UX in AI: Trust in Algorithm-based Investment Decisions

Abstract

This Thesis looks at investors’ loss tolerance with portfolios managed by a human advisor compared to an algorithm with different degrees of humanization. The main goal is to explore differences between these groups (Humanized Algorithm, Dehumanized Algorithm, Humanized Human and Dehumanized Humans) and a potential diverging effect of humanizing. The Thesis is based on prior research (Hodge et al., 2018) but incorporates new aspects such as additional variables (demographics, prior experiences) and a comparison between users and non-users of automated-investment products. The core of this research is an experiment simulating an investment portfolio over time with four different portfolio managers. Subjects were asked to decide if they want to hold or sell a declining portfolio at five points in time to measure their loss tolerance. A cox regression model shows that portfolios managed by the Humanized Human had the highest loss tolerance. Humanizing leads to higher loss tolerance for the human advisor but to lower loss tolerance for algorithmic advisors within the non-user group.

Keywords: Künstliche Intelligenz; Artificial Intelligence; Behavioral Finance; Behavioral Economics; Human-Computer-Interaction; User Experience; Investmententscheidungen; Nutzervertrauen.