Machine learning for financial applications: self-organising maps, hierarchical clustering and dynamic time-warping for portfolio constructive

Abstract

This study investigates how modern machine learning (ML) techniques can be used to advance the field of quantitative investing. A broad literature review evaluated the common applications for ML in finance, and what ML algorithms are being used. The results show ML is commonly applied to the areas of Return Forecasting, Portfolio Construction, Ethics, Fraud Detection Decision Making Language Processing and Sentiment Analysis. Neural Network technology and support vector machine are identified as popular ML algorithms. A second review was carried out, focusing in the area of ML for quantitative finance in recent years finds three primary areas; Return forecasting, Portfolio construction and Risk management.

A practical ML experiment carried out as a proof of concept of ML for financial applications. This experiment was informed by the results of the broad and more focused literature searches. Two forms of ML techniques are used to analyse market return data and equity flow data (provided by State Street Global Markets) and create a portfolio from insights derived from the ML technology. The ML technologies employed are those of Self-Organising Maps and Hierarchical Clustering. The portfolios created were tested in terms of risk, profitability and stability. Stable regimes and profitable portfolios are created. Results show that portfolios obtained by analysing equity flow data consistently outperform those created by analysing return data.