Yupeng Lin

This paper exploits the 2014 U.S. Ebola outbreak as a natural experiment to examine how fear distorts mortgage credit allocation. We find a 3.8-percentage-point increase in mortgage denial rates for Black applicants during the outbreak, despite negligible infection risk. Approved Black borrowers have stronger credit profiles and lower default rates, yet are less likely to receive loan modifications when delinquent. These effects are absent in algorithm-driven FinTech lending and reverse quickly postoutbreak—patterns inconsistent with statistical discrimination. Robustness tests rule out alternative explanations related to political cycles, labor market impacts, applicant pool composition changes, and bank financial conditions. Consistent with diagnostic expectation theory, higher exposure to Ebola fear does not result in a significant learning effect during the COVID-19 pandemic. This paper was accepted by Bo Becker, finance. Funding: S. Agarwal acknowledges the financial support from the National Natural Science Foundation of China [Grant 72495154]. Y. Lin gratefully acknowledges financial support from the NUS Research Grant [Grants A-8001725 and A-8004426]. R. Lu and X. Zhang acknowledge the financial support of the National Natural Science Foundation of China [Grants 71903005, 72402005, and 72472001]. Supplemental Material: The online appendices and data files are available at https://doi.org/10.1287/mnsc.2025.01919 .

Despite persistent academic and regulatory concerns, collateralized loan obligations (CLOs) have consistently demonstrated resiliency over the past two decades. We document that firms whose loans are acquired by CLOs are less likely to experience adverse credit events within 12 months of inclusion in the CLO portfolio, particularly when CLOs maintain institutional-level relationships with originating banks (through repeated transactions) or when individual-level relationships exist (as evidenced by personnel flows between originating banks and CLOs). The effects of these relationships are more pronounced when banks likely possess superior private information about borrowers. Furthermore, conditional upon filing for bankruptcy, borrowers whose loans are held by CLOs with institutional- or individual-level relationships are more likely to successfully reorganize under Chapter 11. Collectively, our findings highlight the dual information channels, both institutional and personnel based, that mitigate information frictions in the leveraged loan market. Data Availability: Data are available from commercial sources cited in the text. JEL Classifications: M41; G32; G34; G12.

We study the effect of subsequent write-off tests on myopic managers’ investment decisions. Write-off tests discipline overinvestment when the likelihood of adverse events is high but can otherwise cause more severe underinvestment. Hence, tightening impairment stringency improves firm value when the likelihood of adverse events is high but destroys firm value when that likelihood is low. With intermediate levels of likelihood, firm value is hump-shaped in impairment stringency. To test the theory, we exploit an increase in goodwill impairment stringency following goodwill-related restatements by peer firms audited by the same auditor office, and we adopt a stacked difference-in-differences (DiD) design to control for any generic effects of peer restatements. Firms facing more stringent goodwill impairment reduce M&As relative to other types of investment, as well as in absolute amounts. The valuation of treated firms improves only during periods of high recession expectations or when facing high distress risk. Data Availability: Data are available from sources identified in the text. JEL Classifications: D82; G31; G34; M41; M42.

We examine how bank shareholder-level dividend tax affects banks' mortgage lending and housing cycles. The 2003 dividend tax cut lowered the cost of equity for C-corporation banks (C-banks) but not for pass-through S-subchapter banks (S-banks). Compared to S-banks within the same county, C-banks raise mortgage approvals by 2.3 percentage points, concentrated among risk-taking banks and risky applicants. These loans show higher delinquency and foreclosure during the 2007-2008 crisis. Counties with greater pre-reform Cbank presence experience stronger 2003-2005 house-price growth and sharper 2006-2008 declines. Our results suggest that the capital tax reform incrementally explains the housing boom-bust in the early 2000s.

Political power reaches markets not only through policy but also through official economic narratives. Using 6,327 videos of Chinese local officials from 2015 to 2025, we apply machine-learning methods to extract narrative themes and sentiment from words, voice, and facial expressions. The extracted themes track national initiatives and changing local priorities, supporting the measures' validity and documenting shifts in China's local political-economic landscape. Further, we find that more positive sentiment is associated with lower LGFV bond spreads at the first secondary-market trade, but not at issuance. Both vocal and visual sentiment drive this relation. The initial spread compression later reverses, and land prices show a similar riseand-reversal pattern. These findings suggest that official economic narratives temporarily shift market beliefs, providing a channel through which political power enters asset prices.

We exploit the public release of ChatGPT 3.5 as an informational technology shock to examine how firms adjust their intellectual property protection strategies in response to increased replication risk from generative AI. Using a difference-indifferences design, we find that firms with high exposure to AI replication risk increase patent filings by 15.7% relative to less exposed firms. The effect is concentrated among firms with pre-existing trade secrets, firms in states without the Inevitable Disclosure Doctrine, firms with prior trade secret litigation, and in industries where patents offer more effective protection. To directly capture secrecy-topatent conversion, we develop a novel measure of trade-secret-converted patents and document a significant post-ChatGPT increase in such patent filings among treated firms. We find no accompanying rise in R&D, capital investment, or innovation efficiency, and no evidence that the response reflects reduced filing frictions. Finally, the market reacts positively to these filings, suggesting that investors view the shift toward patenting as a value-enhancing response to AI-driven imitation risk.

We study the effect of subsequent write-off tests on myopic managers' investment decisions. Write-off tests discipline overinvestment when the likelihood of adverse events is high, but can otherwise cause more severe underinvestment. Hence, tightening impairment stringency improves firm value when the likelihood of adverse events is high, but destroys firm value when that likelihood is low. With intermediate levels of likelihood, firm value is hump-shaped in impairment stringency. To test the theory, we exploit an increase in goodwill impairment stringency following goodwill-related restatements by peer firms audited by the same auditor office, and we adopt a stacked difference-indifferences design to control for any generic effects of peer restatements. Firms facing more stringent goodwill impairment reduce M&As relative to other types of investment, as well as in absolute amounts. The valuation of treated firms improves only during periods of high recession expectations or when facing high distress risk.

We run a large-scale randomized controlled trial (RCT) with 2,722 Chinese listed firms to examine how perceived AI bots shape managers' disclosures and market reactions. In online earnings conferences (OECs), we randomize the questioner's identity as either human or AI while holding information content fixed. Managers are 13% less likely to answer AI-styled questions, with stronger effects when an image signals AI identity. Linking manager and firm traits to responses shows smaller declines at state-owned enterprises and larger declines for managers who are older, or have a computer science background. Conditional on responding, managers provide longer answers with more neutral narratives to AI-styled questions. Using high-frequency data, we find that investors withhold trading during AI-question windows. Overall, our results are consistent with AI playing a disciplinary role on voluntary disclosure.

We examine whether accounting-based signals shape bank supervision. SFAS 142 mandated a shift from goodwill amortization to annual impairment testing. We exploit this adoption as a quasi-natural experiment and find that bank regulators are 3.1 percentage points more likely to issue Enforcement Decisions and Orders (EDOs) to banks with goodwill after the adoption. The effect is stronger when regulators face tighter resource constraints and when banks attract greater public scrutiny. We further document a reallocation of regulatory attention: post-adoption, regulators appear to shift attention to goodwill by issuing more EDOs to highgoodwill banks with strong capital positions. Machine learning analysis of EDO texts confirms that regulators place greater emphasis on asset quality, management, and earnings after the adoption. These are the dimensions most informationally tied to goodwill impairment. Despite this reallocation, regulatory actions become less pro-cyclical, and treated banks exhibit lower risk and failure rates during the 2007-2009 financial crisis. These findings demonstrate that impairment accounting measuring bank goodwill generates actionable forward-looking signals that enhance the efficacy of prudential bank supervision.

This study examines the spillover effect of bank financial misconduct on the uninsured deposits of peer banks within local markets. We first validate that misconduct banks experience an increase in deposit spreads and a corresponding outflow of deposits following the misconduct. We then show local peer banks exhibit divergent deposit responses, contingent on how misconduct is perceived by information recipients in different economic contexts. During normal periods, depositors receiving a negative signal about bank misconduct reallocate their funds from misconduct banks to local peers, a local reallocation effect that decreases deposit spreads and increases deposit inflows for peer banks. Cross‐sectional analysis further reveals that this local reallocation effect is more pronounced for financially sophisticated depositors, amplified when peer banks have strong fundamentals, but attenuated when misconduct banks are financially sound. During financial crisis periods, however, bank misconduct leads to withdrawals from both misconduct banks and their peer banks, a local contagion effect whereby local peer banks face increased deposit spreads and deposit outflows following the misconduct.