FeaturedAI TradingWalk-Forward Research
Would You Take +923 Points With a −1,129 Drawdown, or +1,718 With Only −310? How SAIL Lab's AI Balances Profit and Risk
SAIL Lab TeamOctober 20263 min read
A strategy is only usable if you can live through its worst stretch. We started from our Baseline Trend Policy, added SAIL Lab's proposal step by step, and measured profit and drawdown on 14 months of XAUUSD ticks (Jan 2025 – Feb 2026) the system had never seen.
Profit vs risk, step by step
+923 / −1,129
Baseline Trend Policy, fixed parameters
Profit / max drawdown: 0.8
+2,610 / −702
Baseline + our Optimization
Profit / max drawdown: 3.7
+2,997 / −774
Baseline + our Optimization + our Entry filters
Profit / max drawdown: 3.9
+1,718 / −310
Baseline + our Optimization + our Entry filters + our ML
Profit / max drawdown: 5.5
+2,163 / −665
Gold buy-and-hold (+75%)
Profit / max drawdown: 3.3
Out-of-sample net P&L / max drawdown in points, Mar 2025 – Feb 2026, re-optimizing every 5 days on the last 60 days. 1 pt = $1 per 0.01 lot.
Key findings
- 1Our full proposal cuts the maximum drawdown by 73% versus the Baseline (−1,129 → −310) while nearly doubling the profit (+923 → +1,718).
- 2Our ML is a risk control: compared with the same strategy without ML, it gives up about 40% of the profit but cuts the maximum drawdown by 55–60%.
- 3It is skill, not just trading less: at the same profit, randomly skipping trades leaves a 12–38% deeper drawdown than our ML.
- 4The 60-day re-optimization holds in a falling market (gold −21%) and a rising one (gold +101%), beating 93–98% of 3,001 fixed parameter sets.
- 5For maximum profit, run without ML (+2,997 pts, drawdown −774). Entry filters and long/short bias depend on the market, so we treat them as settings, not edge.
Full report (PDF)
Download PDF ↓All results are out-of-sample backtests on tick data. Past performance does not guarantee future results.