Backtest hygiene: the errors that flatter every strategy

Almost every backtest overstates what the strategy would have earned. The errors are well catalogued, and each one biases results in the same direction.

Survivorship

Testing on today's index constituents excludes every company that was removed for failing. The universe must be reconstructed as it stood on each historical date.

Look-ahead

Fundamental data is reported weeks after the period it describes. A backtest that uses Q4 earnings on 31 December has used information that did not exist on 31 December. Lag every data point to its publication date.

Transaction costs

Assume costs, then assume more. A cost assumption that is conservative for liquid large caps is usually generous for anything smaller.