Do generative AI models, as epitomized and popularized by large language models (LLMs), exhibit systematic behavioral biases in economic and financial decisions? If so, how can we mitigate these biases? Following the cognitive psychology literature and the experimental economics studies, we conduct the most comprehensive set of experiments to date---originally designed to document human biases---on prominent LLM families with variations in model version and scale. We document systematic patterns in the behavioral biases that LLMs exhibit. For experiments concerning the psychology of preferences, LLM responses become increasingly irrational and human-like as the models become more advanced or larger; however, for experiments concerning the psychology of beliefs, the most advanced large-scale models frequently generate rational responses. Further exploring various methods for correcting these behavioral biases reveals that prompting LLMs to make rational decisions according to the Expected Utility framework seems the most effective.

We compare retail option trade execution by placing simultaneous market orders across six brokers. Although option trades are all anonymously executed on exchanges and therefore should be treated equally, we find that execution prices vary significantly: the average round-trip execution cost ranges from 0% to 7% across brokers. Wholesalers create differential pricing by not only systematically varying execution methods, but also the pricing within each method. A primary economic driver for differential pricing seems to be payment for order flow (PFOF). However, specialists affiliated with and without PFOF-paying wholesalers provide similar pricing. Our results have market design and disclosures implications.

Using 150,000 actual trades, we study the U.S. equity retail broker-wholesaler market, focusing on brokers’ order routing and competition among wholesalers. We document substantial and persistent dispersion in execution costs across wholesalers within brokers. Despite this, many brokers hardly change their routing and even consistently send more orders to the more expensive wholesalers, although there is considerable variation among brokers. We also document a case where, after a new wholesaler enters, existing wholesalers significantly reduce their execution costs. Overall, our findings and theoretical framework highlight the heterogeneity across brokers and are inconsistent with perfect competition in this market.

Using a novel setting in which retailers receive bonuses when selling jackpot winning lottery tickets, we show that large windfalls lead to both existing business expansion and new business creation. New ventures are larger and have high survival rates; they tend to emerge in nonretail industries, substituting for existing business expansion. We also show that high-quality owners who are financially constrained respond the most to cash windfalls. Our findings contrast with the prevailing view that small businesses lack the desire to grow and highlight that financial frictions not only impede growth but also limit industry choices for constrained entrepreneurs.

In theory, investors who have low security selection ability trade more, use leverage more, and perform worse if they are overconfident. We confirm these predictions empirically by analyzing the overconfidence, trading, and performance of retail investors who use margin. Using survey data, we measure overconfidence as the difference between an investor’s self-assessment of knowledge and tested knowledge; margin investors have greater overconfidence than cash investors. Using broker data, we find margin investors trade more, speculate more, and have worse security selection ability than cash investors. A long-short portfolio that follows the trades of margin investors loses 35 bps per day.

This paper studies the role of race and ethnicity in the investment decisions of mutual fund managers and mutual fund investors. Mutual funds managed by a white-dominant team account for more than 90% of all funds. Such funds invest a smaller proportion of their portfolios in firms led by minority CEOs compared to funds managed by a minority-dominant team. Fund managers do not deliver superior performance on equity holdings for which the CEO's race coincides with their own, suggesting no race-related informational advantage. Considering flow-performance sensitivity, we find that funds managed by a minority-dominant team are equally penalized when they perform poorly but are not rewarded as much for superior performance as white-dominant funds. Our results uncover the differential treatment of minority-led funds and firms by investors.

We study risk aversion (RA) spillovers from the US to several major devel- oped economies. In our observational study, we use financial market and news data from 2000 to 2017 to identify US events that lead to extreme changes in an option-based risk premium measure but not in uncertainty, which we label as "risk aversion events." The international pass-through of US high RA events (61%) is significantly higher than that of US low RA events (43%), suggesting an asymmetric risk aversion spillover. In our controlled experiment, non-US subjects, when primed with a US financial bust shock, exhibited asymmetrically larger changes in risk aversion and general emotions than those primed with a US boom shock. The “foreign” nature of bust shocks may change emotions more than boom shocks, due to unfamiliarity. This non-fundamental emotion-based channel explained 20% of the RA spillover asymmetry in our experiment.

We study how the existence of cutoffs in credit scores affects the behavior of homebuyers. Borrowers are more likely to purchase houses after their credit scores cross over a cutoff to qualify them for a higher credit score bin. However, the credit accounts of these individuals (crossover group) are more likely to become delinquent within four years following home purchases than the accounts of those who had stayed in the same bin (non-crossover group). The effect is not only concentrated in subprime bins, but in other bins as well. It is neither limited to pre-crisis period nor curtailed by post-bust reforms. Using recent house price growth to proxy for the incentives for home purchases, we find that the gap in the delinquency rates between crossover and non-crossover groups is larger for areas with higher recent house price growth. Overall, our results indicate that the credit score at the time of home purchase may not be sufficiently informative because of individuals’ strategic behavior, and suggest the importance of using the longer history of credit scores rather than just the latest draw in making lending decisions.

We study the impact of the linguistic content of index fund prospectuses on investor demands. Even after controlling for lagged fund returns, fees and fund family effects, we find a significant association between positive sentiment words and investor fund flows. The effect is concentrated in broker-sold flows, where the use of positive sentiment words is associated with a 0.0074% increase in monthly fund flows, which equates to $9.78M monthly at the mean broker fund size. In contrast, direct-sold flows respond negatively to the use of positive sentiment words. In addition, funds that use positive sentiment words have higher expenses than funds without positive sentiment words. These findings suggest that investors relying on brokers are talked into buying expensive funds with positive sentiment words; these funds do not outperform funds without positive sentiment words. Using an online survey that controls for unobserved contractual features, we confirm that the use of positive sentiment words drives the diverging prospectus selections of financially sophisticated and unsophisticated investors.