What it shows: the more strategies you try on the same data, the higher the best backtested Sharpe ratio climbs — purely from luck. Each simulated strategy's returns are drawn from a zero-mean normal distribution, so there is no real edge anywhere. The benchmark line is the expected maximum Sharpe under pure chance (Bailey & López de Prado, 2014). It's the intuition behind the backtesting guide and signal validation: a backtest Sharpe is only meaningful once you discount it for how many things you tested.
All returns are synthetic random noise — nothing here is a performance figure or investment advice. Related tools: Information Coefficient Calculator · Signal Decay Calculator · Cointegration & Pairs Trading Simulator · Signal Combination Simulator · Correlation Heatmap · VPIN & the Volume Clock · Signal Skill Explorer.
Run the experiment
Every strategy below has zero real edge — its "returns" are pure random noise. We build many of them, compute each one's in-sample annualized Sharpe ratio, and keep the best — exactly what happens when you grid-search parameters or test many ideas on one dataset.
60 return observations per strategy
Frequently asked
- How can pure noise produce a Sharpe ratio of 1.5?
- When you try many strategies and keep the best, the maximum in-sample Sharpe is inflated by multiple testing even when every strategy has zero true edge. The expected maximum grows with the number of trials — the simulator shows the Bailey and Lopez de Prado expected-maximum-Sharpe benchmark next to the "winner" you found.
- How do I use this to judge a real backtest?
- Set the number of strategies to roughly how many configurations you actually tried (including the informal ones) and the length to your backtest. If your real Sharpe does not clearly exceed the expected-maximum-by-chance line, it may be a product of the search rather than a real edge — discount it, for example with the Deflated Sharpe Ratio.
- Why does the best Sharpe change each run?
- Each run draws a fresh set of random zero-edge strategies, so the maximum varies. But its expectation tracks the benchmark, which is the point: chance alone reliably manufactures impressive-looking winners.
- What is the practical takeaway?
- The more things you try, the higher a Sharpe you need before believing it. Track your trial count, prefer out-of-sample and walk-forward validation, and deflate for multiple testing.
- Is this real strategy data?
- No. All returns are synthetic Gaussian noise with zero true edge. It is a multiple-testing demonstration, not investment advice.