Do Borrowers Intentionally Avoid Covenant Violations? A Reexamination of the Debt Covenant Hypothesis

• Published date: 01 December 2022
• Author: ADAM BORDEMAN,PETER DEMERJIAN
• Date: 01 December 2022
• DOI: http://doi.org/10.1111/1475-679X.12456

DOI: 10.1111/1475-679X.12456

Journal of Accounting Research

Vol. 60 No. 5 December 2022

Printed in U.S.A.

Do Borrowers Intentionally Avoid

Covenant Violations?

A Reexamination of the Debt

Covenant Hypothesis

ADAM BORDEMAN∗AND PETER DEMERJIAN †

Received 18 May 2021; accepted 21 June 2022

ABSTRACT

In this study, we replicate and extend the Dichev and Skinner [DS: 2002]

study on the debt covenant hypothesis (DCH). We start by replicating DS

and f‌ind results consistent with theirs. We then extend their work by chang-

ing three aspects of the research design: histogram bin width, calculation of

slack, and statistical test of discontinuity. We f‌ind that the inference from

DS is generally robust to varying these choices, although sensitive to differ-

ent bin widths, during their sample period. We extend our analysis to the

period 2000–2019 and f‌ind that support for DCH remains robust. We do,

however, f‌ind a lack of support for DCH when examining the most common

f‌inancial covenant, debt-to-EBITDA. These f‌indings suggest a more nuanced

∗Cal Poly San Luis Obispo; †Georgia State University

Accepted by Luzi Hail. The authors appreciate feedback from two anonymous reviewers,

Herb Hunt and Melissa Martin, and the support of the Orfalea College of Business at Cal Poly

San Luis Obispo and the College of Business Administration at the University of Illinois at

Chicago. An online appendix to this paper can be downloaded at https://www.chicagobooth.

edu/jar-online-supplements.

1741

© 2022 The Chookaszian Accounting Research Center at the University of Chicago Booth School of

Business.

1742 a. bordeman and p. demerjian

perspective on DCH, whereby different types of f‌inancial covenants provide

different incentives and abilities to avoid technical default.

JEL codes: G30, G32, M40, M41

Keywords: debt covenant hypothesis; debt contracting; f‌inancial covenants

1. Introduction

The debt covenant hypothesis (DCH) is one of the key testable theories

of the positive accounting paradigm described in Watts and Zimmerman

[1986]. The hypothesis is based on the idea that f‌inancial covenants in

debt contracts—provisions that require the borrower to maintain a thresh-

old level of an accounting-based metric, such as interest coverage or net

worth—are costly to violate. Watts and Zimmerman, in their “debt/equity”

hypothesis, predict that the cost of covenant violation affects borrower be-

havior. Specif‌ically, they predict that borrowers with high leverage (a proxy

for closeness to covenant violation) will make income-increasing account-

ing policy choices. This theory has been adapted into the broader DCH,

which predicts that f‌irms close to covenant violations take action to avoid

technical default through accounting policy changes, accruals, or real ac-

tivities. DCH has received considerable empirical support. Sweeney [1994]

and DeFond and Jiambalvo [1994] test hand-collected samples of covenant

violations and f‌ind that f‌irms make income-increasing accounting decisions

and manage earnings upward prior to covenant violation. More recently,

Kim, Lei, and Pevzner [2010] f‌ind that f‌irms use real activities manage-

ment to avoid violation, and Franz, HassabElnaby, and Lobo [2014] show

that f‌irms use both accrual earnings management and real activities man-

agement to avoid technical default.

Dichev and Skinner [DS: 2002] provide some of the most convincing

support for DCH. Unlike the papers noted above, which use models of dis-

cretionary accruals or real activities to examine DCH, DS use a histogram-

based analysis in the style of Burgstahler and Dichev [1997] to examine

two f‌inancial covenants, minimum current ratio and minimum net worth,

and measure the difference between the covenant threshold and the ac-

tual value of the accounting-based covenant metric, or slack.Theauthors

organize covenant slack into histogram bins and measure the smoothness

of the distribution of bin density around the threshold of technical default.

The f‌indings show a pronounced discontinuity, with a disproportionately

large number of slack observations just above the technical default thresh-

old (peak) and a disproportionately small number of slack observations just

below (trough). These signif‌icant discontinuities are interpreted as sup-

porting DCH. The f‌indings in DS are robust, showing statistically signif‌icant

discontinuities across a variety of subsamples of covenant slack.

In this paper, we propose a reexamination of the DCH. Although DS pro-

vide clear, strong support for the hypothesis and their paper remains inf‌lu-

ential, we believe that a reexamination is justif‌ied for several reasons. First,

do borrowers intentionally avoid covenant violations? 1743

as is the case in all empirical studies, DS make a variety of measurement and

specif‌ication choices that potentially inf‌luence the inferences from their

study. A reexamination allows us to test the robustness of their f‌indings to

different research design choices. Second, research methodologies evolve

over time, allowing for new methods and tests that can change inferences

from past research. In our reexamination, we use an alternative statistical

test that had not been developed when DS conducted their research.

Third, 20 years have passed since their paper was published. These two

decades have seen signif‌icant developments in f‌inancial reporting and the

broader economy. Some of these changes, including the passage of the

Sarbanes–Oxley Act, the global f‌inancial crisis of 2007 and 2008, and the

rise of securitization in the private loan market, have likely inf‌luenced

the ability and incentives of borrowers to avoid covenant violations. Al-

though past research suggests that borrowers acted in accordance with

DCH in the period that DS study, whether this is still the case is an empirical

question. Finally, current ratio and net worth covenants are not commonly

used f‌inancial covenants and have fallen even more out of favor in recent

years; thus, inferences of DCH based on these covenants may not general-

ize.

We start our reexamination of DCH by reproducing DS. Although

we attempt to precisely replicate the sample, measurement choices, and

empirical tests in their study, changes in the Loan Pricing Corporation

(LPC)/Dealscan data since DS’s analysis make a precise sample replication

impossible. We also note that some covenants in our sample have either

dynamic thresholds (current ratio) or build-up provisions (net worth) that

affect the measurement of slack. We hand-collect data from Securities and

Exchange Commission (SEC) f‌ilings to address these measurement issues.1

Considering changes in data availability, we reproduce their sample as best

as we can; our descriptive statistics suggest that our reproduced samples

for current ratio and net worth covenants are close to those of DS, and

we do not believe that sample differences introduce systematic bias into

our results. Using these samples and following the DS research design, our

f‌indings are consistent with DCH. In fact, our distributional results closely

mirror the f‌indings of DS for each of the f‌ive specif‌ications tested for the

current ratio and net worth covenant samples. We conclude that, despite

possible sample differences, we faithfully reproduce DS’s analysis.

In the next phase of our reexamination, we run three research design-

based extensions of DS, focusing on the subsample of observations up to

and including the f‌irst documented covenant violation. In our f‌irst exten-

sion, we examine alternate histogram bin widths. DS use ad hoc bin widths

that approximate a doubling of the interquartile range-based formula of

Degeorge, Patel, and Zeckhauser [1999]. We explore narrowing and widen-

ing the DS bin widths. For our second extension, we vary the measurement

1DS discuss both dynamic thresholds and build-up provisions. We discuss their approach to

these data issues, along with more detail on our hand-collection procedure, in subsection 3.3.