The Deflated Sharpe Ratio
Lopez de Prado introduced the Deflated Sharpe Ratio in his research to address a fundamental problem: when you test many strategy configurations, the best-performing one will have an inflated Sharpe ratio simply because you selected it from a large pool. If you test 100 parameter combinations and pick the best one, that best one benefits from selection bias even if every combination has zero true edge.
The correction is conceptually simple: adjust the reported Sharpe ratio downward based on the number of configurations tested, the skewness and kurtosis of returns, and the length of the backtest. The result is almost always a substantially lower number than the raw backtested Sharpe.
Other Sources of Sharpe Inflation
Beyond selection bias, several other factors systematically inflate backtested Sharpe ratios. Survivorship bias in the asset universe makes every long strategy look better. Unrealistic execution assumptions (zero spread, zero slippage, instant fills) eliminate costs that reduce live returns. Look-ahead bias, even subtle forms, can creep into data preprocessing. And the absence of capacity constraints means the backtest assumes you can trade any size without market impact.
Each of these factors adds a small positive bias. Together, they can easily account for a full point of Sharpe ratio. A raw backtested Sharpe of 2.5 might become 1.5 after corrections, and live performance of 0.8-1.2 would not be unusual.
Practical Guidelines
Harvey, Liu, and Zhu's research suggested that a newly discovered factor needs a t-ratio above 3.0 (rather than the traditional 1.96) to be credible, because of cumulative data-mining across decades of research. The equivalent rule of thumb for strategy backtests: expect live Sharpe to be 40-60% of backtested Sharpe, and be pleasantly surprised if it is higher.
This does not mean backtesting is useless. It means you should treat backtested performance as an upper bound, not a forecast. A strategy whose backtested Sharpe is below 1.5 is unlikely to be profitable after real-world degradation. A strategy above 2.0 backtested has a reasonable chance of remaining profitable in live trading, though at a lower Sharpe than the backtest suggests.