How to Backtest an ICT Strategy Properly (Without Fooling Yourself)
Scroll through any chart in hindsight and the perfect setups jump out. That is hindsight bias, and it is why most "backtests" of discretionary ideas prove nothing. A proper test converts the idea into written rules and runs them candle by candle.
The rules of an honest backtest
- Write the rules first. Every condition must be objective — the reason this platform defines sweeps, MSS and FVGs so strictly.
- No lookahead. At each step the strategy may only see candles that have closed. A swing point is only known after its confirmation candles.
- Same code live and in testing. BlessedForex runs one shared engine for both, so results are comparable.
- Conservative fills. If a candle reaches both stop and target, assume the stop.
- Record every setup, including rejected and invalidated ones.
Metrics worth reading
- Number of trades (enough to mean something)
- Win rate, average and median R, total R
- Maximum drawdown in R and longest losing streak
- Breakdown by session, weekday, direction and setup type
Overfitting: the silent killer
If you tweak settings until past results look great, you have described the past, not the future. Guard against it:
- Change one thing at a time and keep the rest fixed.
- Save each change as a new strategy version and keep old results.
- Test on a period you did not tune on.
- Prefer settings that work acceptably across a wide range of values.
What a backtest cannot show
Real spreads, slippage, requotes and your own emotions. Follow a backtest with demo forward testing.
Reproducibility
Given the same candles, strategy version and settings, results must be identical every time. If they are not, something is wrong.
Reminder
Back-tested results do not guarantee future performance. They are a tool for understanding behaviour, not a promise.
Questions & answers
How many trades do I need for a meaningful backtest?
As many as practical. Dozens is a minimum for a first impression; hundreds across different market conditions are far more informative.
What is lookahead bias?
Using information that would not have been available at the time of the decision, such as a swing that is only confirmed later.
Why save strategy versions?
So old results stay tied to the rules that produced them. Changing rules silently makes comparisons meaningless.
Can I trust results from the mock data feed?
No. Mock data is synthetic and only for development. Use real XAU/USD candles from a data provider, CSV import or your MT5 terminal.