The Impact of Repossession Risk on Mortgage Default

• Published date: 01 April 2021
• Date: 01 April 2021
• DOI: http://doi.org/10.1111/jofi.12990
• Author: TERRY O'MALLEY

THE JOURNAL OF FINANCE •VOL. LXXVI, NO. 2 •APRIL 2021

The Impact of Repossession Risk on Mortgage

Default

TERRY O’MALLEY∗

ABSTRACT

I study the effect of removing repossession risk on a mortgagor’s decision to de-

fault. Reducing default costs may result in strategic default, particularly during

crises when homeowners can be substantially underwater. I analyze difference-in-

differences variation in repossession risk generated by an unexpected legal ruling in

Ireland that prohibited collateral enforcement on delinquent residential mortgages

originated before a particular date. I estimate that borrowers defaulted by 0.3 per-

centage points more each quarter after the ruling, a relative increase of approxi-

mately one-half. High loan-to-value ratios and low liquidity are associated with a

larger treatment effect, suggesting both equity and consumption-based motivations.

DURING ECONOMIC CRISES,GOVERNMENTS often consider debt relief poli-

cies that reduce repossession risk for mortgage borrowers, such as foreclo-

sure moratoria, because of the negative externalities and social disruption of

widespread evictions.1However, these policies come with a potentially large

moral hazard cost. By lowering the costs of default, reducing repossession risk

makes default more attractive. This association may be particularly strong

during recessions, when underwater homeowners have strong incentives to

default. Absent repossession risk, some borrowers who would have otherwise

continued paying choose to default, substantially reducing the net benef‌its of

∗Terry O’Malley is at the Central Bank of Ireland. I thank Morgan Kelly and Fergal McCann

for detailed comments and guidance throughout this project. I also thank Amit Seru and the ed-

itorial team, whose comments greatly improved the paper. I acknowledge helpful comments from

Benjamin Arold; David Byrne; Paul Devereux; Andreas Fuster; Edward Gaffney; Brian Higgins;

Robert Kelly; Marianna Kudlyak; Philip Lane; Michael McMahon; Rachael Meager; Conor O’Toole;

Kevin Schnepel; Johannes Stroebel; Karl Whelan; as well as from participants at the 2017 Irish

Economic Association Annual Conference, 2017 European Economic Association Annual Congress,

2017 Empirics and Methods in Economics Conference, 2018 European Doctoral Group in Eco-

nomics Jamboree, and both Central Bank of Ireland and University College Dublin seminars.

Special thanks to John Joyce for assistance with the data. I have read The Journal of Finance

disclosure policy and have no conf‌licts of interest to disclose.

Correspondence: Terry O’Malley, Central Bank of Ireland, Dublin 1, Ireland; e-mail: terry.

omalley@ucd.ie.

1A recent literature from the United States foreclosure crisis shows large negative external-

ities of foreclosures in terms of house prices (Campbell, Giglio, and Pathak (2011), Mian, Suf‌i,

and Trebbi (2015)), underinvestment in the housing stock (Melzer (2017)), and aggregate demand

externalities (Mian, Suf‌i, and Trebbi (2015)).

DOI: 10.1111/jof‌i.12990

© 2020 the American Finance Association

623

624 The Journal of Finance®

the policy. In this paper, I estimate the magnitude of the impact of removing

repossession risk on default by examining a recent natural experiment that

occurred in Ireland.

Clean estimates of the moral hazard costs of debt relief policies such as fore-

closure moratoria are diff‌icult to obtain. Since restricting debt relief to a partic-

ular group of borrowers partially negates its benef‌its, natural experiments are

rare. Moreover,while foreclosure moratoria were widely used across the United

States during the Panic of 1819 and the Great Depression (Alston (1984), Skil-

ton (1943)) and internationally during the recent Great Recession albeit less

extensively (Artavanis and Spyridopoulos (2018), Gabriel, Iacoviello, and Lutz

(2020)), to date there have been few situations in which both the moratoriums

apply differentially to similar borrowers and the necessary data are available.

Ireland, however, offers an interesting setting to study the moral hazard cost

of debt relief policies. Owing to a poorly drafted change in Irish law governing

property transfers and its later unexpected discovery during a court case, the

repossession regime of existing mortgage contracts was retroactively amended.

Specif‌ically, in July 2011, a judge ruled that properties mortgaged before De-

cember 2009 could no longer be repossessed in the event of default.

To estimate the impact of repossession risk on mortgage default, I compare

the performance of two loan vintages separated by the December 2009 cutoff

date, both before and after the July 2011 legal ruling (known as the “Dunne

judgment”). Using regulatory panel data, I construct a sample of loans origi-

nated over a 180-day window around the cutoff date, and then, match treat-

ment and control groups to obtain the difference-in-differences (DD) variation

in repossession risk. The two groups are similar on observables before the

Dunne judgment, consistent with the exogeneity of the origination cutoff date.

However, while the groups follow parallel trends before the ruling, their de-

fault rates diverge markedly afterwards, with the largest effect observed in

the quarter immediately following the ruling.

Specif‌ically, I estimate that the Dunne judgment increased the quarterly

default rate by 0.3 percentage points. In the control group, a 0.3 percentage

point higher default rate is associated with variation between the median loan-

to-value (LTV) ratio plus 38 percentage points, or the median interest rate

plus two percentage points. When scaled by the estimated counterfactual, this

translates to a relative effect of between 40% and 60%.2

To gauge the robustness of this result, I estimate a variety of models under

different assumptions about confounding factors. Alongside a straightforward

DD regression, I consider specif‌ications with additional borrower and loan co-

variates and with high-dimensional f‌ixed effects. Since treatment status is

determined by a loan’s origination date, loan vintage is a nuisance parame-

ter under my research design. I therefore also estimate models that adjust

for potentially confounding cohort effects by including vintage-calendar time

parameters in the estimation. The signif‌icance of the treatment effect is ro-

bust to these different specif‌ications. Graphical evidence is also consistent with

2Table II contains a range of absolute effects.