Marianne Andries, University of Southern California
Maxime Bonelli, London Business School
David Sraer, University of California-Berkeley
Abstract: Can financial advisors mitigate their clients’ investment biases? We answer this question by exploiting a natural experiment at a large brokerage firm that provides advisory services to high-net-worth investors. In 2018, the firm changed the information displayed on its internal platform so that financial advisors could no longer observe which of their clients’ holdings were in paper gain or loss. Using data on portfolio stock transactions between 2016 and 2021, we show that, while all investors exhibit a significant disposition effect before 2018, i.e., a greater propensity to realize paper gains than losses, highly-advised investors see their bias significantly reduced after 2018. This decrease in disposition effect bias leads to higher portfolio returns, increased client inflow, and a lower likelihood of leaving the firm. Our study highlights how manipulating advisors’ information can help mitigate investors’ biases.
Antoinette Schoar, Massachusetts Institute of Technology
Yang Sun, Brandeis University
Abstract: In a randomized controlled trial we test how retail investors assess different types of financial advice and update their priors based on the advice. We either emphasize the benefits of passive investment strategies (such as diversification and low fees) or those of active strategies (such as stock picking and market timing). First, we find that participants rate advice significantly higher when it is aligns with their priors rather than contradicting them. Second, people update their beliefs about investment strategies toward the advice they receive. Third, there is significant heterogeneity based on the type of advice and the subjects' financial literacy. Financially more literate subjects positively update in response to seeing the passive advice, but do not update (and rate the advice negatively) when exposed to active advice. In contrast, financially less literate subjects are strongly influenced by both types of advice. Finally, we show that subjects rate the advice lower if the advisor's compensation presents a potential conflict of interest (commission-based pay) compared to when it is more aligned (flat fee).
Discussant: Olivia Mitchell, University of Pennsylvania
Abstract: Digital technologies and fintech firms have rapidly reshaped the consumer financial landscape in recent years, and have the potential to help consumers make better decisions and improve their financial health. Existing technologies such as autopay are also experiencing increased takeup, a trend that could be accelerated by innovations such as open banking. I examine the extent to which autopay affects payment behavior for customers of a credit card serviced by a fintech company. Using sharp changes in the company's practices in a regression discontinuity design, I find that a small nudge accounts for half of all autopay enrollment during the sample period, and that enrollment at account opening is persistent. Autopay increases the likelihood of making the minimum payment by 20 to 29pp, more than doubling the baseline rate. The results show that seemingly minor technological defaults can have economically large effects on consumer outcomes.
Discussant: Taha Choukhmane, Massachusetts Institute of Technology
Abstract: How good is artificial intelligence (AI) at tasks that require persuading humans to perform costly actions? We study the effectiveness of phone calls made to persuade delinquent consumer borrowers to repay their debt. Both a regression discontinuity design and a randomized experiment reveal that AI is substantially less able than human callers to get borrowers to repay. Substituting human callers for AI six days into delinquency closes much of the collection gap, but one year later, borrowers initially assigned to AI and then switched to humans have repaid 1% less than borrowers who were called by humans from the beginning. Even accounting for wage costs and assuming zero costs for AI, using AI is less profitable (with the caveat that we do not observe non-wage costs of labor). AI’s lesser ability to handle complex situations and extract payment promises may contribute to the performance gap.
Discussant: Anthony DeFusco, University of Wisconsin-Madison
