A Backtesting Checklist

Backtesting is useful only when the experiment is honest. Before trusting a result, I want to check the assumptions that can quietly inflate performance.

The basic checklist: define the universe before the test, avoid look-ahead bias, model transaction costs, handle missing data carefully, separate training and evaluation periods, and report drawdowns alongside returns.

A good backtest should make failure modes visible. The result is not just an equity curve; it is an argument about data, execution, risk, and robustness.