We conduct the first study exploring the application of large language models (LLMs) to mortgage underwriting, using an audit study design that combines real loan application data with experimentally manipulated race and credit scores. First, we find that LLMs systematically recommend more denials and higher interest rates for Black applicants than otherwise-identical white applicants. These racial disparities are largest for lower-credit-score applicants and riskier loans, and exist across multiple generations of LLMs developed by three leading firms. Second, we identify a straightforward and effective mitigation strategy: Simply instructing the LLM to make unbiased decisions. Doing so eliminates the racial approval gap and significantly reduces interest rate disparities. Finally, we show LLM recommendations correlate strongly with real-world lender decisions, even without fine-tuning, specialized training, macroeconomic context, or extensive application data. Our findings have important implications for financial firms exploring LLM applications and regulators overseeing AI's rapidly expanding role in finance.

We document significant deposit interest rate differentials along the income distribution

• moving from the bottom to the top income decile increases deposit rates by 55%

of the sample median rate. These spreads persist independent of banking competition,

and instead appear to arise from banks internalizing households’ participation

in nondeposit markets. Consistent with this hypothesis, only income components related

to participation can explain our baseline findings, and quasi-exogenous reductions

in participation incentives through increases in capital gains taxes are associated

with lower spreads along the participation distribution. Our findings highlight lack

of participation as a source of deposit market power.

Using loan-level data from the Brazilian credit registry, this paper investigates whether the pass-through of monetary policy to consumer credit is heterogeneous along the income distribution. We find three novel results on how monetary policy affects different consumers' credit costs. Firstly, the pass-through of monetary policy to consumer interest rates is stronger for lower-income borrowers than for higher earners. Secondly, we decompose the results into a direct heterogeneity effect and portfolio composition channels. We show that the direct heterogeneity channel is operational, implying that pass-through of monetary policy would still be higher to lower-income individuals even if all borrowers had identical loan portfolios. Thirdly, we show that this pass-through heterogeneity is asymmetric between periods of monetary loosening and tightening. During the post-Covid tightening cycle, the pass-through of monetary policy hikes to consumers' borrowing costs was stronger for individuals with lower incomes. Conversely, during the previous loosening cycle, the pass-through of cuts to lower earners was weaker than for higher earners. This partial equilibrium result could therefore shed light on a new channel whereby monetary policy exacerbates inequality through consumers' borrowing costs.