LSLuke C. D. Stein

Luke C. D. Stein

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Luke C. D. Stein has a total of 19 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Generative AI and the Evolution of Mortgage Underwriting: Measuring and Mitigating Racial Disparities

    by Donald E. Bowen, S. McKay Price, Luke C. D. Stein, Ke Yang

    Mortgage underwriting is entering a new phase of technological change as general-purpose generative AI tools begin to appear in mortgage origination workflows. We develop an audit framework to evaluate how large language models process mortgage applications. Using real HMDA applications, we experimentally manipulate borrower race, race proxies, and credit quality, and prompt leading commercial LLMs to make approval and pricing recommendations. LLMs recommend more denials and higher interest rates for Black applicants than otherwise identical white applicants, with disparities largest for riskier loans and persisting when race is signaled through names or locations. A simple instruction to avoid bias substantially reduces disparities when race is explicit, partially reduces the name-based pricing gap, and does little against the geographic pricing gap. In matched HMDA–Freddie Mac data, race-blind LLM recommendations correlate with actual lender pricing and subsequent delinquency, though the delinquency association operates largely through observed credit scores. The results show how AI auditing can inform mortgage underwriting governance before general-purpose models become more deeply embedded in origination workflows.

  • Open Access

    The Gendered Impacts of Perceived Skin Tone: Evidence from African American Siblings in 1870-1940

    by Ran Abramitzky, Jacob Conway, Roy Mill, Luke C.D. Stein

    We study differences in economic outcomes by perceived skin tone among African Americans using full-count U.S. decennial census data from the late-19 th and early-20 th centuries. Comparing children coded as "Black" or "Mulatto" by census enumerators and linking these children across population censuses, we first document large gaps in educational attainment and income among African Americans with darker and lighter perceived skin tones. To disentangle the drivers of these gaps, we identify all 35, 992 families in which enumerators assigned same-gender siblings different Black/Mulatto classifications. Relative to sisters coded as Mulatto, sisters coded as Black had lower educational attainment, were less likely to marry, and had less-educated, likely lower-earning husbands. These patterns are consistent with more severe contemporaneous discrimination against African American women with darker perceived skin tones. In contrast, we find similar educational attainment, marital outcomes, and incomes among differently-classified brothers. Men perceived as African Americans of any skin tone faced similar contemporaneous discrimination, consistent with the "one-drop" racial classification rule that grouped together individuals with any known Black ancestry. Lower incomes for African American men perceived as having darker skin tone in the general population were driven by differences in opportunities and resources that varied across families, likely reflecting the impacts of historical or family-level discrimination.

  • Open Access

    Sector Index Selection in Securities Litigation Event Studies: R-Squared and Model Stability

    by Steven P. Feinstein, Luke C.D. Stein, Miguel Villanueva

    An event study regression estimates the return a stock would have generated in the absence of company-specific news. The sector index in the regression model can affect the predicted return and thus the measured abnormal return and statistical significance. In a securities litigation context, the sector choice can thus affect determinations of market efficiency, price impact, loss causation, and damages. We examine one narrow selection rule: choosing from a set of candidate sector indices the model with the highest in-sample R-squared, a measure of fit. We compare that choice with the sector designation made by S&P Dow Jones Indices, which serves as an external benchmark (though not a uniquely "correct" event-study control). For 30 Dow Jones Industrial Average firms, the R-squared-maximizing index differs from the S&P Dow Jones-designated index for 15 firms in 2023 and 18 firms in 2024. Restricting the comparisons to each firm's designated index and just one predetermined, plausible alternate reduces disagreement, and the same candidate has higher R-squared in both years for 25 of 30 firms. By contrast, when the candidate set comprises all ten sector indices, the index with the highest R-squared changes between years for 18 firms, and is unstable across repeated random holdout periods. These results show that, while R-squared can be an informative metric, it should not be relied upon exclusively. A defensible process should emphasize economic reasoning. When fit is considered, the candidate set should be limited and disclosed, mechanical own-stock contamination should be removed, and the fit should be evaluated out-of-sample and for stability.

  • Racial Discrimination in Asset Prices: Evidence From Horse Betting

    by Spencer Barnes, Luke C. D. Stein

    In the presence of behavioral biases, prices can diverge from fundamentals, and the effects of racial/ethnic bias are evident in many financial and nonfinancial markets. We investigate the determinants and consequences of discrimination in parimutuel horse betting by assessing return differences across horses whose trainers have racially/ethnically distinctive surnames, which bettors may see as a proxy for quality (accurately or inaccurately) or a source of nonpecuniary returns (due to animus). Bets on horses with non‐White‐named trainers earn higher risk‐adjusted returns, and these differences are especially pronounced among riskier bets, which receive lower average returns under the well‐known “favorite–longshot” bias. Racial/ Ethnic return differences are stronger—overall and especially among longshots—for “win” than “place” and “show” bets, among horses with poor prior performance, and in low‐stakes races with “fast” conditions. These results are consistent with the effects of discrimination being strongest among less‐informed and less‐sophisticated bettors.

  • Open Access

    Racial discrimination in asset prices: Evidence from horse betting

    by Spencer Barnes, Luke C.D. Stein

    In the presence of behavioral biases, prices can diverge from fundamentals, and the effects of racial/ethnic bias are evident in many financial and non-financial markets. We investigate the determinants and consequences of discrimination in parimutuel horse betting by assessing return differences across horses whose trainers have racially/ethnically distinctive surnames, which bettors may see as a proxy for quality (accurately or inaccurately) or a source of non-pecuniary returns (due to animus). Bets on horses with nonwhite-named trainers earn higher risk-adjusted returns, and these differences are especially pronounced among riskier bets, which receive lower average returns under the well-known “favorite–longshot” bias. Racial/ethnic return differences are stronger—overall and especially among longshots—for “win” than “place” and “show” bets, among horses with poor prior performance, and in low-stakes races with “fast” conditions. These results are consistent with the effects of discrimination being strongest among less-informed and less-sophisticated bettors.


  • Open Access

    Measuring and Mitigating Racial Disparities in LLMs: Evidence from a Mortgage Underwriting Experiment

    by Donald E. Bowen Iii, S. McKay Price, Luke C.D. Stein, Ke Yang

    We evaluate LLM responses to a mortgage underwriting task using real loan application data. Experimentally manipulated race is signaled explicitly or through borrower name/location proxies. Multiple generations of LLMs recommend more denials and higher interest rates for Black applicants than otherwise-identical white applicants, with larger disparities for riskier loans. Simple prompt engineering can cost-effectively mitigate these patterns. Race-blind recommendations correlate strongly with real lender decisions and predict delinquency, but LLMs incorporate racial signals when available despite similar delinquency rates across groups. Our findings show potential costs of adopting this new technology in financial settings and raise important questions for regulators.


  • Open Access

    The Gendered Impacts of Perceived Skin Tone: Evidence from African-American Siblings in 1870–1940

    by Ran Abramitzky, Jacob Conway, Roy Mill, Luke C.D. Stein

    We study differences in economic outcomes by perceived skin tone among African Americans using full-count U.S. decennial census data from the late-19th and early-20th centuries. Comparing children coded as "Black" or "Mulatto" by census enumerators and linking these children across population censuses, we first document large gaps in educational attainment and income among African Americans with darker and lighter perceived skin tones. To disentangle the drivers of these gaps, we identify all 36,329 families in which enumerators assigned same-gender siblings different Black/Mulatto classifications. Relative to sisters coded as Mulatto, sisters coded as Black had lower educational attainment, were less likely to marry, and had lower-earning, less-educated husbands. These patterns are consistent with more severe contemporaneous discrimination against African-American women with darker perceived skin tones. In contrast, we find similar educational attainment, marital outcomes, and incomes among differently-classified brothers. Men perceived as African Americans of any skin tone faced similar contemporaneous discrimination, consistent with the "one-drop" racial classification rule that grouped together individuals with any known Black ancestry. Lower incomes for African-American men perceived as having darker skin tone in the general population were driven by differences in opportunities and resources that varied across families, likely reflecting the impacts of historical or family-level discrimination.

  • Financial Inclusion, Human Capital, and Wealth Accumulation: Evidence from the Freedman’s Savings Bank

    by Luke C D Stein, Constantine Yannelis

    This paper studies how access to financial services among a previously unbanked group affects human capital, labor market, and wealth outcomes. We use novel data from the Freedman’s Savings Bank—created following the American Civil War to serve free Blacks—employing an instrumental variables strategy exploiting the staggered rollout of bank branches. Families with accounts are more likely to have children in school, be literate, work, and have higher occupational income, business ownership, and real estate wealth. Placebo effects are not present using planned but unbuilt branches, or for Whites, suggesting significant positive effects of financial inclusion.


  • Open Access

    Economic Uncertainty and Earnings Management

    by Luke C.D. Stein, Charles C. Y. Wang

    In the presence of managerial short-termism and asymmetric information about skill and effort provision, firms may opportunistically shift earnings from uncertain to more certain times. We document that firms report more negative discretionary accruals when financial markets are less certain about their future prospects. Stock-price responses to earnings surprises are moderated when firm-level uncertainty is high, consistent with performance being attributed more to luck rather than skill and effort, which can create incentives to shift earnings toward lower-uncertainty periods. We show that the resulting opportunistic earnings management is concentrated in CEOs, firms, and periods where such incentives are likely to be strongest: (1) where CEO wealth is sensitive to change in the share price, (2) where announced earnings are particularly likely to be an important source of information about managerial ability and effort, and (3) before implementation of Sarbanes-Oxley made opportunistic earnings management more challenging. Our evidence highlights a novel channel through which uncertainty affects managerial decision making in the presence of agency conflicts.