How To Be An Expert At Trading Multiple Forecasts Optimally

Introduction

You fancy yourself a machine learning expert, and have produced multiple forecasts over multiple horizons, how do you trade them optimally? How do you combine them?

When you have forecasts at multiple horizons (your 1-minute signal, your 5-minute signal, your 60-minute signal, all refreshing every second), the natural instinct is to combine them somehow and trade toward the combined target. Equal weights. Maybe signal-weighted. Maybe volatility-adjusted.

All of these are wrong.

The correct approach is counterintuitive: you should weight signals by how persistent they are, not JUST how accurate they are. And you should never fully trade to your target. You should trade partially toward an aim portfolio that itself is forward-looking. Trade less per period. But aim farther ahead.

Wayne Gretzky skated to where the puck was going, not to where it was. The same logic applies to your aim portfolio.