sharpe ratio needs five numbers: the observed Sharpe, the sample length in periods, the return skewness, the return excess kurtosis, and the trial count. The first four describe the backtest you ran; the fifth describes the search that produced it. Missing the trial count is the most common failure, and without it the deflation cannot be done honestly. Assemble all five before computing anything.

    Use the Sharpe at the same frequency as your return observations. Mixing an annualized Sharpe with a daily sample length corrupts the standard error.

    Use The ToolCalculators

    Deflated Sharpe Ratio Calculator

    Bailey & López de Prado deflated Sharpe — corrects observed Sharpe for selection bias across K trials. Reports deflated Sharpe, PSR (probability of skill).

    ToolOpen ->
  2. 2

    Count the trials, including informal ones

    The trial count is the number of distinct strategies you evaluated against the data: every grid point, every entry rule, every universe filter, every threshold you nudged after seeing a result. Informal tuning counts. If you changed a parameter because the backtest looked better, that is a trial. The deflated Sharpe uses the trial count to estimate the maximum Sharpe you would expect from pure luck across that many attempts, which becomes the bar your real Sharpe must clear.

    When unsure, overcount rather than undercount. A conservative trial count makes the deflation stricter, which is the safe direction.

    Use The ToolCalculators

    Backtest Overfitting Score

    Upload a backtest trade log and compute Probability of Backtest Overfitting (PBO), Deflated Sharpe Ratio, and the odds your edge survives live trading.

    ToolOpen ->
  3. 3

    Compute the expected maximum Sharpe from luck

    Given the trial count, the formula estimates the expected maximum Sharpe ratio you would see by selecting the best of that many independent random strategies. This benchmark grows with the number of trials: search a thousand variants and the best one will have a respectable Sharpe by chance alone. The deflated Sharpe measures how far your observed Sharpe sits above this luck benchmark, not above zero.

    The luck benchmark rises fast at first and then slowly. Going from 1 to 100 trials matters far more than from 100 to 200, which is why the first few sweeps are the most damaging.

  4. 4

    Adjust for skew and kurtosis

    The standard error of a Sharpe ratio is not the textbook value when returns are skewed or fat-tailed, which financial returns almost always are. Negative skew and high kurtosis inflate the true uncertainty of the Sharpe estimate. The deflated Sharpe plugs the third and fourth moments into the standard error so the test reflects the actual shape of your returns rather than assuming a normal distribution that does not hold.

    Strategies that sell tail risk often show high Sharpe with strong negative skew. The skew adjustment is exactly what catches these flattering-but-fragile backtests.

  5. 5

    Read the probability and decide

    The output is a probability between zero and one: the chance the true Sharpe is positive given the search you performed. The conventional threshold to consider a strategy worth capital is 0.95. A backtest with a raw Sharpe of 1.5 can deflate well below 0.95 once a few hundred trials and fat tails are priced in. If you clear the bar, the edge is plausibly real; if you do not, the result is most likely a selection artifact.

    Treat a marginal value near 0.95 as a fail, not a pass. The cheapest way to raise it is more data or fewer trials, never a higher raw Sharpe found by searching harder.

Common Mistakes

The misses that undo good inputs

1

Computing the deflated Sharpe with a trial count of one

If you searched many variants but record only one trial, the deflation does nothing and the number is as misleading as the raw Sharpe. The trial count is the whole point of the adjustment.

2

Assuming normal returns and skipping the moment adjustment

Financial returns are skewed and fat-tailed. Using the textbook Sharpe standard error understates uncertainty and inflates the deflated value, defeating the purpose of the test.

3

Treating 0.95 as a soft target to negotiate down

The threshold exists to keep selection luck out of your capital allocation. Lowering it for a favored strategy reintroduces exactly the bias the test is meant to remove.

Try These Tools

Run the numbers next

FAQ

Questions people ask next

The short answers readers usually want after the first pass.

The probabilistic Sharpe ratio gives the probability that a strategy's true Sharpe exceeds a benchmark, accounting for sample length, skew, and kurtosis, but for a single strategy. The deflated Sharpe ratio extends it by also accounting for the number of trials you ran, setting the benchmark to the expected maximum Sharpe from that many attempts. Deflated is the version to use whenever you selected a strategy from a search.

Sources & References

Related Content

Keep the topic connected

Planning estimates only — not financial, tax, or investment advice.