Backtesting without accidentally peeking at the answer
Backtesting applies a rule to historical information. The result is only meaningful if the rule uses information that would actually have been available at each decision point. Otherwise the exercise quietly borrows knowledge from the future.
Freeze the rule before testing
Write the entry, exit and cost assumptions before inspecting the chosen evaluation period. If you change the rule after seeing a result, label the revision and test it on different data. Repeated adjustments to the same history can fit noise.
Audit the timestamps
A daily close is not available earlier that same day. A revised economic figure may differ from the first published number. A security or currency dataset can also contain missing periods or inconsistent time zones. Record those limitations in the result.
Report more than a headline return
Include the number of observations, losing stretches, costs, maximum historical drawdown and sensitivity to slightly different assumptions. A backtest does not show what future liquidity or behavior will be. Treat it as a way to discover weaknesses in a hypothesis, not a certificate that a live strategy has become safe or reliably profitable.
Have a correction or a question about this article? Contact the editorial team. Read our editorial policy and disclaimer.
