Look-ahead bias
Look-ahead bias is when a backtest makes a decision using information that would not have existed at that moment in real time. It produces results that are not merely optimistic but impossible.
In plain terms
A strategy that enters at the day's opening price based on a signal calculated from the day's closing price will look extraordinary. It is also unimplementable, because on the morning in question the close had not happened yet.
Most real cases are less obvious than that. Using a daily high to set an intraday stop. Applying an indicator that needs the full bar to a decision taken mid-bar. Using economic data at its release date when it was later revised. Using an index constituent list as it stands today for a period when the constituents were different.
Why it matters
Look-ahead bias is different from overfitting in an important way. An overfit strategy might still work a little. A strategy with look-ahead bias cannot work at all, because it depends on information that will never be available when you need it. The backtest is not optimistic; it is describing something that cannot happen.
It is also hard to spot, because the results look great and nothing errors. The strategy does not complain about seeing the future.
How Mithos handles it
The strategy builder is a defined library of blocks rather than a free-form code editor, which is partly a support decision and partly this. Each block's data access is bounded to what was genuinely available at the decision point, so the most common ways of accidentally peeking are structurally unavailable rather than left to the user to avoid.
Where a test runs against a data stand-in rather than the real instrument — an ETF proxying an index, for instance — that is surfaced explicitly as a distinct verdict rather than being folded silently into a pass.
Answers to the usual ones
- Is survivorship bias the same thing?
- Related but distinct. Survivorship bias is testing on a universe that excludes things that failed and disappeared. Both involve using knowledge that did not exist at the time, but the mechanism differs.
- How would I know if my backtest has look-ahead bias?
- Suspiciously good results are the main clue, particularly a very high win rate or unusually small drawdowns. If a backtest looks too good, look-ahead bias should be the first thing you check for, before you start celebrating.