My research focuses on banking, labor and finance, FinTech, and machine learning. I am interested in how regulation, labor market frictions, and new technologies shape the behavior of financial institutions and affect credit allocation, pricing, and risk-taking. Much of my work uses empirical methods to study questions in banking and household finance, with a particular focus on policy-relevant issues and real-world applications. I am also interested in how data-driven methods can be used to better understand financial markets and decision-making.
This paper examines whether consumer complaint narratives contain early information about bank deposit fragility. Using the CFPB Consumer Complaint Database matched to FDIC bank financial data, I construct a transparent text-based measure of trust-loss intensity that captures complaint language about frozen funds, account access, payment reliability, fraud-related missing money, and other failures in depositor-facing services. I find that trust-loss complaint language predicts weaker future deposit growth beyond complaint counts and standard bank characteristics. A one-standard-deviation increase in trust-loss intensity predicts 0.67 percentage points lower deposit growth in the next quarter. The relation is stronger for banks with thinner liquidity buffers, greater deposit dependence, more loan-intensive balance sheets, and fragile banks during the 2023 banking-stress period. The mechanism is most visible in frozen-funds complaints, where operational failures directly interfere with consumers’ access to money. Public enforcement shocks, operating-capacity tests, retail deposit pricing responses, local deposit-market evidence, and nonbank payment-provider complaints all support the interpretation that complaint narratives reveal deterioration in depositor-facing trust before it is fully reflected in regulatory deposit data. Overall, the evidence suggests that public complaint text provides a high-frequency liability-side signal of bank funding fragility.
Presented at the FMA Annual Conference 2025, Community Bank Research Conference 2025 (poster)
This paper examines how the Community Bank Leverage Ratio (CBLR) affects the behavior of small U.S. community banks. By allowing eligible banks to replace risk-weighted capital requirements with a simple leverage-based constraint, the CBLR removes the regulatory price of risk while tightening the link between capital requirements and total assets. Using a difference-in-differences design with propensity score matching, I find that CBLR adoption increases Tier 1 leverage ratios primarily through asset contraction rather than equity growth. Banks largely preserve lending, but take on greater risk, as reflected in higher nonperforming loans and charge-offs. They also adjust along the pricing margin by charging higher loan rates and offering lower deposit rates. At the local level, adoption is associated with changes in the volume and composition of CRA and small business lending. Despite its stated purpose, however, CBLR does not appear to reduce regulatory costs. Overall, the evidence suggests that simplifying capital regulation weakens risk sensitivity and induces riskier behavior without delivering measurable efficiency gains.
“When Talent Walks: Skilled Labor Mobility and Bank Behavior”, June 12, 2025, International Banker.
Presented at the FMA Annual Conference 2025, FMA Asia/Pacific Conference 2025 (coauthor), UT-Dallas 2025 (coauthor)
Semifinalist for the Best Paper Award in Financial Institutions and Markets at FMA 2025
We investigate how traditional banks react to a shock in high-skilled labor mobility that induces more labor turnover, raises the bargaining power of skilled labor, and translates into higher costs for the banks. Using the natural experiment of the U.S. state courts’ staggered rejections of the Inevitable Disclosure Doctrine (IDD), we document that the rejections of IDD increase labor mobility and wages of skilled workers, which translates to a higher operating cost and lower operating efficiency for traditional banks. This induces banks to increase the loan rates and selectively loan offering without impacting their riskiness. The impact of employee labor mobility on loan rates is more prominent among the banks whose clients have fewer outside options, banks with stronger bargaining power, and banks with a higher likelihood of employee turnover. Neither banks nor customers do seem to enjoy the benefit of such mobility, and the rent of high skilled labor is appropriated by the employees through higher salaries. This has important normative implications for the regulator.
This paper studies how competition for workers affects firm performance. Using resume-level worker transitions among U.S. public firms from 2008 to 2022, I apply weighted directed Node2Vec embeddings to recover firms’ latent talent-market competitors. The resulting measure combines talent-market proximity, realized poaching relationships, and abnormal hiring by rival firms to identify periods of elevated talent-defense pressure. Firm-year estimates show that increased hiring by demonstrated talent poachers predicts significantly lower subsequent operating and EBITDA margins. The evidence suggests that competitors’ hiring imposes retention, replacement, and organizational adjustment costs on exposed firms. The paper introduces an embedding-based approach that captures economically relevant labor-market competition beyond conventional industry classifications and observed bilateral worker flows.
We study whether banks’ artificial intelligence (AI) capabilities shape syndicated lending. Using a resume-based bank AI index from 2005 to 2024, we find that higher-AI banks originate more syndicated loans and larger volumes, especially as lead arrangers. Loans arranged by higher-AI banks have lower spreads: a one-standard-deviation increase in the lead-arranger AI index is associated with an 8.9-basis-point reduction, alongside lower future borrower default risk. Estimates using employee-weighted cyberattack exposure as an instrumental variable show similar effects on loan origination and spreads. Cross-sectional results are stronger for non-relationship loans, longer maturities, and less seasoned borrowers. Higher-AI banks also rely more on covenants that necessitate stronger information processing and intervention, and less on rigid clauses that mechanically protect lenders. The evidence suggests AI expands bank intermediation capacity by reducing information-processing and monitoring frictions.
Liu, R. (2019). Comparison of Bank Efficiencies between the US and Canada: Evidence Based on SFA and DEA. Journal of Competitiveness, 11(2), 113–129. https://doi.org/10.7441/joc.2019.02.08