Delays in Banks’ Loan Loss Provisioning and Economic Downturns: Evidence from the U.S. Housing Market

• Published date: 01 June 2022
• Author: SEHWA KIM
• Date: 01 June 2022
• DOI: http://doi.org/10.1111/1475-679X.12415

DOI: 10.1111/1475-679X.12415

Journal of Accounting Research

Vol. 60 No. 3 June 2022

Printed in U.S.A.

Delays in Banks’ Loan Loss

Provisioning and Economic

Downturns: Evidence from the U.S.

Housing Market

SEHWA KIM∗

Received 30 May 2019; accepted 3 November 2021

ABSTRACT

I study whether banks’ loan loss provisioning contributes to economic down-

turns, by examining the U.S. housing market. Specifically, I examine the ag-

gregate effects of banks’ delayed loan loss recognition (DLR) on house prices

during the Great Recession and the channels through which these potential

effects arose. I construct ZIP-code-level exposure to banks’ DLR before the

∗Columbia University

Accepted by Luzi Hail. This paper is based on my dissertation. I am grateful to my disser-

tation committee members for their guidance and support: Philip G. Berger (chair), John

Gallemore, Christian Leuz, and Valeri Nikolaev. I also thank two anonymous reviewers, Ray

Ball, John Barrios, Thomas Bourveau, Matthias Breuer, Jung Ho Choi, Hans B. Christensen,

Rachel M. Geoffroy, João Granja, Seil Kim, Anya Kleymenova, Mark G. Maffett, Charles Mc-

Clure, Michael Minnis, Maximilian N. Muhn, Shiva Rajgopal, Stephen G. Ryan, Haresh Sapra,

Douglas J. Skinner, Robert Stoumbos, George P. Surgeon, Rimmy E. Tomy, Eric Zwick, and

seminar participants at the 2018 CMU Accounting Mini Conference, University of Chicago,

Rice University, University of California Berkeley,University of North Carolina, Massachusetts

Institute of Technology, Tulane University, Columbia University, and Harvard University for

their helpful comments. I gratefully acknowledge the Fama-Miller Center for Research in

Finance and the Initiative on Global Markets at the University of Chicago Booth School of

Business for providing the DataQuick data, Eric Zwick for providing access to the DataQuick

database in his personal server, Gauri Bhat for the loan-type allowances data, and financial

support from Columbia Business School. All errors are mine. An online appendix to this

paper can be downloaded at http://research.chicagobooth.edu/arc/journal-of-accounting-

research/online-supplements.

711

© 2021 The Chookaszian Accounting Research Center at the University of Chicago Booth School of

Business.

712 s. kim

crisis and compare high- and low-exposure ZIP codes during the crisis to ex-

amine the aggregate effects of banks’ DLR on the housing market. I find

that high-exposure ZIP codes experienced larger decreases in mortgage sup-

ply, larger increases in distressed sales, and larger decreases in house prices

during the crisis. In addition, I conduct individual bank-level analyses and

find that high-DLR banks reduced their mortgage supply more than low-DLR

banks, and mortgages issued by high-DLR banks were more likely to become

distressed during the crisis. Takentogether, these findings suggest that banks’

DLR was associated with nontrivial effects on the housing market during the

Great Recession, and the effects of DLR on house prices were likely driven by

both the credit-crunch and distressed-sales channels.

JEL codes: E21, G01, G21, M41, R20

Keywords: real effects of accounting; banks; loan loss provisioning; housing

market; crisis

1. Introduction

The Great Recession of 2007−2009 sparked debate about accounting’s role

in financial stability and economic cycles. The timeliness of banks’ loan loss

provisioning is one of the most important issues in the debate, and its po-

tential real effects have received considerable attention from bank regula-

tors and central bankers for several reasons. Loan loss provisioning is the

largest accrual in bank accounting, and estimating the amount and tim-

ing of loan loss provisioning can involve significant managerial discretion

(Wahlen [1994], Liu and Ryan, [1995, 2006], Ahmed, Takeda and Thomas,

[1999], Jayaraman, Schonberger and Wu, [2019], Bischof, Laux and Leuz,

[2021]). In addition, delays in loan loss provisioning may reinforce pro-

cyclicality in banks’ lending and weaken market discipline over banks’

risk-taking (Laeven and Majnoni, [2003], Dugan, [2009], Beatty and Liao,

[2011], Bushman and Williams, [2012], 2015; Acharya and Ryan, [2016],

Wheeler, [2019]). Existing evidence on the effects of loan loss provision-

ing focus on bank-level behavior (e.g., Beatty and Liao, [2011], Bushman

and Williams, [2015]), but the individual effects might not be sufficient to

capture the economy-wide effects, due to substitutions and spillovers. Thus,

despite the centrality of the issue, whether loan loss provisioning led to sig-

nificant economy-wide effects during the crisis remains an open question.

In this paper, I examine the aggregate effects of banks’ delayed loan loss

recognition (DLR) on house prices during the Great Recession and the

channels through which these potential effects arose.

I examine the U.S. housing market for several reasons. Households are

major economic agents: their consumption expenditures were 65.3% of

the U.S. GDP in 2006 (FRED Economic Data). In addition, mortgage lend-

ing is the most important business for commercial banks, accounting for

about 70% of lending on bank balance sheets. Thus, the housing market

is likely to be an important way in which bank accounting ripples through

the real economy. Studies suggest that falling household wealth during the

delays in banks’ loan loss provisioning 713

Great Recession, which resulted from the collapse of the housing market,

reduced household consumption and aggravated the crisis (Mian, Rao and

Sufi, [2013], Mian and Sufi, [2014]). For these reasons, examining the ef-

fects of banks’ DLR on house prices can shed light on the role of bank

accounting during economic downturns.

Banks’ DLR could affect house prices through at least two channels: by

influencing mortgage lending (i.e., the credit-crunch channel) and by influ-

encing risk-taking (i.e., the distressed-sales channel). The credit-crunch chan-

nel suggests that high-DLR banks affect house prices through their lend-

ing. As banks delayed loss recognition, they created loss overhangs. Once

the downturn started, banks became concerned about future capital inade-

quacy and thus reduced lending (Van den Heuvel, [2009], Beatty and Liao,

[2011]). As a result, the reduced credit supply pushed down house prices.

On the other hand, the distressed-sales channel suggests that high-DLR

banks affect house prices through their prior risk-taking behavior. High-

DLR banks took more risks during good times, because DLR can weaken

discipline by stakeholders due to less transparent financial reporting or a

higher threshold for liquidation of the loan portfolio (Barth and Lands-

man, [2010], Bushman and Williams, [2012, 2015], Bertomeu, Mahieux

and Sapra, [2020], Gallemore, [2021]). Consequently, once the crisis hit,

homes with mortgages from high-DLR banks were more likely to be sold via

foreclosures or short sales, pushing down house prices (Campbell, Giglio

and Pathak, [2011]).

To investigate the aggregate effects of DLR on the housing market, I

construct the ZIP-code-level exposure to banks’ DLR using the weighted

average of individual banks’ DLRs based on their market shares before the

crisis (2004−2006). Then, I run a difference-in-differences analysis by com-

paring high- and low-exposure ZIP codes before and during the crisis. This

approach is akin to the Bartik instrument, a research design exploiting the

differential regional exposure to a common shock measured by the share

of predetermined characteristics (Breuer, [2021]). I expect larger declines

in the housing market if an area was more exposed to high-DLR (high ex-

posure to treated units) banks before the crisis (a common shock).1

An obvious concern is that economic downturns likely influence both

banks’ behavior and local economic conditions; thus, controlling for de-

mand effects is crucial. I employ several approaches to mitigate this con-

cern. First, I include county-year fixed effects to control for any county-level

time-varying economic conditions. Second, I include ZIP-code fixed effects

to control for any ZIP-code-specific invariant characteristics that may drive

1This approach relies on the assumption that the banks’ mortgage market shares at the ZIP-

code level before the crisis are uncorrelated with the changes in housing market outcomes during

the crisis after controlling for observables. Although the identifying assumption cannot be di-

rectly tested, it is akin to the parallel-trends assumption under the difference-in-differences

approach (Breuer,2021). I provide evidence that house-price changes before the crisis do not

depend on the exposure to banks’ DLR in figure 3.