Global index diversification: the ceiling did not move

The current version of my core trading system (v1.2.0) has reached its performance ceiling, and further attempts to diversify the equity index sleeve only degrade overall results. I began this research track by isolating the index sleeve from my v1.2.0 system. The current setup relies on three U.S. indices (US500, US100, and US30). Its baseline performance is a total return of 152.7%, a drawdown (peak-to-trough decline) of 10.0%, and a Profit Factor (PF) of 1.45. Since these three U.S. indices have a high mutual correlation of 0.76, I hypothesized that adding international indices (DE40, JP225, and UK100) to create a “Global 6” portfolio would reduce system-wide drawdown through better diversification. However, the results showed that geographic diversification is counterproductive.

Index Portfolio	PF	Monthly Return (at 10% DD)	MC Pass Rate
U.S. 3 (Current)	5.38	+1.16%	92%
U.S. 3 + JP225	-	+1.07%	-
Global 6	2.27	+0.92%	88%
Global 6 + Connors	-	+0.87%	-
Note: PF refers to Profit Factor (gross profit divided by gross loss; >1 is profitable). MC refers to the Monte Carlo pass rate (the probability of meeting prop-firm equity rules).
The primary issue is a “quality dilution” effect. While the U.S. indices are high-quality assets with a Sharpe ratio of 0.87, the non-U.S. indices perform poorly on their own. The benefit gained from lowering the correlation from 0.76 to 0.34 is significantly smaller than the loss incurred by adding these lower-quality components. Even when I attempted to re-optimize the budget allocation across these assets, the monthly returns remained capped near +0.87% to +0.92%, failing to exceed the +1.16% return of my current U.S.-only configuration.
My investigation into the system’s drawdown revealed that the indices are not the bottleneck. The real source of my system’s drawdown is the 0.54 correlation between the core trend-following component and the secondary “sat2” strategy. The indices, Connors, and calendar strategies are already effectively uncorrelated with the core.
This brings me back to a lesson I first encountered in earlier research: if a component is not the source of your drawdown, trying to “improve” it through further diversification usually just adds dead weight.
Reducing the core-to-sat2 correlation would require non-price data like interest rate differentials. Because these are generally unavailable or unreliable for prop-firm trading environments, I have hit a wall. For now, v1.2.0 (with its ~0.68% monthly compounded return, 10% drawdown, and 92% MC pass rate) remains the most efficient configuration possible under my current constraints. I will be making no changes to the system.

How this connects

This verification builds on earlier ones (what failed before and what I tried this time, comparisons between approaches).