Backtesting without fooling yourself

A backtest applies rules to historical data. Common traps: look-ahead bias (using data you would not have had at the time, such as revised earnings), survivorship bias (testing only companies that still exist), data snooping (trying hundreds of rules and keeping the best), and ignoring costs, slippage and taxes.

Good practice: define the hypothesis first, use point-in-time data including delisted companies, keep an out-of-sample period untouched until the end, apply realistic costs, and test robustness by varying parameters. Harvey, Liu and Zhu (2016) argued that because so many factors have been tested, new ones should clear a much higher statistical bar (t-statistic above about 3).

Then paper trade (Vestora Paper trading) before risking money.