Hands-On Machine Learning for Algorithmic Trading

The minimum backtest length and the deflated SR

Marcos Lopez de Prado (http://www.quantresearch.info/) has published extensively on the risks of backtesting, and how to detect or avoid it. This includes an online simulator of backtest-overfitting (http://datagrid.lbl.gov/backtest/).

Another result includes an estimate of the minimum length of the backtest that an investor should require given the number of trials attempted, to avoid selecting a strategy with a given in-sample SR during a given number of trials that has an expected out-of-sample SR of zero. This implies that, e.g., if only two years of daily backtest data is available no more than seven strategy variations should be tried, and if only five years of daily backtest data is available, ...