Stats Backtesting and evidence Deflated Sharpe

After 200 tests, a Sharpe of 0.75 is worth 0.32

What was measuredTests assumedWhat it becomesSource
Annual Sharpe 0.75, 20 years of monthly returns, single test p value 0.0008200p value 0.15, adjusted annual Sharpe 0.32, haircut about 60%Harvey and Liu
The least profitable of three well known strategies, earnings to price50Haircut almost 50%Harvey and Liu
The most profitable of the three, betting against beta50Haircut 7.9%Harvey and Liu
Any observed annual Sharpe ratio below 0.4their baselineHaircut usually more than 50%Harvey and Liu
Any observed annual Sharpe ratio above 1.0their baselineHaircut at most about 25%Harvey and Liu
Annualised Sharpe 2.5, 1,250 daily returns, skewness -3, kurtosis 10, trial Sharpe variance 0.5100Noise threshold 0.1132 daily, deflated Sharpe 0.90, fails the 0.95 barBailey and Lopez de Prado
Exactly the same backtest, fewer trials behind it46Deflated Sharpe 0.9505, clears the 0.95 barBailey and Lopez de Prado

Sources: Harvey and Liu, Backtesting, Journal of Portfolio Management, Fall 2015; Bailey and Lopez de Prado, The Deflated Sharpe Ratio, Journal of Portfolio Management 40(5), 2014. The 0.1132 daily threshold is about 1.79 annualised at 250 trading days a year.

A Sharpe ratio from the best of many backtests is not the same measurement as one from a single test. On twenty years of monthly returns, two hundred tests cut a Sharpe of 0.75 to 0.32. The discount depends on a number almost nobody records.

0.320.75 Sharpe, 200 tests
about 60%Haircut
0.90100 trials, deflated
0.950546 trials, deflated

Members only

The rest of this page is for members

Below this point there are 5 sections, 1 chart and 5 named sources, roughly 1250 words of it. Every figure carries the source it came from and the date the data is from.

You can keep browsing every statistic in the library for free. The intro and the summary are always open.