27 February 2025 in PNAS Nexus
In traditional decision-making processes, social biases of human decision makers can lead to unequal economic outcomes for underrepresented social groups, such as women and racial/ethnic minorities (1–4). Recently, the growing popularity of large language model (LLM)-based AI signals a potential shift from human to AI-based decision-making. How would this transition affect the distributional outcomes across social groups? Here, we investigate the gender and racial biases of a number of commonly used LLMs, including OpenAI's GPT-3.5 Turbo and GPT-4o, Google's Gemini 1.5 Flash, Anthropic AI's Claude 3.5 Sonnet, and Meta's Llama 3-70b, in a high-stakes decision-making setting of assessing entry-level job candidates from diverse social groups. Instructing the models to score ∼361,000 resumes with randomized social identities, we find that the LLMs award higher assessment scores for female candidates with similar work experience, education, and skills, but lower scores for black male candidates with comparable qualifications. These biases may result in ∼1–3 percentage-point differences in hiring probabilities for otherwise similar candidates at a certain threshold and are consistent across various job positions and subsamples. Meanwhile, many models are biased against black male candidates. Our results indicate that LLM-based AI systems demonstrate significant biases, varying in terms of the directions and magnitudes across different social groups. Further research is needed to comprehend the root causes of these outcomes and develop strategies to minimize the remaining biases in AI systems. As AI-based decision-making tools are increasingly employed across diverse domains, our findings underscore the necessity of understanding and addressing the potential unequal outcomes to ensure equitable outcomes across social groups.
25 September 2023 in Journal of Financial and Quantitative Analysis
Banks’ information technology (IT) capabilities affect their ability to serve customers during the COVID-19 pandemic, which generates an unexpected and unprecedented shock that shifts banking services from in-person to digital. Amid mobility restrictions, banks with better IT experience larger reductions in physical branch visits and larger increases in website traffic, implying a larger shift to digital banking. Stronger IT banks are able to originate more Paycheck Protection Program loans to small business borrowers, especially in areas with more severe COVID-19 outbreaks, higher internet use, and higher bank competition. Those banks also attract more deposit flows and receive better mobile customer reviews during the pandemic.
11 April 2023 in Journal of Financial and Quantitative Analysis
Adverse market events can affect credit supply not only by hurting financial fundamentals but also by changing the risk-taking behaviors of individual decision-makers. We provide micro-level evidence of this individual decision-making channel in the U.S. mortgage market. We find that mortgage application rejection rates are more sensitive to foreclosure intensity when loan officers are more exposed to foreclosure news, despite the same housing market and bank fundamentals. Loans originated from the affected branches have lower ex post default rates, consistent with higher lending standards being applied. In the aggregate, this effect results in tighter credit supply during housing market downturns.
1 March 2025 in Management Science
Research offers conflicting predictions about the impact of credit conditions on mental health. We first assess how bank regulatory reforms that improved credit conditions, for example, by enhancing the efficiency of credit allocation and lowering lending rates, impacted mental health. We discover that among low-income individuals, these regulatory reforms reduced mental depression, boosted labor market outcomes, eased access to mortgage debt, and reduced the ranks of the “unbanked.” We also find that mergers of large regional banks that led to branch closures and tighter credit constraints in affected counties harmed the mental health of lower-income individuals in treated counties. This paper was accepted by Kay Giesecke, finance. Funding: C. Lin and M. Tai acknowledge financial support from the National Natural Science Foundation of China [Grant 72192841] and the Research Grants Council of the Hong Kong Special Administration Region, China [Project T35/710/20R]. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2023.00194 .
8 June 2025 in Real Estate Economics
While housing market booms improve homeowners' credit access, they could constrain credit availability to renters. Using individual‐level consume credit data, I show that renters have poorer credit access if they live in locations where lenders are more exposed to housing booms and extend more mortgage lending to homeowners across the country. This suggests a reallocation of credit from renters to homeowners. As a result, renters end up using more expensive alternative credit, default more, and also experience lower consumption. This reallocation of credit during the boom also affect renters' financial well‐being in the subsequent bust.
23 January 2025 in Journal of Financial and Quantitative Analysis
We show that attention constraints on decision-makers create barriers to financial inclusion. Using administrative data on retail loan-screening processes, we find that attention-constrained loan officers exert less effort reviewing applicants of lower socioeconomic status (SES) and reject them more frequently. More importantly, when externally imposed increases in loan officers’ workloads tighten attention constraints, loan officers are even more prone to quickly reject low-SES applicants but quickly accept very high-SES applicants without careful review. Such selective attention allocation further widens the approval rate gap between high- and low-SES applicants—a unique prediction of this attention-based mechanism.