How much does in-sample Sharpe ratio overstate out-of-sample?

Bailey et al. (2015) show that selecting the best strategy from N independent ones adds sqrt(2*log(N)/T) bias to the Sharpe estimate. Searching 100 strategies adds ~1.5*sqrt(2*log(100)/T) bias.

What is the haircut Sharpe ratio?

Haircut Sharpe = IS_Sharpe * (1 - bias_factor). Bailey and Lopez de Prado suggest a 50% haircut for strategies not validated out-of-sample, reflecting that published results are typically the best of many tested.

Backtest Overfitting Estimated Sharpe Bias Calculator

Estimate the bias in Sharpe ratio due to backtest overfitting when searching over N strategies. Expected maximum Sharpe = E[max of N independent half-normal] grows with sqrt(2*log(N)).

Principal ($)

Annual Rate (%)

Years

Compounding / Year

$ Finance Disclaimer: These calculators are for educational and illustrative purposes only. Results do not constitute financial advice. Consult a licensed financial advisor before making investment or borrowing decisions.

How to Use

Enter the values for the scenario you are assessing. Estimate the bias in Sharpe ratio due to backtest overfitting when searching over N strategies. Expected maximum Sharpe = E[max of N independent half-normal] grows with sqrt(2*log(N)). Use the backtest sharpe bias result to inform your calculation.

Related Calculators

Probabilistic Sharpe

Finance Ultra-depth

Deflated Sharpe

Finance Ultra-depth

False Discovery Rate

Finance Ultra-depth

Backtest Overfitting Estimated Sharpe Bias Calculator

How to Use

Related Calculators

Frequently Asked Questions

How much does in-sample Sharpe ratio overstate out-of-sample?

What is the haircut Sharpe ratio?