XAUUSD M3 Volatility Breakout EA — First Tradeable Backtest
Backtest Update — XAUUSD M3 Volatility Breakout EA
We initially found a structural issue in the compression filter: the original 1.30 ATR threshold produced zero trades across 19+ months of testing.
After analysing the historical distribution, we tested a 3.0 ATR compression threshold.
Backtest: 01 Jan 2025 → 18 Aug 2026Symbol: XAUUSDTimeframe: M3Trades: 12Net P/L: -$2.24Profit Factor: 0.93Max Drawdown: 2.48%
This is not a profitability claim.
The objective of this test was to verify whether the strategy could become genuinely tradeable while keeping the original architecture intact.
The next step is to test 3.5 and 4.0 thresholds without changing the other parameters and compare the behaviour.
Research first. Optimization later.
Good that you are treating this mainly as a test of trade frequency. But 12 trades is much too small to compare the profitability of 3.0, 3.5 and 4.0 thresholds.
I would test a wider fixed range of values on separate periods, because selecting thresholds after seeing the same historical data can easily become curve fitting. Also check whether most of those 12 trades happened during only one volatility regime or short period.
For now I would care more about stable trade frequency and similar behaviour across different periods than profit factor. With only 12 trades, one result can change PF completely
fxrobustix posted:Good that you are treating this mainly as a test of trade frequency. But 12 trades is much too small to compare the profitability of 3.0, 3.5 and 4.0 thresholds.
I would test a wider fixed range of values on separate periods, because selecting thresholds after seeing the same historical data can easily become curve fitting. Also check whether most of those 12 trades happened during only one volatility regime or short period.
For now I would care more about stable trade frequency and similar behaviour across different periods than profit factor. With only 12 trades, one result can change PF completely
Thanks again for the detailed feedback. I agree — with only 12 trades, comparing profitability across thresholds would not be statistically meaningful.
I’ll focus first on trade frequency and consistency across different market periods/regimes, rather than optimizing for a better PF on the same historical sample.
The goal is to understand whether the underlying behaviour is robust before making any parameter changes. Appreciate the input.
