Build a trading process
The idea
Backtesting applies fixed rules to past observations. Look-ahead bias uses information unavailable at the decision time; overfitting selects rules that explain noise. Include costs and realistic fills, separate development from evaluation data, and report sample size. Historical success does not establish future profitability.
A worked example
Choosing the best settings from hundreds of trials and reporting only that winner overstates evidence. A held-out period offers a less flattering but more useful check.
Try it yourself
Define rules before reviewing a new sample and record every qualifying trade, including missed fills.
A mistake to avoid
Repeatedly adjusting rules after seeing the test-period result.