Published 2026-10-14
Look-ahead bias
Look-ahead bias is when a backtest uses information that would not have been available at the time of the trade — the single most common reason an impressive backtest does not survive contact with the market.
What it is
Look-ahead bias is when a backtest uses information that would not have been available at the time of the trade.
It is the single most common reason a backtest looks excellent and then fails in practice.
Why it matters for AI quant
Large language models make look-ahead bias worse, not better. A model trained on historical data has, in effect, already seen the future. Ask it to “analyse” a period it was trained on and the answer can look prescient for reasons that have nothing to do with analysis.
This is why rule 3 of the standards requires look-ahead protection to be stated explicitly.
Common forms
| Form | Example |
|---|---|
| Training-data leakage | Asking an LLM about a period it was trained on |
| Restated fundamentals | Using a financial figure as restated later, not as first reported |
| Index membership | Testing on today’s constituents, which already excludes the failures |
| Trading on close | Using a closing price to decide a trade executed at that same close |
How to check
Ask one question: at the moment this decision was made, was this number actually available?
If you cannot answer that with a source, the result is not a result.