Using machine learning to predict investors’ switching behaviour

Abstract

Individual investors’ decisions to switch investments very often lead to significantly lower investment returns so having an effective predictive model of these switches would be of value to clients, advisors and investment managers. A random forest algorithm was applied to a new dataset of over 20 million observations relating to 95,685 clients on Momentum Investments’ platform between 2018 and 2024. It identified a combination of investor characteristics (number of holdings, past switching behaviour, total assets) and external features (past returns, macroeconomic variables) as the key features of investor switch behaviour. This model exceeds commercially accepted standards in respect of the AUC and Gini metrics showcasing the model’s strength in its ranking capability. It can thus provide a useful basis for client segmentation and engagement by financial advisors.