What is look-ahead bias in backtesting?

Question

AcadiFi · Accepted Answer

Look-ahead bias occurs when a backtest uses information not available at the time of the hypothetical trade. Classic examples: (1) using revised GDP figures (first release is what traders saw, later revisions weren't available); (2) ranking stocks by fiscal-year earnings on Jan 1 when 10-K filings aren't out until March; (3) using index constituents as of today to backtest history (includes companies not in the index back then — plus a dose of survivorship bias); (4) using restated financials retroactively. Example: Fenwick Strategies backtests a value strategy ranking S&P 500 firms by P/E every January 1. They use Compustat's final, restated book values and earnings. Result: 18% annualized return. Problem: a January 1, 2020 trader wouldn't have had Q4 2019 earnings yet — and certainly not 2023 restatements. Correcting for lag (using only data available 75 days after fiscal year-end, with point-in-time financials) drops returns to 11%. Fixes: (1) use point-in-time databases; (2) enforce realistic reporting lags; (3) use as-reported (not restated) data; (4) for index strategies, use historical constituent lists.

What is look-ahead bias in backtesting?

Master Level II with our CFA Course

Related Questions

Practice Questions