Embracing Machine Learning To Tackle Portfolio Optimisers Limitations

Abstract

The content of this article will be concerned with the mathematical limitations of the early MPT theories and will leave aside other topics related to portfolio optimization such as the factoring of behavioural biases, portfolio optimization criteria (by style, country, industry, etc), or the purpose of the optimization (asset allocation, ALM, long-short portfolios, etc). Furthermore, this article does not intend to cover all the body of research but only to emphasize those models that either propose a brand new approach, or have been broadly adopted by the industry over the last decade.

MVO (Minimum Variance Optimization) was an innovative approach to investments more than fifty years ago, albeit it was not without pitfalls. New machine learning techniques such as PCA, Clustering or Graph Theory have been able to tackle the main challenges proposed by the original MVO portfolio optimization problem.