Level 4 · Lesson 3 / 4 · English
Backtesting, unseen evaluation and forward testing
4 min reading · Exercises take additional time
0 of 4 lessons completed for this visit
A backtest applies a rule to historical observations under execution assumptions. It can expose inconsistencies and estimate sample outcomes, but it cannot guarantee future results. Honest testing protects the chronology of information and separates developing an idea from evaluating it.
What you will learn
- Explain backtesting, unseen evaluation and forward testing.
- Work through the stated hypothetical example with explicit units.
- Recognise limitations and apply a documented practice check.
Protect the information boundary
Look-ahead bias occurs when the rule uses information not available at the simulated decision time. Examples include using a candle’s final high for an earlier entry or using an indicator recalculated with future observations. Store the decision timestamp and identify precisely which data each rule could access then.
Backtests need bid/ask or defensible spread assumptions, commissions, slippage, volume constraints and an order-fill model. A touch of a chart level does not prove an order could fill at that price and size. If both stop and target occur inside one candle and order is unknown, use finer data or explicitly conservative handling rather than assuming the favourable sequence.
Separate development from evaluation
Use a development sample to specify the rule, then freeze it before an unseen evaluation sample. Changing parameters after inspecting evaluation results makes that sample part of development. Repeatedly trying many configurations can produce an attractive winner by chance; document the search rather than showing only the best version.
Forward testing follows the frozen rule as new observations arrive. Demo testing can check workflow and timing but may not replicate real fills, liquidity or emotional pressure. Do not merge demo outcomes with verified live performance. Evaluate a meaningful sample and uncertainty, not a few favourable trades, before making claims.
Key terms
- Look-ahead bias
- Using information that was not available at the simulated decision time.
- Out-of-sample evaluation
- Testing frozen rules on observations not used to select them.
- Forward testing
- Following pre-defined rules as new information arrives.
Hypothetical worked example · Not live market prices
Hypothetical net test outcomes
- Assume three gross outcomes are +30, −20 and +10 USD.
- Assumed transaction cost is USD 2 per trade, totalling USD 6.
- Gross total is USD 20; net total under these assumptions is USD 14. This tiny sample establishes no reliable edge.
| Trade | Gross USD | Net after USD 2 |
|---|---|---|
| 1 | 30 | 28 |
| 2 | −20 | −22 |
| 3 | 10 | 8 |
| Total | 20 | 14 |
Simplified educational calculations. Costs, slippage and provider-specific adjustments apply where relevant; these are not actual returns.
Common mistakes to avoid
- Selecting settings using the evaluation set; freeze rules before unseen testing.
- Ignoring ambiguous intrabar sequences; model uncertainty explicitly.
- Calling demo testing verified real performance; keep the record labelled accurately.
Lesson summary
Test chronology, fills and costs honestly. Historical and forward samples are evidence with limitations, not promises of future profit.
Practical exercise
Create a test checklist with signal timestamp, available data, executable order time, costs and intrabar ambiguity. Calculate gross USD 20 minus USD 6 costs as USD 14 net for the supplied three-trade example, without generalising it.
Knowledge check
Check your understanding
Results apply to this attempt only and are not saved to a student account. No certificate is issued.
Answer all questions correctly and complete the exercise. Progress follows you between Academy pages for this visit only; it is not saved to an account and resets on reload.
Related lessons
Financial education, not personal investment advice. Leveraged Forex and CFDs can lead to substantial losses. Demo practice and correct quiz answers do not guarantee trading success.