Backtesting — Trading Wiki

Publisher and editorial transparency

Published by TradeHQ. Site creator: Anuga Weerasinghe. This page has no recorded editorial review date; the About page explains the site's editorial approach.

What Backtesting means

Backtesting applies stated rules to historical data to examine hypothetical results. It can reveal problems with an idea but does not validate future performance; look-ahead, survivorship and overfitting bias matter.

In depth

Backtesting applies predefined rules to historical observations under stated execution and accounting assumptions. Its results describe that selected dataset; they are not executed returns or proof of a future advantage. Look-ahead bias uses information unavailable at the decision time, survivorship bias omits failed or unavailable instruments, and repeated parameter searches can overfit the same history. Separate rule development from held-out evaluation, document the searches performed and account for costs and missing observations. There is no universal number of years or trades that validates a strategy.

Generated simulator candles are not historical market observations.

Key points

• Simulates strategy performance using historical price data
• Overfitting can make selected historical results misleading
• Out-of-sample testing can challenge overfitting but cannot establish a guaranteed future edge

Why it matters when you are learning

Keep rule development separate from evaluation on unseen historical data. Record costs and failed cases. Neither backtesting nor forward simulation certifies future investment performance.

Worked exercise and its limits

Write a rule using a development sample, then freeze it before examining a held-out sample. Include costs, missing data and delisted instruments where relevant. A generated simulator chart is not historical market evidence, and a held-out gain does not guarantee future performance.