Re-blending every system pushed the efficiency frontier outward
Combining the Core v1.4.0 system with a Per-Pair aggressive strategy creates a more efficient portfolio than running either alone.
Combining the Core v1.4.0 system with a Per-Pair aggressive strategy creates a more efficient portfolio than running either alone. By blending these two, the monthly return adjusted for a 10% drawdown increases from 1.66% to 2.34%, representing a 40% improvement. The correlation between the two systems is 0.49. While this correlation is high enough to suggest they might move together, the diversification benefit still holds. The Per-Pair strategy adds enough return to push the efficient frontier higher without significantly increasing the drawdown, which sits at 8.2%. Interestingly, the previously tested Satellite strategy only drags down the performance; excluding it remains the correct decision.
Performance Breakdown
| Metric | Core v1.4.0 | Per-Pair | Blended (1:1) |
|---|---|---|---|
| Monthly Return | 1.38% | 1.80% | 2.50% |
| Drawdown (DD) | -14.1% | -8.0% | -8.0% |
| Profit Factor (PF) | - | - | 1.40 |
| Note: The monthly return and DD figures for the blend reflect the current market regime (2022 to 2026). The full-period data (which includes weaker market regimes) shows a 1.38% monthly return with a 14.1% drawdown. |
Robustness and M1 Stress Testing
This improvement is not just a product of a specific market environment. When looking at the out-of-sample (OOS) data from 2015 to 2020, the PF for the Core system was 0.54, but the blend reached 0.61. During the stronger period of 2021 to 2026, the Core system’s PF of 2.75 rose to 3.04. The results improve across both periods, indicating a genuine diversification effect rather than a temporary reliance on a specific market regime. I also performed rigorous intraday stress testing using M1 data to ensure the account wouldn’t collapse under extreme volatility. Using a risk setting of 0.0025 (0.25% of capital) per system, the worst single-day loss was 3.90%, with zero days breaching my safety limits. If I push the risk to 0.0035, the system hits a limit on a single day: April 29, 2024. As a result, I have capped the risk at 0.0025 per system. The Monte Carlo pass rate for this configuration is 90%.
Summary of Research Sessions
Over the course of these sessions, I tested 11 different variables to try and improve the system. Most of these (including machine learning sizing, mean reversion, and various breadth indicators) were either redundant or ineffective. However, two changes proved to be genuine improvements:
- Integrating the inter-market equity signal into the Core v1.4.0 system, adding 7.6% to the return.
- The Core and Per-Pair blend described above. This confirms that there is no single ceiling for these systems. By maintaining strict verification disciplines (specifically checking for false positives across multiple time periods), I was able to discard the “placebo” strategies and identify these two paths to better performance. No new code is required to deploy this; it is simply a matter of running both systems in parallel.
How this connects
This verification builds on earlier ones (what failed before and what I tried this time, comparisons between approaches).