Hands-On Machine Learning for Algorithmic Trading

Survivorship bias

Survivorship bias emerges when a backtest is conducted on data that only contains currently active securities and omits assets that have disappeared over time, for example, due to bankruptcy, delisting, or acquisition. Securities that are no longer part of the investment universe often did not perform well, and including these cases can positively skew the backtest result.

The solution, naturally, is to verify that datasets include all securities available over time as opposed to only those that are still available when running the test.