How To Properly Combine Signals Of Different Speeds

Introduction

The signals with the sharpe in your backtest might be destroying your live performance. Not because they stopped working, but instead it’s because trading them costs more than they return.

This is the core problem with combining signals that have different time horizons. A momentum signal that predicts returns over the next week and a value signal that predicts returns over the next year look comparable in a regression. They might even have similar information coefficients. But they have completely different trading implications, and most signal combination approaches treat them identically. This is a big problem, and you’ll see why!