What Optimization Could Not Do, Diversification Did: Five Uncorrelated Markets, No Losing Years

For a year we tried to optimize single strategies into something spectacular. The honest ceiling kept landing in the same place: a good trend system on one market wins roughly 6-7 years out of 8, with losing years of -5% to -8% in between. Push the parameters harder and the backtest improves while the future degrades — that trade is always available and always bad.

We run trend-following with hard stops on markets that genuinely move for different reasons: JPY cross trends driven by rate differentials, and crypto trends driven by their own cycle. Individually, each component still has its losing years. Combined at equal risk, something simple and unglamorous happens: their bad years land in different calendars.

In our combined runs, the portfolio of uncorrelated components produced positive results in every tested year, with the worst year still double-digit positive and the standard deviation of yearly returns roughly halved versus the strongest single component. No component was optimized any harder to get there. The improvement came entirely from the correlation matrix, which — unlike a parameter set — does not care what the optimizer wanted.

1. A parameter edge decays; a structural correlation edge is about what the assets are.

2. Diversification improves the worst case, which is what actually removes accounts from the game. Optimization mostly improves the average case in the past.

3. It is testable honestly: each component is verified separately on real ticks, then the combination is arithmetic, not another fit.

You do not need our products to use the idea. Take two or three strategies you already trust, check that their losing periods do not overlap, and split risk evenly. The whole will be calmer than any part. If you want our packaged version of it: https://www.mql5.com/en/market/product/182751

All products and free utilities: https://www.mql5.com/en/users/app.develop.sk/seller