Methodology

The tests

Probabilistic Sharpe ratio and minimum track record length, from Bailey and López de Prado, "The Sharpe Ratio Efficient Frontier" (2012). They account for the length of the record, skewness and fat tails.

Deflated Sharpe ratio, from Bailey and López de Prado, "The Deflated Sharpe Ratio" (2014). It compares the Sharpe ratio with the best one that the same number of skill-less variants would produce by chance. Our implementation reproduces the paper's worked example (0.9004).

Probability of backtest overfitting, from Bailey, Borwein, López de Prado and Zhu (2017), estimated with combinatorially symmetric cross-validation over all half-and-half splits of 16 time blocks (12,870 splits).

Stress test: 5,000 five-year paths built by block bootstrap from the strategy's own returns.

Verdicts

Likely luck: the Sharpe ratio is about what the best of the variants tried would show by chance (deflated confidence under 75%), or the in-sample winner usually loses money out of sample.

Inconclusive: confidence between 75% and 95%, a record shorter than the minimum needed, or meaningful overfitting. When the number of variants tried is unknown, a strategy must pass as if 100 had been tried.

Edge likely real: at least 95% deflated confidence and a record longer than the minimum needed.

What it can't detect

Look-ahead bias, survivorship bias, fabricated data, or costs missing from the returns you upload. The audit describes the statistics of the returns supplied.

Independence and privacy

Verdicts are never for sale, and no customer can pay to change a result. Uploaded files are processed in memory and deleted once the report is made. We never store, sell or share strategies.