Backtesting Review

Definition

Backtesting review is the audit phase after testing a strategy on historical data. It checks that results are produced without future information (no look-ahead), include real-world frictions (spread/fees/slippage), and remain stable across market regimes. The output is a clear decision: keep, refine, or reject the model, plus a documented playbook of what conditions the model works best in.

Why It Matters

A strategy can look profitable in hindsight while failing live due to bias (look-ahead/data-snooping), unrealistic fills, or overfitting to a single regime. A disciplined review prevents false confidence, forces you to measure what actually matters (risk, drawdown, expectancy), and creates a repeatable improvement loop for ICT models (setup quality, timing windows, and execution rules).

How to Identify

1. Confirm your backtest is rule-based enough to be repeatable (clear entry trigger, stop logic, and exit/partial rules).
2. Verify data integrity: timestamps are correct, session times are consistent, and the candle you use for signals is the candle you could have known at decision time (no future leak).
3. Ensure execution realism: include spread/commission and a slippage assumption (especially during volatility/news).
4. Check sample adequacy: enough trades across multiple weeks/months and more than one market regime (trend, range, high-volatility).
5. Review metrics that matter: expectancy, win rate, average win/average loss, max drawdown, profit factor, and largest losing streak.
6. Perform a compliance audit: did you take only A+ setups, inside your intended session windows, aligned with your HTF bias rules?

How to Trade

1. Convert your setup into a checklist with hard rules: context filter(s), trigger (raid/displacement/MSS), entry model (PD array), stop placement, management, and exit criteria.
2. Use bar replay and record every trade with screenshots/notes at entry and at exit. Log the reason for entry and whether all rules were satisfied.
3. Recompute results after applying spread/commissions and a conservative slippage model; compare gross vs net performance.
4. Break performance by session (Asia/London/NY), day-of-week, and regime (trend vs range). Identify where the model truly works best.
5. Change only one rule at a time (e.g., displacement threshold, session filter, stop logic). Re-test to avoid data-snooping.
6. After a clean historical review, paper-trade/forward-test the same rules in real time before sizing up.

Common Confusions

Related Concepts

More Educational Concepts