Researcher
• 市場觀點
From black box to glass box: Understanding and attributing machine learning models
Machine learning techniques are increasingly important in quantitative investing, thanks to their ability to capture complex financial data dynamics and improve return and risk estimates. But the resulting higher complexity means such models are sometimes perceived as being ‘black boxes’. In a new whitepaper, Robeco researchers provide insights into the tools available to understand and interpret machine learning models.
Resumen
- Machine learning (ML) models add value in stock selection strategies
- Rationalizing ML algorithms is vital for their successful use and clients’ trust
- We apply proprietary techniques to understand the drivers of ML predictions and to attribute the performance of ML strategies
ML techniques, including algorithms such as neural networks and gradient-boosted regression trees, can help quant investors understand and exploit complex financial data dynamics. These algorithms capture nonlinear relationships and interactions in a flexible way and can improve return and risk estimates, but can also add to a sense of complexity and opaqueness.
Prediction and performance
In our new whitepaper, we aim to provide insights into the tools we use to understand and interpret ML models. It’s essential to grasp the relationship between the input features and the resulting ML predictions on one hand, and to understand the ML-based performance of associated investment strategies on the other. The enhanced transparency and interpretation from these tools contribute to the successful use of machine learning techniques in practice, and over the past years, a wide variety of ML signals have been added to Robeco’s quant strategies. Done properly, machine learning approaches to investment need not constitute an opaque black box, but a clearly understood ‘glass box’.
Descargar la publicación
Important information
insights.detail.disclaimer.text
