Unsupervised machine learning to reveal South African risk behaviour archetypes in the domain of discretionary investment decisions

Abstract

The mediated risk models of decision making under conditions of risk propose that individuals’ recent experiences of investments can lead to them making investment switching decisions that would be different to those based on their long term risk preferences. This study used the medoids clustering algorithm applied to a large, novel, dataset to identify four statistically significant patterns of switching behaviour based on the dimensions highlighted by this body of theory. Individual decision makers exhibit three consistent patterns or archetypes of risk-seeking (termed ‘assertive’), loss aversion (termed ‘anxious’) and risk aversion (termed ‘avoider’) behaviour over most time periods. Each consistent behaviour pattern is, however, susceptible to temporary deviations from ‘normal’ behaviour. Loss averse investors are on occasion drawn into both risk averse and risk seeking behaviour. Similarly, risk seeking investors are sometimes drawn into both loss averse and risk averse switching behaviour. This provides evidence to support the mediated perspective on risk-based decision-making behaviour and demonstrates the viability of this machine learning-based method for segmenting investors from a financial advice, marketing, and communications perspective.