This Robeco white paper explains how to interpret machine learning models in quantitative investing by transforming them from “black boxes” into “glass boxes.” It covers methods to understand ML predictions using tools like correlation analysis and Shapley values, which help quantify how each input feature affects predictions. It also introduces a proprietary return attribution framework that breaks down portfolio performance into components such as generic factors, proprietary signals, data-driven effects, and interaction-driven outcomes. The goal is to ensure transparency, validate economic rationale, and build client trust in ML-driven strategies.

Join for free to read