Algorithm Aversion: Theory and Evidence from Robo-Advice

Abstract

Automation can lower costs and democratize access to many consumer services, but human discomfort with automation can pose barriers to technology adoption. We build a structural model of psychological "algorithm aversion," which features ongoing disutility of dealing with an algorithm, pessimism about the algorithm's ability, and uncertainty about the algorithm's performance; all three components can be assuaged by human interaction. We estimate model parameters using unique data from a "hybrid" robo-advising service in which portfolio management is automated, but clients are randomly matched with human advisors who provide different standards of support. Algorithm aversion is mainly driven by ongoing disutility and uncertainty, and human advice is especially important in retaining investors in robo-advice during market downturns.

Keywords: FinTech, Portfolio Choice, Behavioral Finance, Individual Investors, Technology Adoption, Structural Estimation, Algorithmic Aversion, Roboadvising

JEL Classification: D14, G11, O33