From Backtest to OOS: Building Mother Superior as an Eight-Module Gold Portfolio
Mother Superior started from a simple problem: gold does not behave the same way in every market regime. One entry model can look excellent for a while and then struggle when volatility, direction or trade frequency changes. Instead of searching for one perfect setup, I built eight separate XAUUSD modules and tested them as one portfolio.
THE TESTING STRUCTURE
I wanted to keep development data separate from validation data.
2022–2024 was the in-sample (IS) period. This is where the module logic and parameter choices were developed. A strong IS result shows that the design fits its intended job, but it is not independent proof because the development decisions were influenced by those years.
2025–2026 was the out-of-sample (OOS) period. The selected parameters were frozen before this section was evaluated. This is the more important reality check: does the final structure still work on data that was not used to choose its settings?
I also treated 2021 separately as stress/history, while 2017 and 2018 were used for additional survival checks. Those older checks are supporting evidence and are not mixed into the formal IS result.
THE COMPLETE REAL-TICK RUN
The main portfolio report covers January 2021 to September 2026 using real ticks. It started at USD 1,000 and finished at USD 3,903.12.
Closed trades: 1,511
Profit factor: 2.93
Maximum relative equity drawdown: 12.82%
Annual results:
2021: +45.03%, PF 3.03
2022: +32.36%, PF 2.71
2023: +16.22%, PF 2.82
2024: +13.62%, PF 1.94
2025 OOS: +16.84%, PF 3.01
2026 OOS through September: +31.79%, PF 7.18
The annual return changed considerably, which is normal. The useful part is that the edge did not disappear after the development window. Both frozen-parameter OOS years remained profitable. The high 2026 profit factor is still only a partial-year observation, so I do not treat it as a permanent expectation.
WHY EIGHT MODULES?
Each module has its own trade identity and basket management. The final portfolio combines M15, M30 and H1 logic with module-specific spacing, targets and exits. The purpose is not to make every module trade all the time. It is to avoid depending on one narrow market behaviour.
The modules must also be judged together. Several strategies can look independent while creating the same underlying gold exposure. Portfolio-level floating equity therefore matters more than counting attractive individual backtests.
FULL-HISTORY MONTE CARLO
I ran 10,000 block-bootstrap paths using the complete 2021–2026 deal sequence. The simulation resampled 50-deal blocks, preserving short clusters of wins and losses instead of pretending that every deal was independent.
The median simulated ending balance was USD 3,896. The 5th to 95th percentile range was approximately USD 3,504 to USD 4,318. Median maximum realised-balance drawdown was 4.1%, while the 95th percentile was 8.0%.
This calculation only rearranges realised deal results. It does not recreate every floating position combination and cannot invent an unseen future regime. For that reason, the observed MT5 relative equity drawdown of 12.82% remains the more relevant risk figure.
WHAT COMES NEXT
This test does not prove that future performance will match the backtest. It shows that the same final portfolio remained profitable across several years and continued working after the parameters were frozen for OOS validation.
The next stage is live observation. Real spreads, slippage, swap, broker execution and the behaviour of floating baskets now have to confirm what the tester showed. The backtest is the starting evidence. Live trading is the next filter.
