Research

New Working Papers & Work in Progress

Failing Banks, with Stephan Luck and Emil Verner.

Why do banks fail? We create a panel covering most commercial banks from 1863 through 2023 and study the history of failing banks in the United States. Failing banks are characterized by rising asset losses. Losses are typically preceded by rapid lending growth, financed by non-core funding. Bank failures, including those that involve depositor runs, are highly predictable based on bank fundamentals, even in the absence of deposit insurance and a central bank. We construct a new measure of systemic risk using bank-level fundamentals and show that it forecasts the major waves of banking failures in U.S. history. Altogether, our evidence suggests that failures caused by runs on healthy banks are uncommon. Rather, the ultimate cause of bank failures and banking crises is almost always and everywhere a deterioration of bank fundamentals.

The Debt-Inflation Channel of the German Hyperinflation, with Markus Brunnermeier, Stephan Luck, Emil Verner, and Tom Zimmermann (R&R AER).

Unexpected inflation can redistribute wealth from creditors to debtors. In the presence of financing frictions, such redistribution can impact the allocation of real activity. We use the German inflation of 1919-1923 to study how a large inflationary shock is transmitted to the real economy via a debt-inflation channel. In line with inflation reducing real debt burdens and relaxing financial constraints, we document a tight negative and convex relation between firm bankruptcies and inflation in aggregate data. Using newly digitized firm-level data, we further document a significant decline in leverage and interest expenses during the inflation. We show that firms that have more nominal liabilities at the onset of the inflation become more valuable in the stock market, face lower interest payments, and increase their overall employment once the inflation starts. The results are consistent with substantial real effects of the inflation through a financial channel that operates even when prices and wages are fully flexible.

Stock Market Milestones and Mortgage Demand: Evidence from the US, with Sumit Agarwal, Bernardo Morais, and Changcheng Song

We document that after the stock market index Dow Jones Industrial Average reaches crosses a milestone number (round a multiple of 1,000), more households apply mortgage for home purchase, and they are more likely to apply mortgage for second homes compared to non-milestone historical maxima. The loan amount also increases after the milestone event. These results are mainly driven by households with high equity holdings. The applicants are more likely to have high equity holdings and high FICO scores, but the mortgage loan terms are riskier with higher interest rate, LTV, and DTI, and the applicants are more likely to default in the two years after loan originations. We investigate the mechanisms and show that the effect is driven by the attention effect rather than the wealth effect. We also show that the results are unlikely to be driven by other events around the milestone or banks’ supply-side response. Our results highlight the linkage between the stock market and the housing market.

Banking

The Effects of Banking Competition on Growth and Financial Stability: Evidence from the National Banking Era, with Stephan Luck and Mark Carlson. Journal of Political Economy, February 2022. Also see preprint.

How does banking competition affect credit provision and growth? How does it affect financial stability? In order to identify the causal effects of banking competition, we exploit a discontinuity in bank capital requirements during the nineteenth-century National Banking Era. We show that banks operating in markets with lower entry barriers extend more credit. The resulting credit expansion, in turn, is associated with additional real economic activity. However, banks in markets with lower entry barriers also take more risk and are more likely to default. Thus, we provide causal evidence that banking competition can cause both growth and financial instability.

Did the Community Reinvestment Act (CRA) Lead to Risky Lending?, with S. Agarwal, B. Morais, E. Benmelech, N. Bergman and A. Seru

We use exogenous variation in banks’ incentives to conform to the standards of the Community Reinvestment Act (CRA) around regulatory exam dates to trace out the effect of the CRA on mortgage lending activity. Our empirical strategy compares lending behavior of banks undergoing CRA exams within a given census tract in a given month to the behavior of banks operating in the same census tract-month that do not face these exams. We find that adherence to the act led to riskier lending by banks: in the three quarters preceding the CRA exams, lending in CRA-eligible census tracts increases by about 0.8 percent every quarter and delinquency rates by about 7.5 percent. These patterns are accentuated among large banks and in banks with previous non-satisfactory CRA evaluations.

Credit Supply Shocks, Consumer Borrowing and Bank Competitive Response: Evidence from Credit Card Markets

I study local shocks to consumer credit supply arising from the opening of bank-related retail stores. Bank-related store openings coincide with sharp increases in credit card placements in the neighborhood of the store, in the months surrounding the store opening, and with the bank that owns the store. I exploit this relationship to instrument for new credit cards at the individual level, and find that obtaining a new credit card sharply increases total borrowing as well as default risk, particularly for risky and opaque borrowers.

In line with theories of default externality, I observe that existing lenders react to the increased consumer borrowing and associated riskiness by contracting their own supply. In particular, in the year following the issuance of a new credit card, banks without links to stores reduce credit card limits by 24–51%, offsetting most of the initial increase in total credit limits.

Economic History

Pandemics Depress the Economy, Public Health Interventions Do Not: Evidence from the 1918 Flu, with Stephan Luck and Emil Verner. Journal of Economic History, December 2022. Also see preprint.

We study the impact of non-pharmaceutical interventions (NPIs) on mortality and economic activity across U.S. cities during the 1918 Flu Pandemic. The combination of fast and stringent NPIs reduced peak mortality by 50% and cumulative excess mortality by 24% to 34%. However, while the pandemic itself was associated with short-run economic disruptions, we find that these disruptions were similar across cities with strict and lenient NPIs. NPIs also did not worsen medium-run economic outcomes. Our findings indicate that NPIs can reduce disease transmission without further depressing economic activity, a finding also reflected in discussions in contemporary newspapers.

Digitizing Historical Balance Sheet Data: A Practitioner’s Guide, with Stephan Luck. Explorations in Economic History, January 2023. Also see slides, preprint, and Github repo.

This paper discusses how to successfully digitize large-scale historical micro-data by augmenting optical character recognition (OCR) engines with pre- and post-processing methods. Although OCR software has improved dramatically in recent years due to improvements in machine learning, off-the-shelf OCR applications still present high error rates which limit their applications for accurate extraction of structured information. Complementing OCR with additional methods can however dramatically increase its success rate, making it a powerful and cost-efficient tool for economic historians. This paper showcases these methods and explains why they are useful. We apply them against two large balance sheet datasets and introduce quipucamayoc, a Python package containing these methods in a unified framework.

Applied Econometrics

Verifying the Existence of Maximum Likelihood Estimates for Generalized Linear Models, with Paulo Guimarães and Tom Zylkin (R&R Econometric Reviews)

A fundamental problem with nonlinear models is that maximum likelihood estimates are not guaranteed to exist. Though nonexistence is a well known problem in the binary choice literature, it presents significant challenges for other models as well and is not as well understood in more general settings. These challenges are only magnified for models that feature many fixed effects and other high-dimensional parameters. We address the current ambiguity surrounding this topic by studying the conditions that govern the existence of estimates for (pseudo-)maximum likelihood estimators used to estimate a wide class of generalized linear models (GLMs). We show that some, but not all, of these GLM estimators can still deliver consistent estimates of at least some of the linear parameters when these conditions fail to hold. We also demonstrate how to verify these conditions in models with high-dimensional parameters, such as panel data models with multiple levels of fixed effects.

Linear Models with High-Dimensional Fixed Effects: An Efficient and Feasible Estimator. Also see slides, website, and Github repo.

I propose a feasible and computationally efficient estimator of linear models with multiple levels of fixed effects. This estimator builds upon the generalized within-estimator of Guimarães and Portugal (2010) and Gaure (2013), addressing its slow convergence properties with two contributions. First, I replace their projection methods by symmetric ones amenable to conjugate gradient acceleration, which guarantees monotonic convergence. Second, I reformulate the within-transformation problem into one of solving a Laplacian system, and apply recent breakthroughs in spectral graph theory (Spielman and Teng 2004; Kelner et al 2013) to implement a nearly–linear time estimator. This estimator performs particularly well in the cases where the conjugate gradient method performs at its worst.

`ppmlhdfe`: Fast Poisson Estimation with High-Dimensional Fixed Effects, with Paulo Guimarães and Tom Zylkin. Stata Journal. 20(1), 95-115. 2020. Also see working paper, website, and Github repo.

In this paper we present ppmlhdfe, a new Stata command for estimation of (pseudo) Poisson regression models with multiple high-dimensional fixed effects (HDFE). Estimation is implemented using a modified version of the Iteratively Reweighted Least Squares (IRLS) algorithm that allows for fast estimation in the presence of HDFE. Since the code is built around the reghdfe package it has similar syntax and supports many of the same functionalities. ppmlhdfe also implements a novel and more robust approach to check for the existence of (pseudo) maximum likelihood estimates.

Technical Notes and other Research

`require`: Package dependencies for reproducible research, with Matthew P. Seay. Also see slides and Github repo.

The ability to conduct reproducible research in Stata is often limited by the lack of version control for user-submitted packages. This article introduces the require command, a tool designed to ensure Stata package dependencies are compatible across users and computer systems. Given a list of Stata packages, require verifies that each package is installed, checks for a minimum or exact version or package release date, and optionally installs the package if prompted by the researcher.

Singletons, Cluster-Robust Standard Errors and Fixed Effects: A Bad Mix

Maintaining singleton groups in linear regressions where fixed effects are nested within clusters can overstate statistical significance and lead to incorrect inference. Due to this problem, the reghdfe package now automatically drops singletons. However, a broader class of problems related to nested fixed effects and finite-sample adjustments remains.