Deflated Sharpe Ratio
The Deflated Sharpe Ratio (DSR) is a performance statistic introduced by David Bailey and Marcos López de Prado (2014) that corrects the conventional Sharpe ratio for two leading sources of inflation: selection bias under multiple testing, and non-Normally distributed (skewed, fat-tailed) returns. It builds on the Probabilistic Sharpe Ratio, which gives the probability that a true Sharpe exceeds a benchmark while accounting for short track records and return non-normality.
DSR’s key move is that the benchmark Sharpe SR* is no longer chosen by the researcher but estimated from the variance across the strategies tried and the number of independent trials N — because “given a set of SR estimates, its expected maximum is greater than zero, even if the true SR is zero.” An observed Sharpe must beat that selection-bias-inflated threshold to count as a real finding. In this vault the DSR is the quantitative expression of the grading rule: a reported Sharpe from a Markov-model backtest that does not disclose how many configurations were tried cannot be deflated, and is therefore weak evidence by default. It is the companion statistic to Combinatorial Purged Cross-Validation, which supplies the distribution of Sharpe ratios DSR consumes.
Pseudo-Mathematics and Financial Charlatanism [defines] Deflated Sharpe Ratio Deflated Sharpe Ratio [opposes] Data-Snooping Bias Deflated Sharpe Ratio [supports] Out-of-Sample Backtesting
Connections
- Pseudo-Mathematics and Financial Charlatanism — proposes_model, source: https://ssrn.com/abstract=2460551
- Marcos López de Prado — proposes_model, source: https://ssrn.com/abstract=2460551
- Data-Snooping Bias — contradicts, source: https://ssrn.com/abstract=2460551
- Out-of-Sample Backtesting — relates, source: https://www.garp.org/hubfs/Whitepapers/a1Z1W0000054x6lUAA.pdf