Algorithm Aversion : Theory and Evidence from Robo-Advice

Algorithms hold great potential to lower costs and democratize access to a wide variety of consumer services. How do humans interact with algorithms, and what are the major barriers to algorithmic adoption? We answer these questions using a structural model applied to unique data that captures interactions between human clients and “hybrid” robo-advisors that offer different levels and standards of human counseling to complement algorithmic investment. The model features three dimensions of investors’ algorithm aversion, all of which can be influenced by human advice, namely: a per-period disutility of dealing with the algorithm, pessimism about the algorithm’s ability to manage assets, and uncertainty about the algorithm’s performance. We estimate the model’s parameters using quasi-random variation in the matching of clients with human advisors generated by mechanical allocation rules. We find evidence that algorithm aversion is mainly driven by ongoing disutility and uncertainty and that human advice is important in retaining investors in robo-advice during market downturns