Backtesting biases: look-ahead, repainting and survivorship

What it is

Look-ahead bias is using information from the future. Common versions: using a candle's close to decide a trade that would have been entered before the close; using the session's final POC or high instead of the developing one; acting on a higher-timeframe candle that hadn't closed yet.

Repainting is when an indicator's past values change after the fact: ZigZag, some fractal and pivot tools, Heikin-Ashi, and lower-timeframe-based footprints can all look better in hindsight. Survivorship bias is testing only the instruments that still exist (common with stocks), which hides the ones that failed. Hindsight labelling, deciding afterwards which swing was the 'real' order block, is the manual version of all three.

How to spot it

• Ask of every rule: could I have known this at the moment of entry?
• Results that look too good almost always have a bias in them.

How to use it

• In UNFLD's backtester, step through bar by bar and decide before the next bar appears.
• Use only confirmed swings, closed candles and developing values in your rules.

Common mistakes

• Marking structure on the finished chart and then 'testing' entries on it.
• Using indicators that repaint without knowing it.

Related