Track record check
A backtest that looks excellent is easy to produce: try enough variants and the best one always does. Paste a history and see how much of it survives once the number of variants tried, the length of the sample and the shape of the returns are taken into account.
Measure a history
One column of returns or NAV per strategy, optionally with dates. Several columns are read as the variants of one search, which lets the probability of overfitting be computed as well.
Every parameter set, universe, start date or rule change you looked at counts. With several columns, each column is one variant and the count is taken from the data.
Computed in this browser. The data is never sent anywhere.
What each figure means
Four standard results from the literature on backtest overfitting, implemented once and tested against the papers' own worked examples.
Probabilistic Sharpe ratio
The probability that the true Sharpe ratio is above zero, given the length of the history and its skew and fat tails. A high Sharpe over a short, lopsided history earns less confidence than the same Sharpe over a long, symmetric one.
Deflated Sharpe ratio
The same probability, measured against the Sharpe the best of N variants with no skill would show by chance. This is the figure that changes most when the number of variants tried is honest.
Minimum backtest length
How many years of history a result needs before its Sharpe can be told apart from the luckiest of the variants tried. Forty five variants need five years at a Sharpe of one.
Probability of backtest overfitting
When every variant is supplied: how often the best variant in one half of the history ranks below the median in the other half, over every way of splitting it. Close to half means choosing the best did no better than choosing at random.
The haircut Sharpe ratio uses the Bonferroni adjustment of Harvey and Liu (2015), the strictest of the standard corrections. Results describe the history supplied and nothing else. For the same measurement as a signed, dated document for investors, see track record reports.