Sign Reversal in the Relation between Income Smoothing and Cost of Debt
Dan Amiram
Columbia University Graduate School of Business
Coller School of Management, Tel Aviv University
da2477@columbia.edu
Edward Owens
Goizueta Business School, Emory University
ed.owens@emory.edu
Accepted Article
October 2017
We thank Peter Pope (editor), an anonymous reviewer, Edwige Cheynel, Peter Demerjian, Ilia Dichev,
Shane Dikolli, Ron Dye, Ted Goodman (AAA discussant), Trevor Harris, Alon Kalay, Mark Lang, Yun
Lou, Mark Maffett, Nahum Melumad, Grace Pownall, Gil Sadka, Cliff Smith, Jason Wei, Chris Williams,
Regina Wittenberg-Moerman (Colorado conference discussant), Joanna Wu, Paul Zarowin, Jerry
Zimmerman, and workshop participants at the 2011 International Conference on Credit Analysis and Risk
Management, the 2011 American Accounting Association Annual Meeting, the 2011 Columbia
University Burton Workshop, 2012 Colorado Summer Accounting Conference, Emory University,
National University of Singapore, Northwestern University, and Temple University for helpful comments
and suggestions. We are grateful to Ryan Ball and Florin Vasvari for providing a matching table between
Dealscan and Worldscope.
ABSTRACT
Despite the fact that income smoothing by managers is a pervasive phenomenon that has been
widely researched, extant literature provides incomplete evidence on how smoothing is
associated with cost of debt in general, and in the private loan market in particular. The
institutional factors associated with private loan contracts, combined with the theoretical
motivations for smoothing, make it unclear whether smoothing will be positively, negatively, or
not associated with loan spread. Using both cross-country and within-country analyses on an
international sample of private loans, we predict and provide evidence that income smoothing is
associated with lower cost of debt when the threat of private benefits consumption by managers
is low, but is associated with higher cost of debt when the threat of private benefits consumption
by managers is high. We provide the first evidence in the literature that the garbling effect of
smoothing can predictably dominate the signaling view of smoothing in debt contract design, and
we identify private benefits consumption threat as the feature of the contracting environment that
empirically reveals a sign reversal in the relation between smoothing and cost of debt.
Keywords: Income smoothing; Private benefits; Debt contracts
JEL Classification: F34; G15; M41
_______________________________
This article has been accepted for publication and undergone full peer review but has not been through the copyediting, typesetting,
pagination and proofreading process, which may lead to differences between this version and the Version of Record. Please cite this article
as doi: 10.1111/jbfa.12295
This article is protected by copyright. All rights reserved.
1.
Introduction
Despite the fact that income smoothing by managers is a pervasive phenomenon that has
been widely researched, extant literature provides incomplete evidence concerning how borrower
income smoothing is associated with cost of debt in general, and in the private loan market in
particular. The institutional factors associated with private loan contracts, combined with the
theoretical motivations for smoothing, make it unclear ex ante whether smoothing will be
Accepted Article
positively, negatively, or not associated with loan spread. Thus far, extant literature has
examined settings where only the signaling effect of smoothing dominates, and therefore
concludes that smoothing is associated with lower cost of debt. In this study, we extend the
literature. Specifically, we take advantage of an international setting to predict and identify a
theoretically motivated feature of the contracting environment that reveals a sign reversal in the
relation between smoothing and cost of debt, thereby providing the first evidence in the literature
that the garbling effect of smoothing can predictably dominate the signaling effect of smoothing
in debt contract design.
Income smoothing is the exercise of managerial discretion to alter the time profile of
earnings to reduce the variability of a firm's reported income stream (e.g., Beidleman, 1973;
Trueman & Titman, 1988; Fudenberg & Tirole, 1995).1 The literature focuses on two key
alternative motivations for income smoothing. On one hand, managers may smooth earnings in
an effort to truthfully convey their private information about underlying economic earnings and
associated risk to capital providers and other market participants (i.e., the information signaling
1
The overall volatility of a firm’s earnings relative to cash flows (i.e., ―income smoothness‖) is determined by two
components. The first component, which we refer to as ―fundamental smoothness,‖ is determined by the inherent
volatility of the underlying business processes that generate earnings (which is not a function of managerial
reporting discretion). The second component, which we refer to as ―income smoothing‖ or simply ―smoothing,‖ is
the portion of overall income smoothness that is determined by managerial reporting choices (i.e., managerial
discretion), which themselves are affected by managerial incentives.
This article is protected by copyright. All rights reserved.
view of smoothing) (e.g., Trueman & Titman, 1988). Alternatively, managers may smooth
earnings (regardless of their private information about longer-term economic earnings) in an
effort to avoid intervention by outsiders to facilitate consumption of private benefits from the
firm (i.e., the information garbling view of smoothing) (e.g., Fudenberg & Tirole, 1995).2,3
The information signaling and information garbling views of smoothing yield
directionally opposing predictions concerning the association between observed smoothing and
cost of debt. The signaling view suggests that smoothing is an attempt by managers to signal that
Accepted Article
economic earnings are less volatile than would be inferred from unsmoothed (reported) earnings,
and therefore smoothing suggests lower probability of default (e.g., Merton, 1974). In contrast,
the garbling view suggests that smoothing is associated with higher risk of private benefits
consumption. To the extent that private benefits consumption extracts firm wealth at the expense
of debtholders (Lin, Ma, Malatesta, & Xuan, 2011), the information garbling view of smoothing
would suggest higher loss given default and higher probability of default (Fudenberg & Tirole
1995). In the context of these alternative views of smoothing, ceteris paribus, we predict cost of
debt will be lower if lenders take the information signaling view of smoothing, and will be
higher if lenders take the information garbling view of smoothing, ceteris paribus. Moreover, as
lenders only lose money if a borrower actually defaults, we expect the effects of smoothing on
cost of debt to be stronger when the probability of default is higher.
2
We use the term 'private benefits' as a general term that encompasses several concepts, including both valueextractive private benefits and non-value-extractive private benefits, both of which arise from managers' ability to
control the firm. Aghion and Bolton (1992) refer to these categories as "pecuniary" and "non-pecuniary," or
"monetary" and "non-monetary" private benefits, respectively. To be more precise, we choose the terms valueextractive and non-value-extractive, because, for example, there can be non-monetary private benefits that
nonetheless are value extractive (e.g., shirking). Value-extractive private benefits arise from actions such as empire
building, rent extraction, perk consumption and expropriation (Tirole, 2001), which reduces the value of the firm to
minority shareholders and debt holders. In contrast, non-value-extractive private benefits include social and
professional connections, status and opportunities that arise from continually controlling and running the entity
(Aghion & Bolton, 1992).
3
We note that in this context, the terms 'signaling' and 'garbling' refer to the intentions of the manager who is
making the smoothing choice. Extant literature sometimes refers to these as 'informative' and 'opportunistic'
motivations for smoothing, respectively (e.g., Dechow, Ge, & Schrand, 2010).
This article is protected by copyright. All rights reserved.
Extant literature provides mixed evidence on the association between smoothing and cost
of capital in the equity market. This evidence does not translate into debt markets because
creditors have different payoff functions than do equityholders, and can protect themselves
through other contracting mechanisms (e.g., financial covenants). Existing evidence from
research on the effects of smoothing in debt markets provides evidence that either smoothing is
associated with lower cost of debt, or that there is no association. That is, extant literature does
not document any setting or conditions where smoothing is associated with higher cost of debt,
Accepted Article
which is noteworthy given the strong theoretical literature that suggests that smoothing can, in
some situations, represent garbling by management, which could negatively affect lenders.
Our conjecture is that extant literature has not examined the association between
smoothing and cost of debt in a setting or manner in which the garbling effect can reveal itself.
Evidence from the U.S. bond market that smoothing is associated with lower cost of debt is not
generalizable to broader credit markets, because the bond market is populated by relatively high
quality, transparent borrowers. Thus, the garbling view of smoothing seems relatively unlikely to
manifest (Bharath, Sunder, & Sunder, 2008).
Studies that examine smoothing in the private loan market find either a negative or no
association between smoothing and cost of debt, either because they do not focus on the subset
of borrowers where garbling effects likely dominate, or because lenders can obtain private
information from borrowers, which may mitigate any association between smoothing and
contract terms. We view the latter possibility as unlikely, because if managers indeed smooth to
facilitate consumption of private benefits, it is unlikely they will reveal their motivation to
lenders in private communications. On the other hand, if managers smooth to signal, it is indeed
plausible that managers may also supply private information to lenders that reinforce this signal.
This article is protected by copyright. All rights reserved.
Importantly, we do not assert that private lenders are the only (or, even, the primary) audience
for signal-based smoothing.4 Accordingly, managers would still need to smooth income to signal
their private information to other market participants with whom no private channels of
communication exist (e.g., equityholders, bondholders).5
Our conjecture is that lenders’ view of smoothing is shaped by their assessment of the
likelihood that managers will consume private benefits, which is a function of the extent of
enforcement and penalties imposed in a given environment if managers are caught consuming
Accepted Article
private benefits. Hereafter, we refer to environments with weak (strong) enforcement and weak
(strong) punitive consequences for private benefits consumption as posing a ―high (low) threat‖
of private benefits consumption. In a probabilistic sense, we conjecture that in high (low) threat
environments lenders are more likely to take the information garbling (signaling) view of
smoothing.6 Intuitively, if there are harsh penalties for consuming private benefits, lenders will
assess a lower likelihood that managers are attempting to consume private benefits, ceteris
paribus, and their view of smoothing will be shaped accordingly. 7 This intuition provides the
basis for our empirical design, where our key design choice concerns how to empirically
measure the threat of private benefits consumption in the contracting environment.
Firm-level governance measures provide one such option for measuring the threat of
private benefits consumption. However, because managers choose both the extent of smoothing
4
For example, Dou, Hope and Thomas (2013) provide evidence firm-supplier relationships can provide smoothing
incentives.
5
Indeed, the logic of our entire study goes through if, in the extreme, the lender can directly obtain all information
from the borrower other than whether the borrower is consuming private benefits or not, and misreporting audited
financial statements is more costly than misreporting private information..
6
Consistent with our conjecture, in informal interviews with bank loan officers in countries where it is easier to
consume private benefits, loan officers indeed indicated that they become "nervous" about a borrower if its financial
statements look too "stable." In contrast, the loan officers are reassured when they observe smooth earnings in
countries where it is harder to consume private benefits. Having said this, our predictions also follow if lenders infer
their view of smoothing from other information sources (i.e., not relying on environmental characteristics), as long
as borrower smoothing motivations are correlated with private benefits consumption threat.
7
We formalize this intuition in Appendix B.
This article is protected by copyright. All rights reserved.
and firm-level governance mechanisms, there are standard endogeneity concerns with using
firm-level measures. Accordingly, for our primary analyses we use an international sample,
which enables us to exploit plausibly exogenous (from the perspective of a manager) countrylevel variation in the threat of private benefits consumption in the contracting environment.
Consistent with our predictions, we provide evidence that smoothing is associated with
lower cost of debt in countries characterized by low threat of private benefits consumption, and
associated with higher cost of debt in countries characterized by high threat of private benefits
Accepted Article
consumption. Further, we find that the effects of smoothing on cost of debt are particularly
pronounced for firms with relatively high credit risk, consistent with the intuition that lenders
would be particularly concerned about the implications of smoothing for expected credit loss in
cases where it is more likely that borrowers will default.
We note that it is unlikely that lenders view smoothing as information signaling
(garbling) for all firms in a low (high) threat environment. That is, within either a low or high
threat environment, firm-specific threat of private benefits consumption likely influences the
relation between smoothing and cost of debt. Although pursuing an analysis based on firm-level
governance characteristics naturally raises endogeneity concerns, we repeat our tests using
measures of firm-level variation in the threat of private benefits consumption within a given
country. Specifically, we use percentage of closely held shares (e.g., Iliev, Lins, Miller, & Roth,
2015) for our primary non-US sample, and repeat our analysis using a firm-specific managerial
entrenchment index (Bebchuk, Cohen, & Ferrell, 2009) with a larger US-only sample. Consistent
with our primary findings, in both cases smoothing is negatively (positively) associated with cost
of debt for firms with relatively low (high) threat of private benefits consumption.
This article is protected by copyright. All rights reserved.
Although our analyses focus on the association between smoothing and cost of debt,
private lenders have loan contracting terms other than interest rate at their disposal that may
likewise be affected by income smoothing, such as financial covenants, collateral requirements,
and loan maturity. Therefore, in all analyses we control for numerous additional loan-level
variables. Further, results suggest that the implications of observed smoothing are incorporated
into loans primarily through loan spread, although there is some evidence that smoothing is
associated with increased use of financial covenant protections in high threat environments.
Accepted Article
Extant literature documents a correlation among different accounting system
characteristics, including smoothing, accruals quality, conservatism, and earnings persistence
(e.g., Dechow et al., 2010). However, there are key differences in the underlying constructs
captured by smoothing and these other accounting attributes. We find that neither accruals
quality, conservatism, nor earnings persistence change sign in their relation with loan spread
across threat environments (consistent with our expectations), and our inferences regarding
smoothing hold after including controls for potential effects of accounting quality, conservatism,
and earnings persistence.
The primary contribution of the study is to provide novel evidence that the garbling effect
of smoothing can dominate debt contract design, where the sign of the relation between cost of
private debt and earnings smoothing depends (predictably) on the threat of managerial private
benefits consumption in the contracting environment. Our study directly addresses a call for
additional research on the relation between smoothing and cost of debt (Lang & Maffett, 2011a),
and contributes new evidence to the international literature that examines the causes and
consequences of income smoothing in capital markets (e.g., Black, Sellers, & Manly, 1998;
Leuz, Nanda, & Wysocki, 2003; Lang, Lins, & Maffett, 2012; Shuto & Iwasaki, 2014).
This article is protected by copyright. All rights reserved.
2.
Background and Motivation
Private debt contracts represent an important source of capital for most firms (Sufi, 2007;
Ivashina, 2009; Thomson Reuters, 2014). Despite the fact that a deep literature examines income
smoothing, there is incomplete evidence on its association with the cost of private debt. Extant
literature focuses on two competing views of income smoothing—the information signaling view
and the information garbling view.8 The information signaling view suggests that income
smoothing is an efficient mechanism for managers to convey their private information about
Accepted Article
future earnings (e.g., Beidleman, 1973; Barnea, Ronen, & Sadan, 1975; Ronen & Sadan, 1981;
Demski, 1998). Consistent with the information signaling view, Graham, Harvey, and Rajgopal
(2005) provide U.S.-based survey evidence that managers' smooth primarily to convey future
growth prospects and lower investor risk perception. Further consistent with this view, Hunt,
Moyer, and Shevlin (2000) provide evidence in a U.S. setting that income smoothing enhances
the contemporaneous relation between stock price and earnings. Likewise, using a U.S. sample,
Tucker and Zarowin (2006) find that the change in the current stock price of firms with more
earnings smoothing contains more information about future earnings than does the change in
current stock price of firms with lower smoothing.
In contrast, the information garbling view suggests that managers smooth income to hide
their actions and avoid interventions by outsiders to facilitate private benefits consumption
(Fudenberg & Tirole, 1995; Acharya & Lambrecht, 2015). According to extant theory,
smoothing can reduce outside intervention by monitors because more recent signals of
performance are more informative than less recent signals; therefore, a bad earnings outcome
leads to negative consequences for the manager even if the firm had an "offsetting" good
8
We recognize that motivations for smoothing exist aside from signaling or garbling, such as tax minimization.
Consideration of other such smoothing motivations does not affect the key logic of our study, as discussed in
Section 6.
This article is protected by copyright. All rights reserved.
earnings outcome in an earlier period, while an average outcome leads to fewer negative
consequences (Fudenberg & Tirole, 1995). That is, in a firm where unsmoothed earnings are
volatile, managers have incentive to smooth earnings simply to reduce volatility (without regard
to their belief about longer term economic earnings), because doing so reduces the probability
that they will be caught consuming private benefits (if they have chosen to do so) (Fudenberg &
Tirole, 1995; Leuz et al., 2003; Lang & Maffett, 2011b; Defond & Park, 1997). Defond and Park
(1997) find empirical evidence of smoothing that appears to be motivated by management
Accepted Article
concerns about job security, consistent with the theory in Fudenberg and Tirole (1995).
International evidence presented in Leuz et al. (2003) suggests that income smoothing is more
pronounced in countries where country-level institutions do a relatively poor job in limiting
insiders’ consumption of private benefits, which is also consistent with the garbling view of
smoothing. Relatedly, Gopalan and Jayaraman (2012) provide evidence that firms with more
insider control exhibit relatively more income smoothing than non-insider controlled firms in
countries with poor investor protection.
Studies that examine the effect of smoothing in debt markets are few. Li and Richie
(2016) examine whether smoothing affects cost of public debt in the U.S. bond market. That
study predicts and documents a negative association between smoothing and bond yield, and
accordingly concludes that the information signaling view of smoothing dominates in public
credit markets. Li and Richie (2016) further consider the effect of governance characteristics,
and do not provide evidence that the negative relation between smoothing and cost of debt is
statistically significantly different between well-governed and poorly-governed firms. Further,
their results show that in U.S. public credit markets, the garbling effect of smoothing is not
strong enough to dominate debt contract design (i.e., the relation between smoothing and cost of
This article is protected by copyright. All rights reserved.
public debt is never positive), even in poorly-governed firms. Using a sample of Japanese loans,
Takasu (2013) documents a negative relation (no relation) between smoothing and cost of debt
when the lender has a low (high) degree of information production about the borrower.
Demerjian, Donovan, and Lewis-Western (2017) examine whether smoothing affects the
frequency of covenant violation in U.S. private loan contracts. Although that paper focuses on
covenants, the paper also presents a secondary analysis which shows that smoothing is
negatively associated with loan spread in the U.S., again consistent with the information
Accepted Article
signaling view of smoothing in an environment with low threat of private benefits consumption.
Relatedly, Gassen and Fulbier (2015) provide evidence that the negative relation between
smoothing and cost of debt is stronger in environments with poor contracting efficiency,
consistent with lenders’ desire to avoid covenant violations where contract renegotiation is more
difficult. As with Li and Richie (2016), Gassen and Fulbier (2015) do not find any evidence that
there is ever a positive association between smoothing and cost of debt, even in environments
with poor firm governance.9
To summarize, extant evidence leaves a reader of the literature with the unequivocal
inference that smoothing is negatively associated with cost of debt. However, these extant credit
market studies examine settings (i.e., bond markets and/or U.S. samples) where the information
garbling motivation is relatively unlikely to reveal itself. Accordingly, it is not clear whether
extant studies fail to find any evidence in the credit markets consistent with the information
garbling view of smoothing because it does not exist, or because the chosen settings are
inadequate to detect it. Our evidence indicates that it is the latter.
9
Another related literature documents a positive relation between income smoothing and credit ratings, consistent
with dominance of the signaling view of smoothing (e.g., Gu & Zhao, 2006; Jung, Soderstrom, & Yang, 2013).
This article is protected by copyright. All rights reserved.
3.
Development of Predictions
As discussed above, our conjecture is that the sign of the association between smoothing
and debt contract terms will depend on whether lenders take the information signaling or
garbling view of smoothing, because these alternative views have directionally opposite effects
on expected credit loss. If lenders take the information signaling view (i.e., smoothing is
associated with lower probability of default, as in Merton, 1974), we predict a negative
association between smoothing and cost of debt, ceteris paribus. If lenders take the information
Accepted Article
garbling view (i.e., smoothing is associated with higher expected loss given default), we predict a
positive association between smoothing and cost of debt, as prior literature documents that
lenders price protect against the threat of private benefits consumption in the contracting
environment (Lin et al., 2011). Further, we predict that, on-average, lenders will more likely take
the information signaling (garbling) view in environments where there is stronger (weaker)
enforcement and harsher (lighter) penalties associated with managerial private benefits
consumption.
These predictions raise several natural questions. First, if smoothing is indeed associated
with lower cost of debt in environments with low threat of private benefits consumption, why
would not all managers in those environments smooth to the same extent? There are at least two
reasons. We expect cross-sectional variation in the extent of smoothing among borrowers in low
threat environments simply because not all borrowers have the same ability to smooth, or
because borrowers that can smooth but do not have smooth economic earnings will forego
smoothing in order to report truthfully, because not doing so is costly.
A second question that arises is, if firms need to communicate with lenders about firm
risk, why would they need to use smoothing, rather than private information channels that are
This article is protected by copyright. All rights reserved.
available between borrowers and private lenders? It is indeed true that borrowers have private
information channels with lenders, and therefore if a borrower has smooth economic earnings, it
can communicate that privately to lenders. However, if such a hypothetical borrower indeed has
smooth economic earnings, they will likely tell the truth and reflect that in public financial
statements, i.e., they will smooth earnings, if for no other reason than to communicate that reality
to other capital market participants with whom the borrower does not have private information
channels (e.g., equityholders).
Accepted Article
A third question is, if smoothing is associated with higher cost of debt in environments
with high threat of private benefits consumption, why would any firm smooth - that is, would the
lender price protection not remove the smoothing incentive? In short, the answer is no. There are
two categories of utility-providing private benefits that managers can consume—those that
extract firm value (e.g., excess compensation, tunneling), and those that do not (e.g., status of
running the firm, social and professional connections) (Aghion & Bolton, 1992). Lenders would
only price protect against consumption of private benefits that are firm-value extractive.
Therefore, as long as any non-value-extractive private benefits exist, managers will have an
incentive to smooth, even if lenders fully price protect against consumption of private benefits
that are firm-value extractive.10 Other possibilities include tax or hedging benefits from
smoothing, or only partial price protection for value-extractive private benefit consumption. In
any case it is an empirical regularity that smoothing is associated with private benefits
consumption (e.g., Leuz et al., 2003).
10
Consider the following motivating example, which we more formally develop in Appendix B. If a manager
chooses to smooth to consume non-value-extractive private benefits (e.g., to avoid intervention and keep her job),
lenders will observe smooth earnings and will be unable to distinguish whether managers are consuming valueextractive or non-value-extractive private benefits. Thus, the lender will price protect against the potential valueextractive component. The manager knows this, so the manager will go ahead and consume those value-extractive
private benefits, along with enjoying the non-extractive private benefits. Even though the price protection offsets the
value-extractive benefit consumed by the manager, the manager remains better off from the non-extractive private
benefits, which can only be achieved via smoothing.
This article is protected by copyright. All rights reserved.
In Appendix B, we present a highly stylized framework that imbeds the dynamics we
discuss above, which presents one possibility in which the equilibrium we describe can exist in
reality. That is, we consider a firm whose manager has private information about the smoothness
of economic earnings, and unobservable (to outsiders) ability to either smooth or not smooth
earnings. The environment in which the manager operates is characterized as having either high
or low costs of getting caught consuming private benefits, and high or low costs associated with
misreporting true economic earnings (e.g., Desai, Hogan, & Wilkins, 2006). The manager then
Accepted Article
chooses whether to consume private benefits and whether to smooth earnings (if able). We also
characterize potential private benefits of two forms: private benefits that extract firm value (and
will therefore be price protected by capital providers), and those that do not (and therefore will
not be price protected). The firm then seeks a private loan. Lenders can observe the threat of
private benefits consumption in the environment, and whether income is smooth. We show that
in an environment with high consumption threat, smoothing is associated with higher cost of
debt in (partial) equilibrium - indeed, lenders fully price protect against expected consumption of
value-reducing private benefits, but the existence of non-value reducing private benefits still
makes it optimal from the manager's perspective to smooth income. Further, we show that in an
environment with low consumption threat, smoothing is associated with lower cost of debt in
(partial) equilibrium, because smoothing reveals information about firm risk with low likelihood
of private benefits consumption.11
11
Alternatively, for example, we could characterize a framework where the firm is financed with equity, and the
manager is making smoothing decisions in that context. Then, in a future period there is an exogenous increase in
growth opportunities that causes the firm to seek debt financing. It is straightforward to show that our same
predictions would follow in this situation where the smoothing choice is exogenous to the borrowing decision.
This article is protected by copyright. All rights reserved.
4.
Research Design
4.1.
Income smoothing
Earnings smoothness is comprised of smoothness that is driven by the natural business
processes and operating cycle of a firm and the application of non-discretionary accounting
regulations to those processes (i.e., fundamental smoothness), as well as smoothness that is
driven by managerial discretion (i.e., discretionary smoothing). A key empirical challenge is
disentangling observed smoothness into these fundamental and discretionary components
Accepted Article
(Dechow et al., 2010).12 To do so, we closely follow the approach in Lang and Maffett (2011b)
and Lang et al. (2012).13
Following the Lang et al. (2012) approach we first compute two alternative measures of
overall earnings smoothness for firm i in period t, which we denote Smooth1i,t and Smooth2i,t.
Smooth1 is the negative of the ratio of the standard deviation of operating earnings to the
standard deviation of operating cash flows, where both earnings and cash flows are scaled by
lagged total assets prior to computation of the standard deviations. 14 Larger values of Smooth1
indicate more income smoothing. Smooth2 is the negative of the correlation between accruals
and operating cash flows (both scaled by lagged total assets) over the three to five year period
ending in year t, where higher values of Smooth2 indicate more earnings smoothness. Second,
12
We note that, although lenders may attempt to estimate the extent of a borrower’s discretionary smoothing in an
attempt to unravel the manager's intervention relative to the unsmoothed earnings stream, this may not be strictly
necessary to generate our predictions. In other words, it is possible that if lenders are unable to do this unraveling
and simply use the extent of observed smoothness as a proxy for discretionary smoothing (together with their
assessment of whether the threat of private benefits consumption is low or high), cost of debt will be higher (lower)
in high (low) threat environments when the borrower has a relatively smooth income stream, on average.
13
We appreciate the difficulties and concerns inherent in performing this decomposition. However, Lang et al.
(2012) (pp. 768-770) conduct and report extensive construct validity tests on this measure in their appendix.
Moreover, this measure has been used frequently in recent literature (e.g., Hamm, Jung, & Lee, 2017; Lang & SticeLawrence, 2015; Friedman, 2017).
14
We compute operating cash flow as net income before extraordinary items minus accruals. We compute accruals
as the change in current assets less the change in current liabilities less the change in cash plus the change in current
debt less depreciation and amortization. The standard deviations are estimated using no fewer than three and no
more than five annual observations ending in year t.
This article is protected by copyright. All rights reserved.
we regress Smooth1 and Smooth2 on a set of proposed fundamental determinants of earnings
smoothness using the following pooled estimations:
9
Smooth1i ,t  0    f Zi ,ft  industry  year   i ,t
(1)
f 1
9
Smooth2i ,t  0    f Zi ,ft  industry  year   i ,t
(2)
f 1
where Z i ,ft is the following vector of nine fundamental smoothness determinants (Lang et al.,
Accepted Article
2012): natural log of total assets (LSize), a measure of firm size; leverage (Leverage), to capture
differences in financing choices; book-to-market ratio (BTM), to capture asset tangibility and
expected earnings growth; standard deviation of firm i's annual sales (StdSales) over the three-tofive year period ending in year t, to capture underlying operating volatility; percentage of years
firm i experienced negative operating earnings in the three-to-five year period ending in year t
(%Loss), to capture differences in accrual properties of loss observations; operating cycle
(OpCycle); average annual sales growth over the three-to-five year period ending in year t
(AvgSalesGrowth), to capture growth opportunities; operating leverage (OpLev), to capture
capital intensity; and average annual cash flow from operations over the three-to-five year period
ending in year t (AvgOpCash), to capture general profitability level.15 When estimating Eqs. (1)
and (2), we further include industry and year fixed effects to capture different accrual properties
across industries, and to control for macro-economic cycles. We define discretionary smoothing
(fundamental smoothness) as the residual (predicted value) from Eqs. (1) and (2), denoted
DiscSmooth1i,t and DiscSmooth2 i,t (FundSmooth1i,t and FundSmooth2 i,t), respectively.
Next, we construct the firm-year measure of income smoothing, DiscSmoothi,t, as the
average of firm i's within-country percentile rank values of DiscSmooth1 and DiscSmooth2:
15
Detailed variable definitions are presented in Appendix A.
This article is protected by copyright. All rights reserved.
DiscSmoothi ,t 
( PcntileDiscSmooth1i ,t  PcntileDiscSmooth2i ,t )
2
.
(3)
By construction, PcntileDiscSmooth1, PcntileDiscSmooth2, and DiscSmooth each ranges from 0
to 99. We similarly construct a firm-year measure of fundamental smoothness, FundSmoothi,t, as
the average of firm i's within-country percentile rank values of FundSmooth1 and FundSmooth2.
4.2.
Threat of Private Benefits Consumption
By definition, value-extractive private benefits consumption by firm insiders is
Accepted Article
detrimental to all external capital providers, including private lenders. Expropriation of firm
assets by insiders leaves fewer resources inside the firm to satisfy the claims of both creditors
(e.g., Lin et al., 2011) and minority shareholders (e.g., Djankov, La Porta, Lopez-de-Silanes, &
Schleifer, 2008). To capture this broad threat of private benefits consumption in the contracting
environment, for our primary test we use the country-level anti-self-dealing index of Djankov et
al. (2008), which has been used in prior debt literature to establish that lenders price protect
against private benefits consumption threat (Lin et al., 2011). This index, computed for 72
countries, focuses on both public and private enforcement mechanisms (e.g., litigation, fines,
prison terms) that govern a hypothetical self-dealing transaction. Higher (lower) values of the
index imply more (less) protection and enforcement against expropriation by corporate insiders
(i.e., private benefits consumption). We define a country-level indicator variable, PBThreat, that
equals one if the anti-self-dealing index of firm i's country is below the sample observation
median (suggesting a relatively high threat of private benefits consumption), and zero otherwise.
We note that extant literature offers several institutional measures of creditor protection.
For example, Djankov, McLiesh, and Shleifer (2007) develop a debt enforcement index, which
reflects the ability of creditors to enforce their claims once a firm becomes insolvent. However,
these creditor protections would do nothing to alleviate creditors' concerns that managers may
This article is protected by copyright. All rights reserved.
expropriate firm assets, which would leave the lenders with higher loss given default. Therefore,
such measures do not directly capture our construct of interest. Stated differently, we are
interested in capturing the threat of managerial private benefit consumption prior to insolvency,
not the ability of creditors to enforce debt contracts after insolvency. We discuss this issue
further in Section 6.5.
4.3.
General empirical setup
To test our predictions concerning the relation between cost of debt, income smoothing,
Accepted Article
and the threat of private benefits consumption in the contracting environment, we estimate the
following empirical model with country and year fixed effects:16
Spread i ,l   0  1 PBThreati   2 DiscSmoothi ,t   3 DiscSmooth * PBThreat
  4 FundSmoothi ,t   5 FundSmooth * PBThreat
(4)
 X i,t + Yi,l  country  year   i ,l ,
where Spread is the loan spread over LIBOR for firm i's loan facility l. PBThreat is an indicator
that equals one (zero) if firm i's environment suggests a high (low) threat of private benefits
consumption, as described above.17 We conduct supplemental analyses using firm-level threat
measures, which we subsequently describe along with the associated analyses. DiscSmooth
(FundSmooth) is firm i's discretionary income smoothing (fundamental earnings smoothness)
computed for the most recent fiscal year-end prior to entering the debt contract. We include
FundSmooth (and its interaction with PBThreat) to control for any effects of non-discretionary
smoothness on cost of debt. Our prediction that the information signaling view of smoothing is
16
We use two-way clustered standard errors by both nation and loan package throughout our analyses. Inferences
are not qualitatively altered if we replace the package-based clusters with either firm clusters or calendar month-year
clusters (e.g., January 2010, February 2010). Further, although we include industry fixed-effects in our smoothing
regressions (i.e., Eqs. 1 and 2), we consider a specification that includes industry fixed effects, and inferences are
unaltered. Accordingly, for parsimony we do not include industry fixed effects in our primary specification.
17
When using a country-level threat measure with country fixed effects, the main effect on PBThreat simply
captures the incremental average spread of an omitted country. Accordingly, including or excluding this main effect
does not change coefficient estimates on any other variables (including the interactions) in the regression.
This article is protected by copyright. All rights reserved.
dominant in environments with low threat of private benefits consumption suggests β2 < 0 (i.e.,
smoothing is associated with lower cost of debt). Our prediction that the information garbling
view of smoothing prevails in high threat environments suggests   2  3   0 (i.e., smoothing is
associated with higher cost of debt).
Xi,t is a vector of the following firm-level control variables, where t references the most
recent fiscal year-end prior to loan inception: LSize (natural log of total assets in U.S. dollars),
BTM (book to market ratio), Leverage (leverage, measured as the ratio of total liabilities to total
Accepted Article
assets), ROA (return on assets, measured as the ratio of earnings before interest and taxes to total
assets), Tangible (asset tangibility, measured as the ratio of property plant and equipment to total
assets), and StdRet (standard deviation of firm i's monthly returns). Yi,l is a vector of the
following loan-level control variables for firm i's loan facility l: Secure (an indicator that equals
one if the loan requires collateral), LMaturity (natural log of loan maturity in months), LFacility
(natural log of the loan facility face amount in U.S. dollars), and NCov (the number of financial
covenants attached to the loan package that contains facility l). All variables are defined in
Appendix A.
5.
Data and Descriptive Statistics
We obtain international data on bank loans from Dealscan.18 The most primitive unit of
observation is a loan facility, where multiple facilities can be included in a loan package between
a borrower and lender. Because each loan facility within a package typically has a unique
combination of terms (e.g., spread, maturity, collateral requirements, amount) and abstracting
away from this detail can lead to biased results, we follow a common approach in the accounting
literature and use a loan facility (rather than loan package) as our base unit of observation (e.g.,
18
International Dealscan loan data has been used extensively in extant accounting and finance literature (e.g., Qian
& Strahan, 2007; Bae & Goyal, 2009; Kim, Tsui, & Cheong, 2011; Lin et al., 2011).
This article is protected by copyright. All rights reserved.
Bharath et al., 2008; Amiram, Kalay, & Sadka, 2017; De Franco, Hope, & Lu, 2017), and
include loan package as a standard error clustering dimension. We collect accounting and stock
price data from Worldscope and Datastream, respectively, and convert all non-ratio variables
into U.S. dollars. We obtain data on firm-level probability of default (PD) from the Credit
Research Initiative (CRI) of the National University of Singapore.19
Our sample begins with 117,817 loan facilities in Dealscan with non-missing loan spread
with loan initiation dates ranging from 1996 to 2009, and where borrowers are not banks or
Accepted Article
utilities (two digit ICB codes 70, 83, 85, and 87). We next match the Dealscan sample to
Worldscope by company name, reducing our sample to 48,458 loan facilities. We retain in our
sample loans with expected maturity (at loan inception) of greater than twelve months (e.g.,
Demiroglu & James, 2010), further reducing our sample to 37,550 loan facilities. We next
eliminate observations that have missing values for our computed discretionary smoothing
metrics, leaving 16,410 facility-level observations. Because of the disproportionate number of
U.S. observations in Dealscan, for our primary country-level analysis we exclude all
observations for U.S. borrowers, reducing our sample to 4,588 facilities.20 Next, we truncate all
continuous variables used in our analyses at the lower 1% and upper 99% values by countryyear, and delete observations with missing values for loan characteristics (i.e., spread, maturity,
number of covenants, face amount), accounting data, or any other data necessary for our
analyses, as described below. This leaves us with 1,817 facility-level observations across 1,084
loan packages for 639 distinct non-U.S. borrowers across twenty countries.
19
The firm-level probability of default measures we employ are computed by the CRI using a forward default
intensity model, as outlined in Duan, Sun, & Wang (2012). Please refer to www.rmimcri.org for more details.
20
We estimate a firm-level analysis using a U.S. sample. We discuss the sample along with the associated results in
Section 6.2.
This article is protected by copyright. All rights reserved.
Table 1 presents details of the sample country distribution. The country distribution of
firms and facility-level observations are similar, with Taiwan and the United Kingdom
representing a large fraction of the sample (22% and 31% of facility-level observations,
respectively), where both countries are classified as having low threat of private benefit
consumption. Among countries classified as having high threat of private benefit consumption,
France and Germany are prominent within the sample (11% and 4% of facility-level
observations, respectively). We acknowledge that our sample composition reflects potential
Accepted Article
selection bias driven by two requirements: firms must exist in both Dealscan and Datastream
with data for all of our required variables, and we must be able to obtain valid matches across the
two datasets. This may limit the generalizability of our results to loans between relatively large
banks and large borrowers. Further, as the data only allow us to examine loans that were actually
issued, we note that our sample may underrepresent firms in countries where borrower access to
credit is limited because of generally poor contracting environments. However, these limitations
are standard in this literature (e.g., Qian & Strahan, 2007).
Table 2 presents descriptive statistics for firm and facililty-level variables. Mean total
assets in U.S. dollars is $3.8 billion (i.e., mean logged assets of 14.01). Mean book-to-market
ratio is 0.71, and mean leverage ratio is 0.59. The median sample loan has a 100 basis point
spread over LIBOR, a face amount of U.S. $140 million, a five-year maturity, no collateral
requirements and no financial covenants.21 Table 3 presents correlations for facility-level
observations.
21
When Dealscan reports no covenant data for a given loan package, we make the assumption that the number of
covenants on the loan package is zero. While this assumption is common in the literature, Drucker and Puri (2009)
points out that this assumption is questionable, particularly for non-U.S. loans. Throughout the study, inferences are
unchanged if we do not include the number of covenants in our analyses.
This article is protected by copyright. All rights reserved.
6.
Empirical Results
6.1.
Country-level threat of private benefits consumption
Table 4 presents results from estimation of Eq. (4). Column (3) presents results from our
primary specification, which tests our two key predictions. As predicted,  2 is significantly
negative (coefficient estimate 0.145; t-statistic 2.38), which documents a negative association
between income smoothing and cost of debt for firms within countries characterized by a low
threat of private benefits consumption. This is consistent with our prediction that, on average,
Accepted Article
lenders perceive smoothing to reflect signaling in such environments. In stark contrast, the
interaction between smoothing and the high threat indicator ( 3 ) is significantly positive
(coefficient estimate 0.410; t-statistic 2.86), resulting in a significantly positive total coefficient
on firm-level smoothing in high threat countries of 0.265 ( 2  3 ). As predicted, this provides
evidence of a positive association between smoothing and cost of debt for firms within countries
characterized by a high threat of private benefits consumption, consistent with lenders onaverage perceiving smoothing to reflect garbling in high threat countries.
Although statistically insignificant, there is a negative sign on the association between
non-discretionary earnings smoothness (FundSmooth) and loan spread, which is consistent with
such smoothness being reflective of lower risk. More importantly, consistent with our intuition
there is no difference in the association between non-discretionary smoothness and spread across
consumption threat regimes, as reflected by the insignificant FundSmooth*PBThreat interaction
(coefficient estimate -0.035; t-statistic -0.11). For parsimony, hereafter we omit this interaction
from our (country-level) empirical tests.
The results for the firm-level control variables are generally consistent with our
expectations. Market-to-book has been used by numerous studies as a proxy for conservatism
This article is protected by copyright. All rights reserved.
(e.g., Roychowdhury & Watts, 2007). Because book-to-market is decreasing in conservatism, the
positive coefficient on book-to-market (BTM) is consistent with literature that documents that ex
ante conservatism lowers cost of debt (e.g., Zhang, 2008). Larger (LSize) firms have lower cost
of debt. Firms with more leverage (Leverage) and volatility (StdRet) have higher cost of debt.
Consistent with Bharath et al. (2008) and Costello and Wittenberg-Moerman (2011), there is a
negative relation between loan amount (LFacility) and spread, and a positive association between
maturity (LMaturity) and spread. Further consistent with Bharath et al. (2008) and Berger and
Accepted Article
Udell (1990), there is a strong positive relation between collateral requirement (Secure) and
spread. These relations reflect a complex set of unobservable tradeoffs in the loan contracting
process.
In terms of economic significance, there is a material difference in the effects of
smoothing on cost of debt across low and high private benefit consumption threat environments.
In low threat countries, movement across the interquartile range of DiscSmooth results in an
approximate 7 basis point decrease in loan spread (i.e., interquartile range of 45 times the
coefficient estimate of 0.15), which represents a 7% decrease in spread relative to the median
sample spread. In high threat countries, movement across the interquartile range of DiscSmooth
results in an approximate 12 basis point increase in loan spread (i.e., interquartile range of 45
times the total coefficient estimate of 0.265), which represents a 12% increase in loan spread
relative to the median spread.22
Column (1) reports results from a more naive specification of the relation between
smoothing and cost of debt, where we do not partition based on private benefits consumption
threat, which indicates an insignificant relation between smoothing and cost of debt. However,
22
These effects are economically significant, with magnitudes comparable to effects documented in Bharath et al.
(2008). Specifically, Bharath et al. (2008) finds a 14 basis point increase in loan interest spread over LIBOR in
going from firms in the worst to best quintiles of accounting quality.
This article is protected by copyright. All rights reserved.
we now know from column (3) that this insignificance simply reflects the contrasting negative
and positive relations across environments. Accordingly, without considering the forces we
reveal, researchers may make inappropriate conclusions concerning the relation between
smoothing and cost of private debt.
Although our sample includes only a limited number of firms that receive loans more
than once during our sample period (i.e., 271 firms), we repeat our main test with firm fixed
effects. This design is extremely strict, in that all of the variation in the effect of smoothing on
Accepted Article
cost of debt comes from within-firm time series variation in smoothing using a very small
number of firms with multiple observations. Untabulated findings are nonetheless consistent with
our main inferences. That is, within firm there is a significant negative (positive) relation
between the extent of smoothing and cost of debt if that firm is operating in a low (high) threat
environments. This design further mitigates concern that our primary inferences are driven by
correlated omitted variables.
6.2.
Firm-level threat of private benefits consumption
Our primary results provide evidence that, on average, lenders interpret smoothing in a
manner consistent with the signaling (garbling) view of smoothing in countries with low (high)
threat of private benefits consumption. However, our results do not imply that smoothing is
associated with lower (higher) cost of debt for all firms in low (high) threat countries. To the
extent that lenders distinguish among firms within a given country in terms of their view of
smoothing, it is likely that there are firm-specific cases in low (high) threat countries where
smoothing is associated with an increased (decreased) cost of debt.
One issue with pursuing a within-country analysis based on firm-level threat
characteristics is that within a country, such firm-level characteristics are likely to be endogenous
This article is protected by copyright. All rights reserved.
(i.e., a firm can choose both smoothing and governance structure). However, with that caveat in
place, we estimate Eq. (4) using a time-varying firm-level measure of the threat of private
benefits consumption. That is, using our foreign sample we replace PBThreat with
CloseHeldSharesi,t, which is an indicator that equals one if firm i's percentage of closely held
shares in year t is in the top quartile of sample observations (i.e., greater than 45.75%), and
equals zero otherwise. Ownership concentration enables more efficient expropriation of private
benefits, thus a high level of closely held shares reflects relatively high threat of private benefits
Accepted Article
consumption (La Porta, Lopez-de-Silanes, Shleifer, & Vishny, 2000).23
Column (1) of Table 5 presents results from this analysis, which are remarkably similar
to results from our primary analysis using the country-level threat measure. These results provide
nice identification, in that the effect we document in our primary analysis likewise operates
within country based on time-varying firm-level threat characteristics. That is,  2 is significantly
negative (coefficient estimate 0.182; t-statistic 1.65), which documents a negative association
between income smoothing and cost of debt for firms that have low threat of private benefits
consumption, holding country-level threat characteristics constant. Moreover, there is a
significantly positive total coefficient on firm-level smoothing in high-threat firms of 0.471
( 2  3 ), which provides evidence of a positive association between smoothing and cost of debt.
To fix intuition with an example, these results illustrate that, whereas our primary analysis in
Table 4 suggests that smoothing firms in the U.K (a low threat country) tend to have lower cost
of debt on average, and that smoothing firms in Germany (a high threat country) tend have
23
We acknowledge that in some settings, ownership concentration can provide governance benefits to a firm.
However, the prevailing interpretation in the literature is that ownership concentration enables private benefits
extraction.
This article is protected by copyright. All rights reserved.
higher cost of debt on average, within either the U.K. or Germany smoothing firms with low
(high) private benefits consumption threat have relatively low (high) cost of debt.
We repeat our firm-level threat analysis for a separate sample of U.S. firms for which
Bebchuk et al. (2009) compute a firm-level composite managerial entrenchment index.24 The
entrenchment index reflects the potential for the threat of private benefits consumption using six
Investor Responsibility Research Center provisions (accordingly, the measure ranges from zero
to six) (e.g., John, Litov, & Yeung, 2008; Rego & Wilson, 2012). Specifically, we construct an
Accepted Article
indicator MgrEntrenchi,t that equals one if firm i's entrenchment index in year t is greater than
three, and equals zero otherwise.25 The U.S. sample we use in this analysis consists of 6,033
facility-level observations across 4,570 loan packages for 1,453 distinct borrowers. As reported
in column (2) of Table 5, inferences from the within-U.S. firm-level analysis mirror those from
our primary analysis.
6.3.
Borrower credit risk
Our results thus far suggest that lenders adjust loan spread based on their interpretation of
the implications of observed smoothing for expected credit loss. However, because lenders only
lose money if a borrower actually defaults, lenders may be relatively insensitive to the positive or
negative implications of smoothing for borrowers with relatively low credit risk. In high private
benefits consumption threat environments, if it is more likely that a borrower will default,
lenders may be more concerned with the implications of private benefits consumption because
such actions will more likely affect their ultimate payoff. Accordingly, we predict that the
positive association between smoothing and cost of debt in high threat countries will be
24
The U.S. is a low threat country, having an anti-self-dealing index of 0.65.
We note that, in contrast to our primary measure based on the anti-self-dealing index, literature has shown that
managerial entrenchment can benefit debtholders by providing takeover defenses (in the presence of shareholderdebtholder conflicts) (e.g., Ashbaugh-Skaife, Collins, & LaFond, 2006). However, as this is a limited setting, we
rely on the more general interpretation of entrenchment as enabling private benefits extraction.
25
This article is protected by copyright. All rights reserved.
exacerbated if the borrower has relatively high risk of default. To test this prediction, we
estimate the following modification to Eq. (4), which includes our new variable of interest,
DiscSmooth*PBThreat*PD, where PD is the borrower’s one-year-ahead probability of default as
of the end of the month immediately preceding the loan contracting date:
Spreadi ,l   0  1 PBThreati   2 DiscSmoothi ,t   3 DiscSmooth * PBThreat
  4 DiscSmooth * PBThreat * PD  5 FundSmoothi ,t   6 PDi ,t
  7 DiscSmooth * PD  8 FundSmooth * PD  9 PBThreat * PD
(5)
 X i,t + Yi,l  country  year   i ,l ,
Accepted Article
where all variables are as previously defined, and DiscSmooth*PBThreat*PD captures the
incremental effect of probability of default on the relation between smoothing and loan spread in
high consumption threat environments, which we predict will be positive.
We report results in Table 6. As predicted, the coefficient on DiscSmooth*PBThreat*PD
is positive and significant (coefficient estimate 0.793; t-stat 2.14). Because PD is a continuous
measure, all lower order coefficients are technically uninterpretable; however, signs and
significance levels are consistent with prior results. For example, the coefficient on DiscSmooth
is significantly negative, reinforcing the result that smoothing reduces spread in countries with a
low threat of private benefits consumption. To summarize, these results provide evidence that
lenders pay particular attention to smoothing and its implications for potential private benefits
consumption for firms that are closer to default, particularly when smoothing more likely reflects
garbling which may increase loss given default.
6.4.
Non-price loan terms
Although our interest in this paper is the relation between smoothing and the cost of debt,
we analyze whether income smoothing differentially affects several non-price loan terms that
lenders may adjust to reflect perceived borrower risk across environments, including financial
This article is protected by copyright. All rights reserved.
covenants, maturity, and collateral requirements. To do so, we use the basic variable structure of
Eq. (4), while changing the dependent variable and estimation approach as appropriate.
Conceptually, there are competing forces concerning the effect of smoothing on the use
of financial covenants. Consider environments with high threat of private benefits
consumption—if smoothing represents garbling, then lenders may want to increase the use of
financial covenants to provide further protections; however, garbling suggests that the resulting
financial data are less reflective of underlying economics, which suggests that lenders may want
Accepted Article
to rely less on financial covenants (e.g., Costello and Wittenberg-Moerman, 2011).
Aside from this tension, investigating the effect of smoothing on the inclusion of loan
covenants is problematic because of international Dealscan data issues (Drucker & Puri, 2009).
In particular, in Dealscan the absence of covenant data does not necessarily imply that no loan
covenants exist, and this problem is particularly acute for the non-U.S. Dealscan sample.
Therefore, to empirically explore this relation we first delete all observations with no covenant
data, and estimate a variation of Eq. (4) with the number of financial covenants attached to the
loan (NCov) as the dependent variable using the resulting sample of 724 observations. As
reported in Table 7, we find that smoothing is indeed positively associated with the number of
covenants attached to the loan in environments with high threat of private benefits consumption
(coefficient estimate on DiscSmooth*Threat of 0.007; t-statistic 2.15). This result suggests that in
addition to increasing loan spread, lenders may also use increased covenant protections when
smoothing more likely reflects garbling to facilitate private benefits consumption.
To examine whether smoothing affects loan maturity (collateral requirements), we
estimate an OLS model of the form used in Eq. (4) with LMaturity (Secure) as the dependent
variable. To summarize, in untabulated results we find no statistically significant evidence that
This article is protected by copyright. All rights reserved.
the threat of private benefits consumption in the contracting environment affects the relation
between smoothing and either maturity or collateral requirements.26
6.5.
Additional considerations
6.5.1. Private benefits consumption vs. creditor rights
As discussed previously, our construct of interest is the threat that firm insiders will
consume private benefits as the firm continues its ongoing operations. Our construct of interest is
not creditor rights in the event of default, as such creditor protections would do little to alleviate
Accepted Article
a lender's concern that managers may expropriate firm assets prior to default, which would leave
the lenders with fewer recoverable assets in default. Therefore, we do not view creditor rights as
a meaningful delineator of the effect of smoothing on cost of debt. Accordingly, we predict that
there will be no differential association between smoothing and loan spread across partitions
based on the degree of creditor rights that exist in the event of default.
To test this prediction, we repeat our primary analysis after replacing PBThreat with
WeakCredRightsi, an indicator variable that equals one if firm i's country-level creditor rights
index (La Porta, Lopez-de-Silanes, Shleifer, & Vishny, 1998; Djankov et al., 2007) is zero or
one, and equals zero if the index is two, three or four. We then repeat our primary analysis after
adding WeakCredRights and its interactions to our main specification. As reported in column (1)
of Table 8, using creditor rights to partition the effect of smoothing on cost of debt is essentially
equivalent to partitioning on noise, i.e., we recover the apparent result from column (1) of Table
4 that there is no relation between smoothing and cost of debt. Column (2) of Table 5 reports
results from a specification that includes smoothing interactions with both creditor rights and
private benefits consumption threat. As reported, our main result obtains, in that there is a
26
In addition to equation-by-equation estimation using different loan terms as dependent variables, we estimate the
equations simultaneously using seemingly unrelated regression, and find inferences consistent with our reported
results across all loan terms we consider, including Spread and NCov.
This article is protected by copyright. All rights reserved.
differential relation between smoothing and cost of debt across private benefits consumption
threat environments, even after controlling for the effect of creditor rights (which remains
insignificant).
6.5.2. Other accounting attributes
Extant literature documents a negative relation between cost of debt and both accruals
quality and accounting conservatism (e.g., Bharath et al., 2008; Zhang, 2008). Further, extant
literature documents that different accounting system characteristics, including accruals quality,
Accepted Article
conservatism, and smoothness, are correlated with one another (e.g., Dechow et al., 2010).
To verify that our results are indeed capturing effects of smoothing and not an alternative
(correlated) accounting attribute, we estimate a modification to Eq. (4) where we replace
DiscSmooth with a measure of accruals quality (AQ) computed following the approach of
Dechow and Dichev (2002), and estimate a specification that includes both DiscSmooth and AQ
and their interactions with PBThreat.27 As reported in Column (1) of Table 9, we find a negative
relation between AQ and Spread, consistent with Bharath et al. (2008), and this negative relation
does not vary across private benefits consumption threat environments (as indicated by the
insignificance of AQ*PBThreat). Further, our key inference that DiscSmooth changes sign in its
relation with Spread across threat environments remains.
We repeat (separately) this modification of Eq. (4) with both BTM (as an inverse proxy
for conservatism) and the standard deviation of NIEXS computed over the three-to-five year
horizon ending in year t as an inverse proxy for earnings persistence (Dichev & Tang, 2009). As
with AQ, the signs of the relations between DiscSmooth and these alternative attributes
(conservatism and earnings persistence) do not vary across threat environments. Moreover, our
primary inferences regarding DiscSmooth hold in the presence of these alternative attributes and
27
Data requirements for computing AQ cause a significant decrease in our sample size, to 976 observations.
This article is protected by copyright. All rights reserved.
their interactions with PBThreat. Accordingly, we conclude that our inferences are indeed
attributable to smoothing, and not an alternative correlated accounting attribute.
6.5.3. Measurement error in smoothness proxies
Extant literature presents tests that lend construct validity to the smoothing measure we
use in this study (e.g., Lang & Maffett, 2011b; Lang et al., 2012). However, we acknowledge
that this proxy for smoothing may contain significant measurement error. Given that
measurement error typically introduces attenuation bias in coefficient estimates, any such
Accepted Article
measurement error would work against our finding significant relations between smoothing and
cost of debt. A separate but related concern is whether the relation between smoothing and cost
of debt is driven by correlated omitted variables. In order for omitted variables to be the driver of
our results, it would be necessary for the omitted variables to be negatively associated with
smoothing in low consumption threat environments, but positively associated with smoothing in
high consumption threat environments. We view this as an unlikely explanation for our results.
6.5.4. Smoothing motivations
Although we position our study within the signaling versus garbling framework of
smoothing motivations, we recognize that there exist other smoothing motivations, such as
removing risk from earnings-based compensation contracts and tax minimization. However, it is
unlikely that these motivations differ across high and low private benefit consumption threat
environments in a manner that would affect our inferences. Relatedly, smoothing may be
accomplished by methods aside from accounting choice, such as real operational decisions
(Fudenberg & Tirole, 1995), hedging via derivative use (Barton, 2001), or asset sales (Black et
al., 1998; Peasnell,, 1998). In our framework, smoothing that results from these activities is
likely captured by the discretionary smoothing component. We do not attempt to disentangle the
This article is protected by copyright. All rights reserved.
sources of smoothing. Because smoothing via these alternative mechanisms can likewise be used
to either signal or garble, the logic of our study applies regardless of the smoothing mechanism.
6.5.5. Relationship lending
As pointed out in extant literature (e.g., Bharath et al., 2008; Sufi, 2007), lenders with
prior relationships with borrowers likely have both extensive operational information about the
borrower and well-developed information channels with firm insiders. Thus, existing
relationships could lead to mitigated effects of smoothing on loan terms. However, as also
Accepted Article
pointed out in extant literature (e.g., Rajan, 1992; Murfin, 2012), relationships between
borrowers and lenders may lead to lender ability to extract information rents. Following this
logic, if smoothing creates opacity as under the garbling view, information rents may increase,
enabling lenders to increase loan spread. Therefore, the effects of relationship banking on the
association between smoothing on loan spread is an empirical question. In untabulated analysis,
we find no difference in the effect of smoothing on loan spread for relationship versus nonrelationship loans. This is consistent with findings in Costello and Wittenberg-Moerman (2011)
who find no difference in the effects of internal control weaknesses on financial covenant use for
relationship versus non-relationship loans.
7.
Conclusion
Income smoothing by managers is a pervasive phenomenon that has been widely
researched. Despite the fact that private debt markets provide a major source of financing used
by most corporations, we have incomplete evidence on how smoothing is associated with cost of
debt in the private loan market. The institutional factors associated with private loan contracts,
combined with the theoretical motivations for smoothing, make it unclear ex ante whether
smoothing will be positively, negatively, or not associated with loan spread. In this study, we fill
This article is protected by copyright. All rights reserved.
this gap in the literature. Extant evidence regarding smoothing in the credit markets is
incomplete, and presents a one-sided inference that income smoothing lowers cost of debt. This
state of the literature is puzzling, given that there are two coexisting views of smoothing that
theoretically suggest opposite signs in the association between smoothing and cost of debt. That
is, the information signaling view of smoothing suggests a negative association, and the
information garbling view of smoothing suggests a positive association.
Our conjecture is that lenders’ view of smoothing is a function of the extent of the threat
Accepted Article
of managerial private benefits consumption in the contracting environment. In high (low) threat
environments, we predict that lenders are more likely to take the information garbling (signaling)
view of smoothing. Consistent with our predictions, we provide evidence that smoothing is
associated with lower cost of debt when the threat of private benefits consumption by managers
is low, and is associated with higher cost of debt when the threat of private benefits consumption
by managers is high. In so doing, our study is the first to document a positive association
between smoothing and cost of debt, and we are the first to identify a feature of the contracting
environment that empirically reveals the sign reversal in the association.
We use an international sample of private loans so that we can obtain plausibly
exogenous variation in the threat of private benefits consumption in the contracting environment.
Accordingly, our study is subject to the typical small sample concerns that plague international
debt research, which leads to concerns about generalizability of our findings. However, we find
consistent inferences using both country-level and firm-level threat measures in our primary nonU.S. sample, as well as using firm-level measures in a much larger U.S.-only sample. The
consistency in inference using these multiple approaches and samples provides some comfort
with regard to these concerns. In summary, notwithstanding the study’s natural limitations, we
This article is protected by copyright. All rights reserved.
contribute important new evidence both to the debt contracting literature and to the literature that
examines the effects of income smoothing in international capital markets.
Appendix A - Variable definitions
Variables prefixed by DS-, WS- and DL- are the mnemonic identifiers of the raw data items obtained
from Datastream Advance, Worldscope, and Dealscan, respectively. Subscripts i, t, and l refer to firm,
fiscal year, and loan facility, respectively.
Accruals,i,t
Accepted Article
AccrualsS,i,t
AQi,t
AvgOpCashi,t
AvgSalesGrowthi,t
BTMi,t
CloseHeldSharesi,t
DiscSmoothi,t
DiscSmooth1i,t
DiscSmooth2i,t
Facilityi.l
FundSmoothi,t
FundSmooth1i,t
FundSmooth2i,t
Leveragei,t
LFacilityi,l
change in current assets (WS-WC02201) minus change in cash (WS-WC02001)
minus change in current liabilities (WS-WC03101) plus change in short-term debt
(WS-WC03051) minus depreciation and amortization (WS-WC01151)
Accruals,i,t scaled by lagged total assets (TAi,t-1)
"accrual quality," measured as negative one times the standard deviation of firm
i's residuals from a regression of AccrualsS,i,t on OpCashi,t-1, OpCashi,t and
OpCashi,t+1 using no fewer than three nor more than five residuals.
average operating cash flow (OpCash) over the three-to-five year horizon ending
in year t
average sales growth (SalesGrowth) over the three-to-five years ending in year t
book-to-market, measured as total assets (WS-WC02999) minus total liabilities
(WS-WC03351), divided by market value of equity (WS-MV)
firm-level indicator of the threat of private benefits consumption; an indicator that
equals one if firm i's percentage of closely held shares in year t is in the top
quartile of sample observations (i.e., greater than 45.75%), and zero otherwise
composite measure of discretionary income smoothing, calculated as the average
percentile ranking (by country) of DiscSmooth1 and DiscSmooth2
residual from the panel regression of Smooth1i,t on industry (two-digit WS-ICB)
and fiscal year fixed effects and the following variables for fiscal year t: LSize,
Leverage, BTM, StdSales, %Loss, OpCycle, OpLev, AvgSalesGrowth and
AvgOpCash
the residual from the panel regression of Smooth2i,t on industry (two-digit WSICB) and fiscal year fixed effects and the following variables for fiscal year t:
LSize, Leverage, BTM, StdSales, %Loss, OpCycle, OpLev, AvgSalesGrowth and
AvgOpCash
face amount of loan facility l (DL-facilityamt), in millions of U.S. dollars
composite measure of fundamental earnings smoothness, calculated as the average
percentile ranking (by country) of FundSmooth1 and FundSmooth2
predicted value from the panel regression of SMTH1i,t on industry (two-digit WSICB) and fiscal year fixed effects and the following variables for fiscal year t:
LSize, Leverage, BTM, StdSales, %Loss, OpCycle, OpLev, AvgSalesGrowth and
AvgOpCash
predicted value from the panel regression of SMTH2i,t on industry (two-digit WSICB) and fiscal year fixed effects and the following variables for fiscal year t:
LSize, Leverage, BTM, StdSales, %Loss, OpCycle, OpLev, AvgSalesGrowth and
AvgOpCash
leverage, measured as total liabilities (WS-WC03351) divided by total assets
(WS-WC02999)
natural logarithm of Facility
This article is protected by copyright. All rights reserved.
LMaturityi,l
LSizei,t
Maturityi,l
MgrEntrenchi,t
NCovi,l
NIEXSi,t
σNIEXSi,t
OpCashi,t
Accepted Article
OpCyclei,t
OpLevi,t
PBThreati
%Lossi,t
PDi,m
ROAi,t
Securei,l
SalesGrowthi,t
Smooth1i,t
Smooth2i,t
Spreadi,l
StdReti,l
StdSalesi,t
TAi,t
Tangiblei,t
natural logarithm of Maturity
natural log of total assets (WS-WC02999) in U.S. dollars
the term in months of loan facility l (DL-maturity)
firm-level measure of the threat of private benefits consumption; an indicator that
equals one if firm i's Bebchuk et al. (2009) entrenchment index in year t is greater
than three, and zero otherwise.
number of distinct financial and net worth covenants attached to the loan facility
l's loan package
net income before extraordinary items (WS-WC01551) scaled by lagged total
assets (TAi,t-1)
standard deviation of NIEXS computed over the three-to-five year horizon ending
in year t
operating cash flow, measured as net income before extraordinary items (WSWC01551) minus Accruals, scaled by lagged total assets (TAi,t-1)
operating cycle, measured as the natural logarithm of ((average accounts
receivable/sales)*360 + (average inventory/cost of goods sold)*360); accounts
receivable (WS-WC02051), sales (WS-WC01001), inventory (WS-WC02101),
cost of goods sold (WS-WC01051)
operating leverage, measured as property, plant and equipment (WS-WC02501),
divided by total assets (TA)
country-level indicator of the threat of private benefits consumption; an indicator
that equals one if the Djankov et al. (2008) anti-self-dealing index of firm i's
country is below the sample observation median, and zero otherwise
percentage of years where net income before extraordinary items (WS-WC01551)
is less than zero over the three-to-five year horizon ending in year t
firm i's one-year ahead probability of default as of month m, obtained from the
Credit Risk Institute of the National University of Singapore (NUS CRI). The
default probabilities are estimates using a forward intensity model, as outlined in
Duan et al. (2012).
earnings before interest and taxes divided by total assets (WS-WC02999)
an indicator variable that equals one if loan facility l requires collateral, and zero
otherwise (DL-secured)
sales growth, measured as percentage change in sales (WS-WC01001) from year
t-1 to t
(standard deviation of NIEXS divided by the standard deviation of OpCash)
multiplied by -1, where the standard deviations are computed over the three-tofive year horizon ending in year t
correlation between OpCash and AccrualsS multiplied by -1, where the
correlation is computed over the three-to-five year horizon ending in year t
interest rate on loan facility l in excess of LIBOR, in basis points (DL-allindrawn)
standard deviation of monthly return (computed from DS-ret_index) for firm i
over the twelve-month horizon immediately preceding loan facility l
standard deviation of sales (WS-WC01001) over the three-to-five year horizon
ending in year t
total assets (WS-WC02999)
asset tangibility, measured as property, plant and equipment (WS-WC02501),
scaled by total assets, TA
This article is protected by copyright. All rights reserved.
WeakCredRightsi
country-level measure of creditor rights given default, based on the country-level
creditor rights index developed in La Porta et al. (1998) and Djankov et al. (2007),
which ranges from 0 to 4; our variable is an indicator that equals one if firm i's
country-level creditor rights index is 0 or 1, and zero otherwise.
Appendix B - Stylized framework that illustrates our key predictions
In this appendix we describe a partial equilibrium framework that provides one example
of how the forces we describe in this paper interact to yield our predictions under a set of
plausible assumptions.28 We are not attempting to create a complete analytical model of a private
Accepted Article
lending interaction. For example, the framework does not include features such as private
information flows, financial covenants, etc. Rather, we use this framework to simply demonstrate
key forces that can plausibly lead to our predictions concerning the sign reversal in the
association between smoothing and loan spread. As discussed in the paper, these institutional
features of the private debt setting provide some tension concerning our predicted associations.
Setup
Consider an existing all-equity firm with a self-interested manager/owner, i.e., the
manager has an incentive to maximize her own utility, where the firm has no minority
shareholders (although this framework can easily generalize to a case where there are minority
shareholders). At time t = 0, the manager knows that at time t = 1, the firm will require additional
financing in the amount $K to fund a project, which we assume will be obtained in the form of
debt with interest rate I.29 At time t = 2, the project outcome is realized, with probability of
failure (success) PD (1PD). If successful, the project has a gross percentage return R greater
28
For example, an alternative framework could consider a relationship between minority and controlling
shareholders, where smoothing is done for the same reasons that are described below. The lending decision could
then come later, and exogenously to the smoothing decision by the borrower (i.e., the lender observes smoothing
that is the result of the interactions between controlling shareholders and minority shareholders and responds to the
observed smoothing). The predictions of this alternative framework regarding the sign reversal in the relation
between smoothing and cost of debt will be identical to those we present below.
29
Debt can be the optimal financing source for a variety of reasons, e.g., tax benefits.
This article is protected by copyright. All rights reserved.
than the risk-free rate (which we normalize to zero for convenience), the lender is repaid, and the
firm continues. If the project fails, the firm defaults and the lender receives gross recovery rate V
as a percentage of K.
At time t = 0, nature endows the manager with two key things that are unobservable to
external capital suppliers. First, with probability P  (0,1) the manager receives private
information that the firm's economic earnings are less volatile than reported cash flows suggest
(thus, setting up the possibility that the manager can signal using income smoothing, assuming
Accepted Article
she has the ability to smooth). Without loss of generality, we characterize firms in binary fashion
as either having smooth economic earnings (denoted as 1 ) or not (denoted as  0 ). Further, we
assume that smooth economic earnings lowers probability of default, i.e. 0  PD1  PD 0  1
(Merton, 1974). Second, with probability P   0,1 nature endows the manager with the ability
to smooth reported income.30 This ability could relate to either managerial skill, or innate
smoothness of underlying cash flows (i.e., if cash flows are innately smooth, there will naturally
be very little a manager can do to further smooth earnings by applying discretion). Without loss
of generality, we simply characterize managers in binary fashion as either able to smooth
(denoted as 1 ) or unable to smooth (denoted as 0 ) (e.g., Trueman & Titman, 1988).31
Accordingly, there are four types of managers across these two unobservable dimensions, which
we denote as follows:   11;10 ; 01; 00  . For example, type   10 refers to a
manager that knows economic earnings are relatively smooth, but is unable to smooth reported
income to signal that information.
30
Under certain values of other parameters that can be characterized, the introduction of this "ability" parameter is
not necessary. However, it makes the intuition of the framework easier to follow.
31
Alternatively, we could assume differences in smoothing ability across a continuum, or differences in managerspecific costs of smoothing.
This article is protected by copyright. All rights reserved.
At a time between t = 0 and t = 1 (for simplicity we will call it t = 0.5), the manager
chooses whether or not to consume private benefits, and whether or not to smooth earnings, and
then externally reports financial results (i.e., earnings, cash flows and accruals). These choices
are made with the understanding that she will pursue a loan at t=1, and that lenders will price
protect against any anticipated private benefits consumption that extracts firm value. Without
loss of generality, we assume if private benefits are consumed, they are consumed as a
percentage of firm value that translates into a percentage B   0,1 of the loan K (i.e., $ B  K ).
Accepted Article
Further, assume that the choice to smooth earnings can provide the manager with private benefits
that do not extract firm wealth from external capital providers and are therefore not price
protected, for example, power, respect, credibility, and connections (e.g., Demsetz & Lehn,
1985; Aghion & Bolton, 1992).32 That is, importantly, these additional benefits will leave an
incentive to smooth even if capital providers fully price protect against the anticipated private
benefits which the manager consumes to their potential detriment. We denote these additional
benefits as  , and for simplicity hereafter will refer to them simply as "non-value-extractive"
private benefits.
For notational convenience we denote the choice to consume private benefits as B1 and
the choice to not consume private benefits as B0. We characterize the firm's environment as
having either low threat (LT) or high threat (HT) of private benefits consumption (which is
observable to capital providers), where low (high) threat of consumption is an environment
where there are heavy (light) penalties if "caught" (we denote the probability of being caught as
PC). We capture this threat with a punishment parameter    LT ,  HT  that reflects the cost to
32
Alternatively, smoothing can provide managers with other pecuniary benefits that are not price protected by
external capital providers, for example, smoothing can lower the firm's total tax obligations, which increases the size
of the pie for the manager and capital suppliers (e.g., Hepworth, 1953; Graham & Smith, 1999).
This article is protected by copyright. All rights reserved.
the manager of being caught, which without loss of generality we express as a percentage of the
amount consumed, where 1   HT   LT   . That is, the expected cost to the manager of
consuming private benefits is $   B  K  PC  . We assume that the minimum punishment if
caught is repayment of the amount consumed plus a slap on the wrist, i.e.,  HT  1   , where
  0 . Heavier penalties that characterize low threat environments include more substantial
fines, reputational costs, criminal penalties, etc. Although we could characterize the point above
Accepted Article
which the punishment is "high enough", for simplicity we assume that if the manager gets caught
consuming private benefits in a low threat environment,  LT   .
If the manager has the ability to smooth income (i.e., type 1 ), the manager chooses to
smooth income (S = 1, denoted S1) or not (S = 0, denoted S0 ) at time t = 0.5. This choice is made
with the manager's understanding that there is a positive probability at t = 2 that both α and  will
be revealed, and that there will be an associated cost of lying (L > 0) if the manager had the
ability to smooth ( 1 ) but lied about her true type α (i.e., 1 and S0, or  0 and S1) (Desai et al.,
2006). Consistent with our characterization of costs of private benefits consumption, we assume
that LHT  LLT .33 Although we could characterize the point below (above) which the cost is low
(high) enough, for simplicity we set LHT    0 ( LLT   ). We assume that if the manager
chooses to consume private benefits, income smoothing reduces the probability of being caught
(Fudenberg & Tirole, 1995). Further, for simplicity we assume that if the manager consumes
private benefits but does not smooth, she will be caught. That is, 0  PCS1  PCS 0  1. After the
manager chooses B and S at t = 0.5, she reports financials.
33
Alternatively, to reduce the number of parameters we could have used the punishment parameter to describe the
misreporting costs.
This article is protected by copyright. All rights reserved.
The manager's choice of private benefits consumption and smoothing
Based on our binary exposition, there are four B-S choice combinations, which we denote
 BS  B0 S1; B0 S0 ; B1S1; B1S0  . For example, type BS  B0 S1 refers to a manager that smooths
income and does not consume value-extractive private benefits. At time t = 0.5, the manager
chooses  BS to maximize her payoff, subject to her endowed (unobservable) type
  11;10 ; 01; 00  and (observable)    LT ,  HT  . Table A1 outlines the payoff
Accepted Article
structures to the manager under each scenario. Note that the manager's general payoff equals the
sum of five terms:
(1  PD)  K  ( R  I )  ( PD  B  K )  (  B  K  PC )  L  .
(A1)
The first term, (1  PD)  K  ( R  I ) , is the expected payoff if the project succeeds. The second
term, ( PD  B  K ) , is the expected payoff if the project fails (i.e., the manager receives the
amount of private benefits consumed). The third term, (  B  K  PC ) , is the expected cost of
consuming private benefits. The fourth term, L, is the expected cost of lying about whether the
manager has smooth economic earnings (e.g., Desai et al., 2006). The fifth term,  , is the
additional non-value-extractive private benefits obtained from smoothing.
Consider a manager operating in a high threat of consumption environment, as depicted
in Panel A of Table A1. For endowment   11 , choice 2 dominates choice 4 (because the
sum of terms two and three is zero in choice 2 and negative in choice 4), choice 3 dominates
choice 1 (because the sum of terms two and three is positive in choice 3 and zero in choice 1).
The determination of whether choice 2 or 3 is optimal depends on the net benefit of private
benefits consumption (the sum of terms two, three, and five in choice 3) relative to the interest
rate differential reflected in the first term in both choices. It is straightforward to show (as
This article is protected by copyright. All rights reserved.
outlined below) that in setting interest rates, the lender will fully price protect against anticipated
private benefits consumption, and the borrower will be indifferent between choices 2 and 3
before considering additional non-value-extractive private benefits,  . Therefore, it follows
directly that for any positive  , the borrower will choose to smooth and consume, even in the
face of full lender price protection against private benefits consumption. Specifically, setting the
borrower's choice 2 and 3 payoff functions equal, the borrower will choose choice 3 (smooth and
Accepted Article
consume) if I S 1  I S 0 
PD  B

. As we show in the section below, based on full lender

1  PD 1  PD
price protection, I S1  I S 0 
PD  B
. Thus, the borrower will find choice 3 optimal for any
1  PD
positive  .
Continuing across the other endowment possibilities, it follows that the optimal choice
for   10 is 2, for   01 is 2 or 3, and for   00 is 2. As depicted in Panel B,
following the same approach in analyzing the manager's choice in the low consumption threat
environment, the optimal choice for   11 is 1,   10 is 2, for   01 is 2, and for
  00 is 2. This structure is understood by the lender, who will use these insights when
choosing the interest rate I.
The lender's pricing decision
At time t = 1, the lender chooses the interest rate I on the loan amount $K. Theoretically,
I will be increasing in expected cost of default, i.e., (probability of default)*(loss given default).
Literature has established that smoother economic earnings imply lower probability of default
(e.g., Merton, 1974). That is, PD1  PD 0 . Further, it is straightforward that managerial
consumption of private benefits from capital providers increases loss given default, i.e.,
This article is protected by copyright. All rights reserved.
decreases the recovery rate in the event of default (V = R-B). That is, VB1  VB0 . Accordingly, the
lender would like to base I on the smoothness of economic earnings and the extent of managerial
private benefit consumption, but neither factor is directly observable at the contracting date.
Therefore, the lender will base I on the observable private benefits consumption threat (  ) and
the smoothness of earnings (S), where smooth earnings may reflect either signaling or garbling.
Based on our binary characterizations, there are four possible interest rates, which we denote as


Accepted Article
I  S  I  HT S 0 ; I  HT S 1 ; I  LT S 0 ; I  LT S 1 .
Consider the high consumption threat environment. It follows from the analysis outlined
in Panel A of Table A1 that if the manager smooths income (S1), the lender knows that the
manager consumed private benefits.34 Further, if the manager does not smooth income (S0), the
lender knows the manager did not consume private benefits. In either case (S1 or S0), the lender
will assess that the firm has smooth economic earnings ( 1 ) with probability P and that the
firm does not have smooth economic earnings (  0 ) with probability ( 1  P ). It is
straightforward to then show that I HT ,S1  I HT ,S 0 because VB1  VB0 . Our prediction that
smoothing is positively associated with loan spread in countries with high threat of private
benefits consumption follows immediately. More formally, we can solve for the interest rate
differential I HT ,S1  I HT , S 0 . Consider again endowment   11 . The lender will set rates to
equalize expected profit across choices 2 and 3, where the general form of the lender's profit
equals (1  PD)  K  (1  I )  PD  K  ( R  B)  K . Setting this lender profit function equal across
choice 2 and 3, it is straightforward to show that I HT , S1  I HT , S 0 
34
PD  B
0 .
1  PD
We make an implicit assumption that lenders will have the expectation that if managers consumed private benefits
before loan initiation, they will continue to do so after receiving the loan at t = 1.
This article is protected by copyright. All rights reserved.
Next, consider the low consumption threat environment. It follows from the analysis
outlined in Panel B of Table A1 that the manager will never consume private benefits. If the
manager smooths income (S1), the lender knows the firm has smooth economic earnings ( 1 )
with probability 1.0. If the manager does not smooth income (S0), the lender does not know
whether the firm has smooth economic earnings or not, and assesses the probability of smooth
economic earnings at less than 1.0. It is then straightforward to show that I LT ,S1  I LT , S 0 , because
Accepted Article
PD1  PD 0 . Our prediction that smoothing is negatively associated with loan spread in
countries with low threat of private benefits consumption follows immediately. More formally,
in the low threat environment, setting the lender's profit function equal across borrower
smoothing
choices
yields
the
following
(1  PD1 )  K  (1  I S1 )  PD1 (K  R)  K  (1  PD 0 )  K  (1  I S 0 )  PD 0 (K  R)  K .
expression:
After
 1  PD 0  1  R  PD 1  PD 0 
algebraic simplification: I S1  I S 0  
. Because PD1  PD 0 ,

1

PD
1

PD
1 
1

the term on the right hand side is negative, and the ratio multiplier attached to IS0 on the left hand
 1  PD 0 
side is less than one. Therefore, if I S1  I S 0  
  0 , it must be the case that I S1  I S 0  0 ,
 1  PD 1 
thus I S1  I S 0 .
This article is protected by copyright. All rights reserved.
Table A1
Framework-based predictions of managerial private benefits consumption and smoothing choice
This table outlines payoffs to a manager who chooses whether to consume private benefits and smooth reported
income prior to obtaining a loan to undertake a project. B0 (B1) denotes her choice to not consume (consume) private
benefits. S0 (S1) denotes her choice to not smooth (smooth) reported income. α0 (α1) denotes her endowment of
private information that economic earnings are not smooth (smooth). λ0 (λ1) denotes her endowment of the inability
(ability)
to
smooth
reported
income.
The
general
structure
of
her
payoff
is
(1  PD)  K  ( R  I )  ( PD  B  K )  (  B  K  PC )  L   . PD is the probability of project default. K is the dollar
amount borrowed. R is the project's gross percentage return. I is the interest rate charged by the lender. B is the
amount of private benefits she consumes. γ is the punishment parameter for consuming private benefits. PC is the
probability of getting caught consuming private benefits. L is the cost of lying about her type α. φ is an additional
smoothing benefit that accrues to the firm. * denotes non-dominated strategies under each possible
threat/endowment combination.
Panel A: Payoffs to manager in high threat of private benefits consumption
 HT  1   ; LHT   ; PCS1   ; PCS 0  1
Accepted Article
Endowment:   11 (has smooth economic earnings and can smooth reported income)
Choice 1: BS  B0 S1
(1  PD)  K  ( R  I HT ,S1 )  0  0  0  
Choice 2:  BS  B0 S0 *
(1  PD)  K  ( R  I HT ,S 0 )  0  0  
Choice 3: BS  B1S1 *
(1  PD)  K  ( R  I HT ,S1 )  PD  B  K  (1   )  B  K    0  
Choice 4: BS  B1S0
(1  PD)  K  ( R  I HT ,S 0 )  PD  B  K  (1   )  B  K 1 
Endowment:   10 (has smooth economic earnings but cannot smooth reported income)
Choice 1: BS  B0 S1
N/A (cannot have S1 for type 0 )
Choice 2:  BS  B0 S0 *
(1  PD)  K  ( R  I HT ,S 0 )  0  0  0
Choice 3: BS  B1S1
N/A (cannot have S1 for type 0 )
Choice 4: BS  B1S0
(1  PD)  K  ( R  I HT ,S 0 )  PD  B  K  (1   )  B  K 1 0
Endowment:   01 (does not have smooth economic earnings but can smooth reported income)
Choice 1: BS  B0 S1
(1  PD)  K  ( R  I HT ,S1 )  0  0    
Choice 2:  BS  B0 S0 *
(1  PD)  K  ( R  I HT ,S 0 )  0  0  0
Choice 3: BS  B1S1 *
(1  PD)  K  ( R  I HT ,S1 )  PD  B  K  (1   )  B  K      
Choice 4: BS  B1S0
(1  PD)  K  ( R  I HT ,S 0 )  PD  B  K  (1   )  B  K 1 0
Endowment:   00 (does not have smooth economic earnings and cannot smooth reported income)
Choice 1: BS  B0 S1
N/A (cannot have S1 for type 0 )
Choice 2:  BS  B0 S0 *
(1  PD)  K  ( R  I HT ,S 0 )  0  0  0
Choice 3: BS  B1S1
N/A (cannot have S1 for type 0 )
Choice 4: BS  B1S0
(1  PD)  K  ( R  I HT ,S 0 )  PD  B  K  (1   )  B  K 1 0
This article is protected by copyright. All rights reserved.
Table A1, continued
General structure of manager's payoff: (1  PD)  K  ( R  I )  ( PD  B  K )  (  B  K  PC )  L  
Panel B: Payoffs to manager in low threat of private benefits consumption
 LT   ; LLT   ; PCS1   ; PCS 0  1
Endowment:   11 (has smooth economic earnings and can smooth reported income)
Choice 1: BS  B0 S1 *
(1  PD1 )  K  ( R  I LT ,S1 )  0  0  0  
Choice 2:  BS  B0 S0
(1  PD 0 )  K  ( R  I LT ,S 0 )  0  0  
Choice 3: BS  B1S1
(1  PD1 )  K  ( R  I LT ,S1 )  PD1  B  K   B  K    0  
Choice 4: BS  B1S0
(1  PD 0 )  K  ( R  I LT ,S 0 )  PD1  B  K   B  K 1  
Accepted Article
Endowment:   10 (has smooth economic earnings but cannot smooth reported income)
Choice 1: BS  B0 S1
N/A (cannot have S1 for type 0 )
Choice 2:  BS  B0 S0 *
(1  PD 0 )  K  ( R  I LT ,S 0 )  0  0  0
Choice 3: BS  B1S1
N/A (cannot have S1 for type 0 )
Choice 4: BS  B1S0
(1  PD 0 )  K  ( R  I LT ,S 0 )  PD1  B  K   B  K 1  0
Endowment:   01 (does not have smooth economic earnings but can smooth reported income)
Choice 1: BS  B0 S1
(1  PD1 )  K  ( R  I LT ,S1 )  0  0    
Choice 2:  BS  B0 S0 *
(1  PD 0 )  K  ( R  I LT ,S 0 )  0  0  0
Choice 3: BS  B1S1
(1  PD1 )  K  ( R  I LT ,S1 )  PD1  B  K   B  K      
Choice 4: BS  B1S0
(1  PD 0 )  K  ( R  I LT ,S 0 )  PD 0  B  K   B  K 1  0
Endowment:   00 (does not have smooth economic earnings and cannot smooth reported income)
Choice 1: BS  B0 S1
N/A (cannot have S1 for type 0 )
Choice 2:  BS  B0 S0 *
(1  PD 0 )  K  ( R  I LT ,S 0 )  0  0  0
Choice 3: BS  B1S1
N/A (cannot have S1 for type 0 )
Choice 4: BS  B1S0
(1  PD 0 )  K  ( R  I LT ,S 0 )  PD 0  B  K   B  K 1  0
This article is protected by copyright. All rights reserved.
Accepted Article
REFERENCES
Acharya, V., Lambrecht, B., 2015. A theory of income smoothing when insiders know more than
outsiders. Review of Financial Studies 28, 2534-2574.
Aghion, P., Bolton, P., 1992. An incomplete contracts approach to financial contracting. Review
of Economic Studies 59, 473-494.
Amiram, D., Kalay, A., Sadka, G., 2017. Industry characteristics, risk premiums, and debt
pricing. The Accounting Review 92, 1-27.
Ashbaugh-Skaife, H., Collins, D., LaFond, R., 2006. The effects of corporate governance on
firms’ credit ratings. Journal of Accounting and Economics 42, 203-243.
Bae, K., Goyal, V., 2009. Creditor rights, enforcement, and bank loans. Journal of Finance 64,
823-860.
Barnea, A., Ronen, J., Sadan, S., 1975. The implementation of accounting objectives: an
application to extraordinary items. The Accounting Review 50, 58–68.
Barton, J., 2001. Does the use of financial derivatives affect earnings management decisions?
The Accounting Review 76, 1-26.
Bebchuk, L., Cohen, A., Ferrell, A., 2009. What matters in corporate governance? Review of
Financial Studies 22, 783-827.
Beidleman, C., 1973. Income smoothing: the role of management. The Accounting Review 48,
653-667.
Berger, A., Udell, G., 1990. Collateral, loan quality, and bank risk. Journal of Monetary
Economics 25, 21-42.
Bharath, S., Sunder, J., Sunder, S., 2008. Accounting quality and debt contracting. The
Accounting Review 83, 1-28.
Bhattacharya, U., Daouk, H., Welker, M., 2003. The world price of earnings opacity. The
Accounting Review 78, 641-678.
Black, E., Sellers, K., Manly, T., 1998. Earnings management using asset sales: An international
study of countries allowing noncurrent asset revaluation. Journal of Business Finance &
Accounting, 25, 1287-1317.
Costello, A., Wittenberg-Moerman, R., 2011. The impact of financial reporting quality on debt
contracting: Evidence from internal control weakness reports. Journal of Accounting
Research 49, 97-136.
Daske, H., Hail, L., Leuz, C., Verdi, R., 2008. Mandatory IFRS reporting around the world: early
evidence on the economic consequences. Journal of Accounting Research 46, 1085-1142.
Dechow, P., Dichev, I., 2002. The quality of accruals and earnings: The role of accrual
estimation errors. The Accounting Review 77, 35-59.
Dechow, P, Ge, W., Schrand, C., 2010. Understanding earnings quality: a review of the proxies,
their determinants and their consequences. Journal of Accounting and Economics 50,
344-401.
DeFond, M., Park, C., 1997. Smoothing income in anticipation of future earnings. Journal of
Accounting and Economics 23, 115–139.
De Franco, G., Hope, O., Lu, H., 2017. Managerial ability and bank-loan pricing. Journal of
Business Finance & Accounting DOI: 10.1111/jbfa.12267.
Demerjian, P., Donovan, J., Lewis-Western, M., 2017. Income smoothing and debt covenants:
evidence from technical default. Working paper.
Demiroglu, C., James, C., 2010. The information content of bank loan covenants. Review of
Financial Studies 23 (10), 3700-3737.
This article is protected by copyright. All rights reserved.
Accepted Article
Demski, J., 1998. Performance measure manipulation. Contemporary Accounting Research 15,
261–285.
Demsetz, H., Lehn, K., 1985. The structure of corporate ownership: causes and consequences.
Journal of Political Economy 93, 1155-1177.
Desai, H., Hogan, C., Wilkins, M. 2006. The reputational penalty for aggressive accounting:
Earnings restatement and management turnover. The Accounting Review 81, 83-112.
Dichev, I., Tang, V., 2009. Earnings volatility and earnings predictability. Journal of Accounting
and Economics 47, 160-181.
Djankov, S., McLiesh, C., Shleifer, A., 2007. Private credit in 129 countries. Journal of Financial
Economics 84, 299-329.
Djankov, S., La Porta, R., Lopez-de-Silanes, F., Shleifer, A., 2008. The law and economics of
self-dealing. Journal of Financial Economics 88, 430-465.
Dou, Y., Hope, O., Thomas, W., 2013. Relationship-specificity, contract enforceability, and
income smoothing. The Accounting Review 88, 1629-1656.
Drucker, S., Puri, M., 2009. On loan sales, loan contracting, and lending relationships. Review of
Financial Studies 22, 2835-2872.
Duan, J., Sun, J., Wang, T., 2012. Multiperiod corporate default prediction - a forward intensity
approach. Journal of Econometrics 170, 191-209.
Francis, J., LaFond, R., Olsson, P., Schipper, K., 2004. Costs of equity and earnings attributes.
The Accounting Review 79, 967-1010.
Friedman, H., 2017. Capital market development and confidence in disclosure quality. Working
paper.
Fudenberg, D., Tirole, J., 1995. A theory of income and dividend smoothing based on
incumbency rents. Journal of Political Economy 103, 75–93.
Gassen, J., Fulbier, R., 2015. Do creditors prefer smooth earnings? Evidence from European
private firms. Journal of International Accounting Research 14, 151-180.
Gopalan, R., Jayaraman, S., 2012. Private control benefits and earnings management: Evidence
from insider controlled firms. Journal of Accounting Research 50, 117-157.
Graham, J., Smith, C., 1999. Tax incentives to hedge. Journal of Finance 54, 2241-2262.
Graham, J., Harvey, C., Rajgopal, S., 2005. The economic implications of corporate financial
reporting. Journal of Accounting and Economics 40, 3–73.
Gu, Z., Zhao, Y., 2006. Accruals, income smoothing and bond ratings. Working paper.
Hamm, S., Jung, B., Lee, W., 2017. Labor unions and income smoothing. Contemporary
Accounting Research, forthcoming.
Hepworth, S., 1953. Smoothing periodic income. The Accounting Review 28, 32-39.
Hunt, A., Moyer, S., Shevlin, T., 2000. Earnings volatility, earnings management and equity
value. Working Paper.
Iliev, P., Lins, K., Miller, D., Roth, L., 2015. Shareholder voting and corporate governance
around the world. Review of Financial Studies 28, 2167-2202.
Ivashina, V., 2009. Asymmetric information effects on loan spreads. Journal of Financial
Economics 92, 300-319.
John, K., Litov, L., Yeung, B., 2008. Corporate governance and risk-taking. Journal of Finance
63: 1679-1728.
Jung, B., Soderstrom, N., Yang, S., 2013. Earnings smoothing activities of firms to manage
credit ratings. Contemporary Accounting Research 30, 645-676.
This article is protected by copyright. All rights reserved.
Accepted Article
Kim, J., Tsui, J., Cheong, H., 2011. The voluntary adoption of International Financial Reporting
Standards and loan contracting around the world. Review of Accounting Studies 16, 779811.
Lang, M., Maffett, M., 2011a. Economic effects of transparency in international equity markets:
a review and suggestions for future research. Foundation and Trends in Accounting 5,
175-241.
Lang, M., Maffett, M., 2011b. Transparency and liquidity uncertainty in crisis periods. Journal of
Accounting and Economics 52, 101-125.
Lang, M., Lins, K., Maffett, M., 2012. Transparency, liquidity, and valuation: International
evidence on when transparency matters most. Journal of Accounting Research 50, 729774.
Lang, M., Stice-Lawrence, L., 2015. Textual analysis and international financial reporting: large
sample evidence. Journal of Accounting and Economics 60, 110-135.
La Porta, R., Lopez-de-Silanes, F., Shleifer, A., Vishny, R. 1998. Law and finance. Journal of
Political Economy 106, 1113-1155.
La Porta, R., Lopez-de-Silanes, F., Shleifer, A., Vishny, R., 2000. Investor protection and
corporate governance. Journal of Financial Economics 58, 3-27.
Leuz, C., Nanda, D., Wysocki, P., 2003. Earnings management and investor protection: an
international comparison. Journal of Financial Economics 69, 505–527.
Li, S., Richie, N., 2016. Income smoothing and the cost of debt. China Journal of Accounting
Research 9, 175-190.
Lin, C., Ma, Y., Malatesta, P., Xuan, Y., 2011. Ownership structure and the cost of corporate
borrowing. Journal of Financial Economics 100, 1-23.
Merton, R., 1974. On the pricing of corporate debt: The risk structure of interest rates. Journal of
Finance 29, 449-470.
Murfin, J., 2012. The supply-side determinants of loan contract strictness. Journal of Finance 67,
1565-1601.
Peasnell, K., 1998. Discussion of earnings management using asset sales: An international study
of countries allowing noncurrent asset revaluation. Journal of Business Finance &
Accounting 25, 1319-1324.
Qian, J., Strahan, P., 2007. How laws and institutions shape financial contracts: The case of bank
loans. Journal of Finance 62, 2803-2834.
Rajan, R., 1992. Insiders and outsiders: the choice between informed and arm's‐length debt. The
Journal of Finance 47, 1367-1400.
Rego, S, Wilson, R., 2012. Equity risk incentives and corporate tax aggressiveness. Journal of
Accounting Research 50, 775-810.
Ronen, J., Sadan, S., 1981. Smoothing income numbers: Objectives, means and implications.
Addison-Wesley, Reading, MA.
Roychowdhury, S., Watts, R., 2007. Asymmetric timeliness of earnings, market-to-book and
conservatism in financial reporting. Journal of Accounting and Economics 44, 2-31.
Shuto, A. and Iwasaki, T., 2014. Stable shareholdings, the decision horizon problem and
earnings smoothing. Journal of Business Finance & Accounting, 41, 1212-1242.
Sufi, A., 2007. Information asymmetry and financing arrangements: evidence from syndicated
loans. Journal of Finance 62, 629-668.
Takasu, Y., 2013. Does income smoothing affect the cost of bank loans? Working paper.
Thomson Reuters., 2014. Global Syndicated Loans Review. Accessed October 19, 2016 at
http://dmi.thomsonreuters.com/Content/Files/4Q2014_Global_Syndicated_Loans_Review.pdf.
This article is protected by copyright. All rights reserved.
Tirole, J., 2001. Corporate governance. Econometrica 69, 1-35.
Trueman, B., Titman, S., 1988. An explanation for accounting income smoothing. Journal of
Accounting Research 26, 127–139.
Tucker, J., Zarowin, P., 2006. Does income smoothing improve earnings informativeness? The
Accounting Review 81, 251-270.
Zhang, J., 2008. The contracting benefits of accounting conservatism to lenders and borrowers.
Journal of Accounting and Economics 45, 27-54.
Table 1 – Non-U.S. sample composition
Accepted Article
Table 1 presents the country distribution of the sample firms and facility-level observations used in our primary
analyses. Anti-self-dealing index (Djankov et al., 2008) takes values in the range zero to one, where higher values
indicate a lower threat of private benefit consumption. A country is classified as having high (low) threat of private
benefits consumption if its anti-self-dealing index is below (above) the sample observation median. PBThreat is an
indicator variable that =1 (=0) for high (low) threat countries. For simplicity, we follow prior research (e.g., Daske,
Hail, Leuz, & Verdi, 2008) and refer to Hong Kong as a country.
NATION
Australia
Brazil
Canada
France
Germany
Hong Kong
India
Italy
Korea (South)
Mexico
Netherlands
Singapore
South Africa
Spain
Sweden
Taiwan
United Kingdom
Firms
N
%
14
2.2
2
0.3
69
10.8
72
11.3
28
4.4
25
3.9
18
2.8
8
1.3
23
3.6
5
0.8
8
1.3
8
1.3
2
0.3
12
1.9
4
0.6
143
22.4
198
31.0
Total
639
100.0
Facilities
N
%
29
1.6
2
0.1
160
8.8
281
15.5
90
5.0
44
2.4
33
1.8
12
0.7
71
3.9
7
0.4
15
0.8
9
0.5
6
0.3
19
1.1
4
0.2
377
20.8
658
36.2
1,817
Anti-Self
Dealing Index
0.76
0.27
0.64
0.38
0.28
0.96
0.58
0.42
0.47
0.17
0.20
1.00
0.81
0.37
0.33
0.56
0.95
100.0
This article is protected by copyright. All rights reserved.
PBThreat
0
1
0
1
1
0
0
1
1
1
1
0
0
1
1
0
0
Table 2 - Sample descriptive statistics
Table 2 presents descriptive statistics for the facility-level observations used in our analyses. Variable definitions are
presented in Appendix A.
Accepted Article
Variable
N
Mean
Std
P1
P25
Median
P75
P99
DiscSmooth
1,817
49.323
26.839
2.000
26.500
50.500
71.500
97.000
FundSmooth
1,817
51.796
27.598
2.500
29.000
51.500
74.500
97.000
Spread
1,817
134.413
108.854
19.000
55.000
100.000
200.000
525.000
Facility
1,817
462.163
987.631
0.183
23.164
140.000
440.767
4708.474
LFacility
1,817
4.301
2.563
-1.699
3.143
4.942
6.089
8.457
Maturity
1,817
64.216
21.929
24.000
60.000
60.000
75.000
144.000
LMaturity
1,817
4.105
0.346
3.178
4.094
4.094
4.317
4.970
Secure
1,817
0.254
0.436
0.000
0.000
0.000
1.000
1.000
NCov
1,817
0.569
0.929
0.000
0.000
0.000
1.000
4.000
LSize
1,817
14.006
1.477
11.015
12.967
13.923
14.982
17.689
BTM
1,817
0.714
0.480
0.009
0.380
0.612
0.942
2.117
Leverage
1,817
0.591
0.149
0.265
0.491
0.588
0.680
0.980
ROA
1,817
0.080
0.068
-0.109
0.044
0.076
0.114
0.235
NetWorth
1,817
11.396
19.601
0.037
0.666
2.761
12.073
86.318
Tangible
1,817
0.389
0.245
0.015
0.178
0.362
0.565
0.935
StdRet
1,817
0.106
0.052
0.031
0.070
0.097
0.129
0.292
Table 3 - Correlation matrix for facility-level variables
Table 3 reports Pearson (Spearman) correlations for our sample above (below) the diagonal.
Variable
(1)
DiscSmo
oth (1)
FundSmo
oth (2)
0.07
7
Spread
(3)
0.04
9
LFacility
(4)
LMaturit
y (5)
0.05
1
0.01
3
Secure
(6)
0.01
6
NCov (7)
0.00
(2)
0.07
7
(3)
0.02
8
0.11
5
0.13
5
0.01
4
0.11
5
0.00
7
0.35
3
0.09
5
-
0.33
3
-
(4)
0.04
1
0.01
2
(5)
0.01
9
(6)
(7)
0.01
5
0.00
1
0.02
2
0.09
6
0.31
6
0.34
0
0.03
9
0.16
1
0.01
8
0.01
8
0.25
0
0.11
4
0.06
1
0.02
1
0.15
8
-
0.20
5
0.22
6
-
0.19
4
0.15
(8)
0.11
0
0.00
6
0.22
5
0.56
3
0.11
1
0.14
9
-
(9)
(10)
0.08
1
0.08
6
0.06
7
0.07
5
0.36
9
0.11
9
0.09
6
0.28
2
0.06
2
0.07
8
0.01
0
0.02
5
0.24
5
0.16
5
0.23
0
0.03
5
0.12
7
0.09
8
0.08
0
-
0.07
5
-
0.08
7
-
0.01
6
0.10
This article is protected by copyright. All rights reserved.
(11)
0.05
1
(12)
0.00
2
(13)
0.04
1
0.09
3
0.04
3
0.02
9
0.04
8
0.00
8
-
(14)
0.03
0
0.20
8
0.10
8
0.21
3
0.09
1
0.03
2
0.09
4
LSize (8)
0.11
5
BTM (9)
0.04
1
Leverage
(10)
0.09
1
ROA (11)
Accepted Article
NetWorth
(12)
Tangible
(13)
StdRet
(14)
0.10
1
0.04
7
0.05
8
0.03
9
0.02
1
0.06
1
0.34
0
0.05
1
0.36
3
0.35
8
0.15
7
0.12
6
0.13
3
0.04
4
0.25
4
0.07
4
0.27
3
0.04
1
0.17
9
0.13
4
0.11
6
0.03
6
0.46
0
0.12
2
0.07
6
0.15
8
0.05
2
0.06
6
0.26
4
0.06
7
0.10
9
0.00
3
0.04
5
0.08
5
0.15
4
0.58
9
3
0.03
9
0.13
4
0.07
6
0.00
8
0.21
4
0.07
4
0.04
3
0.08
1
0.00
1
0.04
9
0.15
9
0.15
1
0.34
9
1
0.03
6
0.04
1
0.25
8
0.08
5
0.42
5
0.07
2
0.24
8
0.23
7
0.27
7
0.40
1
0.04
3
0.05
7
0.02
6
0.17
1
0.14
0
0.19
2
0.13
4
0.12
0
0.12
0
0.30
3
0.23
1
0.20
0
0.13
7
0.17
4
0.02
8
0.14
1
0.00
3
0.31
7
0.01
3
0.13
4
0.06
9
0.14
2
0.12
2
0.13
3
0.19
0
0.04
1
0.00
5
0.05
3
0.16
1
0.13
4
0.05
8
0.15
0
0.10
0
0.28
2
2
0.22
2
0.16
1
0.03
7
0.07
6
Table 4 – Country-level threat of private benefits consumption
Table 4 presents results of OLS estimation of Eq. (4):
Spread i ,l   0  1 PBThreati   2 DiscSmoothi , t   3 DiscSmooth * PBThreat   4 FundSmoothi , t
  5 FundSmooth * PBThreat   X i,t +  Yi,l  country  year   i , l .
Spread is the loan interest rate over LIBOR in basis points. PBThreat is an indicator that equals one (zero) if a firm
is in a country with high (low) threat of private benefits consumption. DiscSmooth is a rank variable increasing in
firm-level discretionary smoothing. FundSmooth is a rank variable increasing in firm-level fundamental smoothness.
All variables are further defined in Appendix A. Country, year, and industry fixed effects (including the intercept
and main effect of PBThreat) are included where indicated but not reported. Robust t-statistics based on two-way
clustered standard errors are reported in parentheses, where C, M, and P indicate clustering by country, calendar
month-year, and loan package, respectively. *, **, and *** indicate significance (two-sided) at the 10%, 5% and 1%
levels, respectively.
Column:
(1)
(2)
(3)
(4)
DiscSmooth
-0.038
-0.145**
-0.145**
-0.131*
(-0.39)
(-2.14)
(-2.38)
(-1.93)
DiscSmooth *PBThreat
0.410***
0.410***
0.433**
(2.64)
(2.86)
(2.46)
FundSmooth
-0.184
-0.166
-0.166
-0.368**
(-1.16)
(-1.38)
(-1.37)
(-2.38)
FundSmooth *PBThreat
-0.035
-0.035
-0.101
(-0.11)
(-0.11)
(-0.36)
LSize
-14.238***
-14.086***
-14.086***
-12.240***
(-5.91)
(-5.65)
(-6.11)
(-5.84)
BTM
22.867***
22.941***
22.941***
20.040***
(4.37)
(3.95)
(3.74)
(3.13)
Leverage
46.003***
43.602***
43.602***
35.640***
(4.30)
(3.79)
(4.07)
(2.70)
ROA
-3.530
-0.809
-0.809
-1.346
This article is protected by copyright. All rights reserved.
(-0.07)
1.156
(0.10)
213.822**
(1.98)
4.030
(1.16)
-10.831***
(-3.09)
70.433***
(3.36)
64.170***
(2.98)
Tangible
StdRet
NCov
LFacility
LMaturity
Accepted Article
Secure
DiscSmooth + DiscSmooth*PBThreat
Included Fixed Effects
Standard Error Clustering
N
Adj. R2
C, Y
C, M
1,817
0.437
(-0.02)
2.970
(0.27)
225.143**
(2.07)
4.253
(1.23)
-10.725***
(-3.03)
70.278***
(3.36)
63.676***
(3.00)
0.265*
C, Y
C, M
1,817
0.438
(-0.02)
2.970
(0.26)
225.143**
(1.98)
4.253
(1.30)
-10.725***
(-3.02)
70.278***
(3.34)
63.676***
(2.99)
0.265*
C, Y
C, P
1,817
0.438
(-0.03)
-5.908
(-0.58)
213.077*
(1.94)
4.271
(1.23)
-11.197***
(-3.28)
68.224***
(3.14)
63.475***
(2.97)
0.302*
C, Y, I
C, P
1,817
0.442
Table 5 - Firm-level threat of private benefits consumption
Table 5 presents results of OLS estimation of Eq. (4) using firm-level threat measures:
Spread i ,l   0  1 PBThreatFirmi   2 DiscSmoothi ,t   3 DiscSmooth * PBThreatFirm   4 FundSmoothi ,t
  5 FundSmooth * PBThreatFirm   X i,t +  Yi,l  country  year   i ,l .
Spread is the loan interest rate over LIBOR in basis points. CloseHeldShares is an indicator that equals one (zero) if
a firm has a high (low) percentage of closely-held shares. MgrEntrench is an indicator that equals one (zero) if a
firm has a high (low) Bebchuk et al. (2009) entrenchment index. DiscSmooth is a rank variable increasing in firmlevel discretionary smoothing. FundSmooth is a rank variable increasing in firm-level fundamental smoothness. All
variables are further defined in Appendix A. Column (1) uses our primary sample with two-way clustered standard
errors by country and loan package. Column (2) uses a sample of U.S. firms with standard errors clustered by loan
package. Where indicated, country and year fixed effects are included but not reported, including the intercept and
the main effect of PBThreatFirm. Robust t-statistics are reported in parentheses. *, **, and *** indicate significance
(two-sided) at the 10%, 5% and 1% levels, respectively.
PBThreatFirm Var.:
CloseHeldShares
MgrEntrench
Column:
(1)
(2)
DiscSmooth
-0.182*
-0.101*
(-1.65)
(-1.73)
DiscSmooth *PBThreatFirm
0.653***
0.389**
(3.18)
(2.56)
FundSmooth
-0.056
-0.396***
(-0.43)
(-6.49)
FundSmooth *PBThreatFirm
-0.399
-0.089
(-1.37)
(-0.62)
LSize
-15.419***
-8.503***
(-5.94)
(-5.15)
BTM
23.391***
9.693***
(3.35)
(3.32)
Leverage
43.972*
113.466***
(1.66)
(13.39)
ROA
10.825
-141.145***
(0.16)
(-6.97)
This article is protected by copyright. All rights reserved.
Tangible
StdRet
NCov
LFacility
LMaturity
Secure
Accepted Article
DiscSmooth + DiscSmooth*PBThreatFirm
Included Fixed Effects
Standard Error Clustering
N
Adj. R2
2.977
(0.24)
220.143***
(2.78)
8.142**
(2.02)
-10.892***
(-5.13)
71.423***
(6.57)
69.988***
(7.16)
0.471**
C, Y
C, P
1,608
0.454
6.310
(0.97)
180.274***
(6.24)
-1.119
(-1.08)
-13.063***
(-7.99)
23.156***
(4.53)
78.072***
(22.66)
0.288**
Y
C, P
6,033
0.465
Table 6 - Borrower credit risk
Table 6 presents results of OLS estimation of Eq. (5):
Spread i ,l   0  1 PBThreati   2 DiscSmoothi ,t   3 DiscSmooth * PBThreat   4 DiscSmooth * PBThreat * PD
  5 FundSmoothi ,t   6 PDi ,t   7 DiscSmooth * PD   8 FundSmooth * PD   9 PBThreat * PD
 X i,t + Yi,l  country  year   i ,l .
Spread is the loan interest rate over LIBOR in basis points. PBThreat is an indicator that equals one (zero) if a firm
is in a country with high (low) threat of private benefits consumption. DiscSmooth is a rank variable increasing in
firm-level discretionary smoothing. FundSmooth is a rank variable increasing in firm-level fundamental smoothness.
PD is firm-level probability of default. All variables are further defined in Appendix A. Country and year fixed
effects (including the intercept and main effect of PBThreat) are included but not reported. Robust t-statistics based
on two-way clustered standard errors by country and loan package are reported in parentheses. *, **, and ***
indicate significance (two-sided) at the 10%, 5% and 1% levels.
Column:
(1)
DiscSmooth
-0.188***
BTM
19.321*
(-2.76)
(1.90)
DiscSmooth*PBThreat
0.188
Leverage
32.707*
(0.51)
(1.71)
DiscSmooth* PBThreat*PD
0.793**
ROA
32.072
(2.14)
(0.55)
FundSmooth
0.120
Tangible
13.163
(0.55)
(0.87)
PD
79.111***
StdRet
140.198
(3.17)
(1.09)
DiscSmooth*PD
0.318
NCov
5.647**
(1.47)
(2.25)
FundSmooth*PD
-1.542***
LFacility
-9.257**
(-3.66)
(-2.07)
PBThreat*PD
13.327
LMaturity
62.597**
(0.56)
(2.20)
LSize
-14.894***
Secure
52.883**
(-6.42)
(2.43)
Included Fixed Effects
C, Y
This article is protected by copyright. All rights reserved.
Standard Error Clustering
N
Adj. R2
C, P
1,309
0.414
Table 7 - Financial Covenants
Accepted Article
Table 7 presents results of OLS estimation of a modified version of Eq. (4) where NCov (the number of financial
covenants on the loan) is used as the dependent variable:
NCovi ,l   0  1 PBThreati   2 DiscSmoothi , t   3 DiscSmooth * PBThreat   4 FundSmoothi , t
  5 FundSmooth * PBThreat   X i,t +  Yi,l  country  year   i ,l .
Spread is the loan interest rate over LIBOR in basis points. PBThreat is an indicator that equals one (zero) if a firm
is in a country with high (low) threat of private benefits consumption. DiscSmooth is a rank variable increasing in
firm-level discretionary smoothing. FundSmooth is a rank variable increasing in firm-level fundamental smoothness.
All variables are further defined in Appendix A. Country and year fixed effects (including the intercept and main
effect of PBThreat) are included but not reported. Robust t-statistics based on two-way clustered standard errors by
country and loan package are reported in parentheses. *, **, and *** indicate significance (two-sided) at the 10%,
5% and 1% levels.
Column:
(1)
DiscSmooth
0.001
(0.44)
DiscSmooth *PBThreat
0.007**
(2.15)
FundSmooth
0.001
(1.13)
LSize
0.045
(1.19)
BTM
0.028
(0.44)
Leverage
0.116
(0.64)
ROA
-0.289
(-0.72)
Tangible
-0.459***
(-2.97)
StdRet
1.494
(0.99)
Spread
-0.000
(-0.59)
LFacility
-0.001
(-0.09)
LMaturity
-0.046
(-0.74)
Secure
0.001
(0.44)
Included Fixed Effects
C, Y
Standard Error Clustering
C, P
N
724
Adj. R2
0.358
This article is protected by copyright. All rights reserved.
Table 8 - Creditor rights
Accepted Article
Table 8 presents results of OLS estimation of an extension of Eq. (4) which includes a variable that captures the
extent of country-level creditor rights:
Spread i ,l   0  1 PBThreati   2WeakCredRightsi   3 DiscSmoothi ,t   4 DiscSmooth * WeakCredRightsi
  5 DiscSmooth * PBThreat   6 FundSmoothi ,t   X i,t +  Yi,l  country  year   i ,l .
PBThreat is an indicator that equals one (zero) if a firm is in a country with high (low) threat of private benefits
consumption. WeakCredRights is an indicator that equals one if a firm is in a country with weak creditor rights.
DiscSmooth (FundSmooth) is a rank variable increasing in firm-level discretionary (fundamental) smoothness. All
variables are further defined in Appendix A. Country and year fixed effects (including main effects on PBThreat and
WeakCredRights) are included, but not reported. Robust t-statistics based on two-way clustered standard errors by
country and loan package are reported in parentheses. *, **, and *** indicate significance (two-sided) at the 10%,
5% and 1% levels, respectively.
Column:
(1)
(2)
DiscSmooth
-0.052
-0.120**
(-0.50)
(-2.01)
DiscSmooth*WeakCredRights
0.055
-0.184
(0.28)
(-1.25)
DiscSmooth*PBThreat
0.501***
(2.71)
FundSmooth
-0.183
-0.177
(-1.12)
(-1.13)
LSize
-14.210***
-14.177***
(-6.66)
(-6.59)
BTM
22.916***
22.828***
(4.02)
(3.69)
Leverage
46.071***
42.844***
(4.52)
(3.94)
ROA
-2.635
-2.557
(-0.05)
(-0.05)
Tangible
1.440
2.282
(0.12)
(0.20)
StdRet
214.971*
223.200*
(1.90)
(1.94)
NCov
4.035
4.294
(1.22)
(1.31)
LFacility
-10.837***
-10.678***
(-3.07)
(-3.05)
LMaturity
70.366***
70.619***
(3.35)
(3.40)
Secure
64.090***
63.790***
(2.97)
(3.00)
DiscSmooth+DiscSmooth*PBThreat
0.381*
Included Fixed Effects
C, Y
C, Y
Standard Error Clustering
C, P
C, P
N
1,817
1,817
Adj. R2
0.437
0.439
This article is protected by copyright. All rights reserved.
Table 9 – Other accounting attributes
Accepted Article
Table 9 presents results of OLS estimation of an extension of Eq. (4) which includes variables that capture other
accounting attributes, along with their interactions with PBThreat, where PBThreat is an indicator that equals one
(zero) if a firm is in a country with high (low) threat of private benefits consumption:
Spread i ,l   0  1 PBThreati   2 DiscSmoothi ,t   3 DiscSmooth * PBThreat   4 Attributei ,t
  5 Attribute * PBThreat   6 FundSmoothi , t   X i,t +  Yi,l  country  year   i ,l .
DiscSmooth (FundSmooth) is a rank variable increasing in firm-level discretionary (fundamental) smoothness. AQ is
a proxy for accruals quality. BTM is an inverse proxy for accounting conservatism. σNIEXS is an inverse proxy for
earnings persistence. All variables are further defined in Appendix A. Country and year fixed effects (including the
intercept and main effect of PBThreat) are included, but not reported. Robust t-statistics based on two-way clustered
standard errors by country and loan package are reported in parentheses. *, **, and *** indicate significance (twosided) at the 10%, 5% and 1% levels, respectively. # indicates significance (one-sided) at the 10% level.
Attribute:
AQ
BTM
σNIEXS
Column:
(1)
(2)
(3)
DiscSmooth
-0.392***
-0.151***
-0.098#
(-5.90)
(-2.69)
(-1.57)
DiscSmooth*PBThreat
0.682***
0.428***
0.497**
(3.43)
(2.62)
(2.34)
Attribute
-0.319***
26.690***
89.564***
(-4.13)
(4.10)
(2.94)
Attribute*PBThreat
0.350
-10.401
240.724
(1.04)
(-0.65)
(0.66)
FundSmooth
-0.270
-0.177
-0.067
(-1.33)
(-1.12)
(-0.62)
LSize
-6.141*
-14.158***
-13.958***
(-1.66)
(-6.62)
(-6.53)
BTM
19.865***
24.727***
(2.71)
(4.14)
Leverage
65.949***
46.167***
47.346***
(2.79)
(4.99)
(4.64)
ROA
-62.748
3.118
2.928
(-0.76)
(0.06)
(0.06)
Tangible
12.062
1.817
6.513
(1.28)
(0.15)
(0.59)
StdRet
261.633*
225.563**
224.549**
(1.85)
(2.00)
(2.01)
NCov
10.374***
4.245
4.365
(2.88)
(1.29)
(1.32)
LFacility
-10.638***
-10.652***
-10.679***
(-3.04)
(-3.02)
(-3.15)
LMaturity
23.554
70.479***
69.971***
(1.61)
(3.36)
(3.36)
Secure
58.520**
63.518***
63.490***
(2.14)
(2.97)
(3.04)
DiscSmooth+DiscSmooth*PBThreat
0.290#
0.277*
0.399*
Included Fixed Effects
C, Y
C, Y
C, Y
Standard Error Clustering
C, P
C, P
C, P
N
976
1,817
1,817
Adj. R2
0.439
0.439
0.441
This article is protected by copyright. All rights reserved.