Free tool

Overfitting calculator

By Ignacio Arias, founder of TRAVIDENCE and author of the methodology · Last reviewed: August 2026

A backtest reports the Sharpe of the strategy you kept. It doesn't report how many versions you discarded along the way. This calculator estimates the best Sharpe that statistical noise produces given that many trials — and the probability that your result is real. Method: Bailey & López de Prado (2014).

Everything runs in your browser. Your numbers are never sent to any server.

The Sharpe your backtesting platform reports, annualized. For example: 1.8.

How many observations back that Sharpe, matching the frequency you picked above. For example: 756.

Every version you tried, not just the one that survived.

Where do I get this?

This number is the heart of the calculation: it measures how much you searched before keeping this version.

One optimizer pass counts in full: 3 parameters with 10 values each = 1,000 trials.

Every asset, timeframe or filter you tried and discarded counts too.

If you didn't keep count, estimate high: declaring fewer trials makes the result look better than it is.

Advanced (optional)

If you don't know them, keep the defaults: they correspond to a normal distribution.

Asymmetry of your returns: negative = extreme losses outweigh extreme gains.

How often extremes happen: 3 = the normal bell curve; above 3 = fatter tails.

What this calculator doesn't do. This calculator evaluates a single overfitting signal: selection bias on the Sharpe ratio. It doesn't analyze your trade record trade by trade, or stability across subperiods, or consistency across volatility regimes, or per-component contribution. That's what the full audit does.

Audit your strategy before you risk capital →

Method and limits

Deflated Sharpe Ratio formulas from Bailey & López de Prado (2014): the expected best Sharpe out of N trials when no real signal exists is estimated with the paper's extreme-value approximation, and the DSR is the probability that the true Sharpe clears that bar, adjusting for sample size, skewness and kurtosis. Without the list of Sharpe values from your N trials, the cross-trial variance is approximated by the sampling error of the Sharpe estimator (Lo 2002; Mertens 2002) — a standard, disclosed approximation. Reference: Bailey, D. H., & López de Prado, M. (2014). "The Deflated Sharpe Ratio: Correcting for Selection Bias, Backtest Overfitting and Non-Normality." The Journal of Portfolio Management, 40(5), 94–107.

This tool answers whether your result is real. To see how much the ride can hurt if it is, use the Monte Carlo Simulator →

Notice. TRAVIDENCE is an independent validation service. It is not financial advice or an investment recommendation. Past performance does not guarantee future results; no validation eliminates the risk of loss.