Professional Documents
Culture Documents
*
University of Graz.
Accepted by Haresh Sapra. We thank an anonymous reviewer, Hans Christensen, Pingyang Gao, Ian
Gow, David Grünberger, Trevor Harris, Bjorn Jorgensen, Sebastian Kronenberger, Peter Pope, Joshua
Ronen, Stefan Schantl, Ulf Schiller, Cathy Schrand, participants at the GEABA Conference in
Hamburg, NYU Accounting Summer Camp, EIASM Workshop on Accounting and Regulation, IWP
Wissenschaftsforum, and seminar participants at Columbia University, University of Würzburg,
University of Cluj, London School of Economics, University Carlos III Madrid, University of Chicago,
University of North Carolina at Chapel Hill, London Business School, WHU – Otto Beisheim School of
Management, University of Chile, and University of Melbourne for helpful comments.
November 2018
Abstract
A widely held assumption in policy making and empirical research is that increasing the strength of
public enforcement improves financial reporting quality and audit quality. This paper provides a
more nuanced view. In a model with a manager who can manage earnings, a strategic auditor, and
an enforcement institution, we show that enforcement and auditing are complements in a weak
enforcement regime but can be substitutes in a strong regime. Although stronger enforcement
always mitigates earnings management, the effects of different instruments of strengthening
enforcement are ambiguous. We show that they can improve or impair financial reporting quality
and audit quality, depending on production risk, accounting system characteristics, and the scope of
auditing relative to enforcement.
Keywords: earnings management; enforcement; auditing; financial reporting quality; audit quality.
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:
10.1111/1475-679X.12251.
A widely held assumption in policy making and empirical research is that greater enforcement is
desirable because it improves financial-reporting and audit quality, and several empirical studies
provide supporting evidence.1 Under this view, primarily the direct costs of enforcement limit its
desirability. Recent papers by Christensen, Liu, and Maffett [2017] and Florou, Morricone, and Pope
[2018] find support for high costs of enforcement to firms, exploiting a change of the enforcement
by the Financial Reporting Review Panel in the United Kingdom. Christensen, Liu, and Maffett [2017]
find that stronger enforcement reduces shareholder wealth. Florou, Morricone, and Pope [2018]
document that stronger enforcement increases audit effort, risk, and fees, particularly for firms
listed on the Alternative Investment Market.
The view that greater enforcement improves financial-reporting and audit quality implicitly is correct
if one assumes that it does not alter the auditor’s strategy. Once one considers direct and indirect
effects, the results can be different. Our paper analyzes the effects of stronger enforcement on
earnings management, financial-reporting and audit quality, and audit fees in an equilibrium setting.
We develop a model in which the owner of a firm hires a manager and a strategic auditor. The
compensation contract designed to motivate the manager to provide productive effort also creates
incentives for upward earnings management. The auditor strategically chooses an effort that
identifies and corrects errors in the preliminary financial report, based on a conjecture of earnings
management and enforcement outcomes. After publication of the audited financial report, the
enforcer investigates. It identifies remaining errors with some probability (the enforcement
intensity) and imposes penalties on the owner, manager, and auditor. Stronger enforcement is
achieved by increasing enforcement intensity or any of the penalties inflicted to the parties.
Our main finding is that auditing and enforcement are complements in a low-intensity enforcement
regime but can become substitutes in a strong regime. The auditor’s incentives to perform a high-
1
See Hope [2003], Ernstberger, Stich, and Vogler [2012], Christensen, Hail, and Leuz [2013], Brown, Preiato,
and Tarca [2014], and Silvers [2016].
We use the equilibrium strategies to examine the effects of strengthening enforcement on financial
reporting quality and on audit quality in equilibrium. Our results suggest that the effects depend on
the proxy for financial reporting quality. If it measures intentional errors only, as does the modified
Jones model (Dechow, Sloan, and Sweeney [1995]), then stronger enforcement always increases
financial reporting quality, because it mitigates earnings management. If it measures both
unintentional and intentional errors, as does the Dechow and Dichev [2002] model, then the impact
of stronger enforcement on financial reporting quality depends on the accounting system and the
production environment. The reason is that the accounting system is not perfect and produces
errors that lead to an understatement or overstatement of economic earnings. Through upward
earnings management, the manager makes more overstatements but fewer understatements. If the
corrective function of earnings management on average prevails, deterring it through stronger
enforcement reduces financial reporting quality. Another reason why financial reporting quality can
decline occurs if auditing is broader in scope than enforcement. If enforcement crowds out audit
effort, then the auditor produces less useful information, which decreases financial reporting
quality.
Audit quality can increase or decrease with stronger enforcement. An increase in enforcement
intensity or in the auditor’s penalty strictly increases audit quality in a weak enforcement regime but
can decrease it otherwise. There are two effects: a positive direct effect of the higher intensity or
penalty and a negative indirect effect from mitigating earnings management. Which one prevails
depends on the strength of the enforcement. An increase in the manager’s penalty strictly decreases
audit quality, because it alleviates audit effort. We show that audit fees and the number of
enforcement actions strictly increase with a higher enforcement intensity in a weak enforcement
regime, but that either can decrease otherwise. The number of enforcement actions strictly
decreases with higher penalties for the auditor, manager, or both. Yet the functional form of these
two measures does not fully correspond with that of financial-reporting and audit quality,
respectively.
The main empirical implications of our analyses are as follows. We show that different enforcement
instruments have different, sometimes ambiguous, implications on auditor’s behavior, financial
This paper contributes to recognize and better understand the interactions of earnings
management, auditing, and enforcement and on the resulting accounting outcomes, financial
reporting quality and audit quality. We are not aware of other analytical papers that explicitly study
the economic effects of enforcement and the interaction of enforcement and auditing. However,
this paper relates to several streams of literature that examine earnings management and auditing.
The paper’s productive setting relates to studies of production effort and earnings management in
multi-action agency models. For example, Feltham and Xie [1994] model productive effort and
earnings management (“window dressing”), which are simultaneously induced by the same
information system, and provide insights into the properties of an optimal information system in a
LEN setting. Glover and Levine [2018] assume the manager learns the actual outcome before
managing earnings and can only correct a wrong (low) report, thus allowing that earnings
management can be beneficial. In our paper, the manager can also bias a correct report, but
earnings management can still yield benefits on average. Laux and Laux [2009] study management
compensation by the board of directors, which also decides on oversight effort, and show that these
two decisions are related. Bertomeu, Darrough, and Xue [2017] consider production and earnings
management choices and focus on the optimal bias (conservatism) of the underlying accounting
system. Chan [2016] studies the effect of increasing internal controls on earnings management and
auditing. Laux and Stocken [2018] consider a similar setting but focus on the interaction between
accounting standards and enforcement. Enforcement in their model discovers noncompliance with
some probability and imposes a penalty that increases with stronger enforcement. None of these
papers consider auditing and enforcement jointly.
Other models study earnings management in rational expectations equilibria, in which a manager
biases financial reports to increase the market price of the firm (e.g., Fischer and Verrecchia [2000],
Ewert and Wagenhofer [2011] survey this literature). In these models, auditing and enforcement
implicitly drive the cost of earnings management. Königsgruber [2012] addresses enforcement in
such a model, with the added feature that the manager decides on the investment in a risky project.
He finds that stronger enforcement strictly increases reporting quality but may reduce investment
efficiency, due to over-deterrence of investment in viable projects.3 In contrast, we find that
2
See the composite index constructed by Brown, Preiato, and Tarca [2014].
3
Deng, Melumad, and Shibano [2012] find a related result for increasing auditor liability.
The auditing literature analyzes audit strategies but does not explicitly consider enforcement. Most
of the papers model a strategic auditor, who maximizes expected utility by the choice of audit effort
(e.g., Antle [1982], Baiman, Evans, and Noel [1987]), as we do. Given that contingent audit fees are
not allowed in most jurisdictions, the motivation for auditors to exert effort in these models usually
results from the risk that the auditor may be held liable if an error is uncovered. Our enforcement
mechanism is explicitly based on its interaction with the audit results. Other papers assume that the
liability arises from shareholder litigation. In that case, the cost to the auditor depends on decisions
by shareholders and on the liability regime (e.g., Ewert [1999], Hillegeist [1999]). In a setting in
which the manager is interested in a high market price of the firm, Patterson, Smith, and Tiras [2016]
study how earnings management and audit effort change if the liability of management, of the
auditor, or of both increase. Other auditing literature considers internal controls implemented
before auditing happens (e.g., Smith, Tiras, and Vichitlekarn [2000], Pae and Yoo [2001]). Ye and
Simunic [2017] study the effect of audit oversight and its interaction with the liability regime on
audit effort and audit market structure.
The paper proceeds as follows. Section 2 describes the model and introduce the underlying
production technology, the accounting system, the discretion for earnings management, auditing,
and public enforcement. Section 3 establishes the equilibrium in the earnings management and
auditing game. Section 4 derives effects of varying the strength of enforcement on equilibrium
earnings management and audit effort. Section 5 examines how strengthening enforcement affects
financial reporting quality, number of enforcement actions, audit quality, and audit fees. Section 6
solves for the optimal compensation contract. Section 7 considers extensions to the model, and
Section 8 concludes with empirical implications.
2. Model
We develop a one-period agency model with the owners of a firm, a manager, an auditor, an audit
oversight institution, and an enforcement institution (the enforcer). In the following, we describe
these elements, their relations, and the production and information technology. The notation is
summarized in the appendix.
We consider a risk neutral owner representing the owners of the firm or the board of directors and
abstract from potential conflicts of interest among different owners or board members. The owner
The manager has a reservation utility of zero and, because of limited liability, the compensation paid
must be positive. We assume that x is not contractible or not observable to the owner and to the
manager by the time the manager receives the compensation. Compensation s() ≥ 0 is written on
the audited financial report r {rL, rH}. It is easy to see that a contract s(ri) that induces aH must
satisfy s(rL) < s(rH); that is, the owner pays more compensation if the report is good, because rH is
more likely to occur if the manager chose aH. Moreover, because the manager’s reservation utility is
0 and si ≥ 0, there is no need to overpay the manager, so that s(rL ) = 0 and s ≡ s(rH) > 0, which
reflects a simple bonus contract.
After observing y, the manager can manage earnings and misrepresent the signal to achieve a
financial report m ≠ y. m is the preliminary financial report, which is subject to auditing (see below).
Earnings management includes the choice of probabilities bL ≡ b(yL) and bH ≡ b(yH) with which it is
successful in reporting mi ≠ yi, i = L, H. The cost of earnings management effort is 0 at bi = 0 and
increasing and convex in bi. It captures disutility from, for example, searching for earnings
management opportunities, future disadvantages, loss of reputation, or unethical behavior but not
enforcement penalties that we consider explicitly. We assume a quadratic cost function, vbi2 / 2 ,
where v is a constant scaling factor that is sufficiently high that bi < 1. We show below (Lemma 1)
that this contract induces earnings management only if yL occurs; that is, bH = 0 and bL ≥ 0. Figure 1
summarizes the information flow and indicates out-of-equilibrium strategies by dotted arrows.
2.3. Auditing
The firm’s financial report is subject to mandatory auditing. The owner hires an auditor at the
beginning of the game. Before contracting, the auditor learns the precision of the accounting system
{, } and uses it to assess the audit risk and plan the audit procedures. After engagement but
before determining audit effort, the auditor receives the manager’s preliminary report m that is
subject to the audit.
The auditor chooses the audit effort, which equals the probability gi, i = L, H, that he observes the
accounting signal y. Providing gi is privately costly to the auditor with a cost of 1
2 kgi2 , where k > 0 is
a parameter that scales the quadratic cost and is sufficiently large to prevent perfect auditing (gi =
1). The audited financial report is as follows.
yi with probability gi ,
ri (1)
mi with probability (1 gi ).
The audit market consists of auditors with similar technologies and is competitive. Consistent with
typical audit regulations, we assume the auditor receives a noncontingent audit fee A > 0. In a
After accepting the engagement, the auditor’s objective is to minimize the expected cost of the audit
and of a penalty incurred from any remaining uncorrected errors that are uncovered by
enforcement.
2.4. Enforcement
A public enforcement institution independently investigates audited financial reports. Examples are
the SEC’s Division of Corporation Finance in the United States and national enforcement agencies in
European Union countries. A typical feature is that enforcement investigates financial reports of a
sample of firms after they were audited and published. The sample can be strictly random or chosen
based on some patterns ex ante.
We model public enforcement as a technology. An investigation uncovers the signal yj with some
probability f ∈ *0, 1+, which is uncorrelated with the auditor’s effort gi. We refer to f as the
enforcement intensity that is determined by the budget the enforcer has available for investigations.
For example, allocating a higher budget to the enforcer increases f. Without loss of generality, we
cast our analysis directly in terms of f.
If the enforcer detects yi, the investigation ends without a finding if the report ri equals yi (i = L, H);
otherwise the enforcer initiates an enforcement action, which imposes costs on all parties involved.
These costs include fines, penalties, potential legal liability as well as indirect effects, such as a
reputation loss, and they can be monetary or nonmonetary.
The costs of an enforcement action are penalties CO > 0 on the owner (the firm), C M > 0 on the
manager, and CA > 0 on the auditor. We assume these penalties are not excessive. If CO is large, the
owner will always prefer to induce the low effort aL, and the manager’s compensation can be set
independent of y, which provides no earnings management incentives and avoids enforcement
penalties. We assume fC
M
s to account for the manager’s limited liability and to avoid situations
in which the penalty is so high that the manager never overstates earnings (bL = 0). Since we vary f in
our subsequent analysis, we make the stronger sufficient condition CM ≤ s. A necessary condition for
gi < 1 is fCA/k < 1, and we assume C A / k 1 . Note that the enforcement intensity f and the
penalties CM and CA act as substitutes in determining enforcement strength.
4
The equilibrium we establish does not depend on the competition in the audit market. Once the fee is fixed,
it does not affect the equilibrium strategies.
Auditor chooses audit effort g, learns y with probability g, and corrects errors (m ≠ y)
Enforcer investigates audited report r, learns y with probability f, and initiates enforcement
action if r ≠ y
O M A
Owner, manager, and auditor suffer penalties C , C , and C respectively from enforcement
action
3. Equilibrium
In this section, we characterize the equilibrium earnings management and auditing strategy given an
incentive-compatible compensation contract. We endogenize the contract in Section 6. All formal
proofs are in the appendix.
The manager chooses bi after observing y = yi to maximize her expected utility, conjecturing the
auditor’s audit strategy gˆ L and gˆ H , where the hat denotes conjectures. The expected utilities are
v
E[U M yL ] prob(rH yL ) s prob(error yL )C M V bL2 ,
2 (2)
v
E[U M yH ] prob(rH yH ) s prob(error yH )C M V bH2 .
2
The terms are the conditional expected compensation, the expected enforcement cost, the disutility
of effort aH, V, and the disutility of earnings management. The probabilities depend on the
productive effort aH (which induces a success probability p > q), the accounting system errors and
, earnings management, and audit effort. They are stated in the proofs.
The lemma confirms that the manager overstates earnings if the initial signal is low, yL, but never
understates high earnings yH. Note that, because gˆ H < 1, even a perfect enforcement (f 1) cannot
fully deter upwards earnings management as long as CM < s.5
The probability of an enforcement action determines the expected cost of enforcement for each of
the parties involved. Observe that there is no risk of an enforcement action if the manager does not
manage earnings. Since the manager has only an incentive to overstate earnings, she manages
earnings bL if the accounting system reports yL. The auditor detects earnings management with
probability gH and corrects it. With probability (1 – gH), earnings management prevails and is
detected by the enforcer with probability f. Taken together, the ex ante probability that the enforcer
initiates an enforcement action and imposes enforcement costs is
The auditor’s strategy is determined as follows. After accepting the audit engagement, the auditor
chooses gi to maximize his expected utility, conditional on the preliminary report mi, i = L, H,
k
U A mi A gi2 prob(error mi ) C A . (5)
2
The expected utility comprises the endogenously determined audit fee A, the cost of audit effort,
and the expected penalty of an enforcement action. Exerting higher effort reduces the probability of
an enforcement action but incurs higher disutility.
Lemma 2: The optimal audit effort levels for mL and mH are: gL = 0 and
prob(yL )bˆL CA
gH f . (6)
prob(yL )bˆL +prob(yH ) k
prob(yL mH )
A perfect Bayesian equilibrium in this manager-auditor game requires that the following conditions
hold, where i, j = L, H:
(i) {bi*} maximizes the manager’s conditional expected utility given conjectures {gˆ j };
5
For example, a clawback provision (e.g., Section 304 of the Sarbanes-Oxley Act and rule 954 of the Dodd-
M
Frank Act) of bonuses based on erroneous financial reports implies C = s, so full deterrence, bL = 0, would
occur for f = 1.
Proposition 1: For any s that induces aH, there exists a unique equilibrium with earnings management
s CM CA
bL* = (1 g H* ) 1 f > 0 and bH
*
= 0 and audit effort g *
L = 0 and g *
H = prob( y L mH ) f
v s k
> 0.
The equilibrium strategies equal those stated in Lemma 1 and 2 (equations (3) and (6)), after
replacing conjectures by the respective other party’s equilibrium strategies. We show in the proof
that this equilibrium is unique. The equilibrium strategies are stated implicitly because they depend
on each other in nontrivial ways.
Enforcement strength increases in the enforcement intensity f with which the enforcer detects
errors and in the enforcement penalties CO, CM, and CA that it imposes on the owner, manager, and
auditor. A regulator can increase enforcement intensity by increasing the (benevolent) enforcer’s
budget. Penalties are typically bounded from above by considerations that they must be
proportional to the crimes. We show how a variation of each of these parameters individually affects
the equilibrium outcomes. Because we do not include costs of enforcement (by increasing f) and
other considerations (such as the appropriateness of penalties in a legal system), we do not study
the optimal design of an enforcement regime.
1 s k
f 0 M g H* ( f 0 ) A . (7)
2C C
Proposition 2 (ii) states that equilibrium audit effort g H* strictly increases for low f but can strictly
decrease for high f, depending on whether its maximum is attained within the feasible range of f.
Thus, the proposition establishes a complementary relation between audit effectiveness and
enforcement intensity if it is low and a substitutive relation, or crowding out, if it is high. To see this
result, note that higher f directly increases g H* because it raises the likelihood that the enforcer finds
earnings management. This again increases the expected penalty on the auditor. Then there is an
indirect effect: a higher f mitigates earnings management bL* , which reduces the likelihood that the
audit detects errors. This effect leads to a decrease in audit effort through a decrease in the
probability prob(yLmH) that the low accounting signal prevails, given that the manager’s report is
high. This probability is increasing and strictly concave in earnings management, which implies that
the effect of a lower bL* on prob(yLmH) is stronger the lower bL* becomes. As is apparent from the
expression (6) for the optimal audit effort, the marginal direct increase of g H* for higher f declines in
f, while the indirect negative effect grows. Thus, when earnings management is large, then
increasing f, starting from a low level, increases audit effort strongly and only slightly reduces the
probability prob(yLmH), which determines the auditor’s expected penalty; hence, the direct effect
dominates. When f is already high, then a further increase has a stronger effect on prob(yLmH),
inducing the auditor to exert less audit effort.
*
The enforcement intensity for which g H attains its unique maximum is
1 s k
f 0 M g H* ( f 0 ) A .
2C C
*
Equilibrium audit effort g H decreases in f for f > f0. Given our constraint that CM ≤ s, the value of f0
> 1/2. However, f0 can be greater than 1, so that there is no feasible interval of f in which audit effort
decreases in f. Note that enforcement intensity and enforcement penalties determine the strength
of enforcement. CM and CA drive the threshold f0 directly downward, although equilibrium audit
effort also depends on these penalties.
* *
Figure 3 plots the equilibrium earnings management bL and audit effort g H for the full range of
enforcement intensity f ∈ [0, 1]. The underlying parameter values are p = 0.3, = 0.05, = 0.05, s =
*
0.6, v = 10, CM = 0.5, and CA/k =1. Equilibrium earnings management bL always decreases in
*
enforcement f, whereas equilibrium audit effort g H increases in f up to f0 ≈ 0.62 and then
decreases.
The owner’s penalty CO has no immediate effect on the reporting equilibrium because it affects
neither the manager’s nor the auditor’s expected utility directly. Its only effect is that it raises the
owner’s cost of motivating high productive effort aH, which ultimately can become so high that the
owner foregoes implementing aH and prefers the low effort aL, which can be induced by a fixed wage
contract that avoids earnings management altogether.
Proposition 3 (iii) states the effect of an increase of the auditor’s penalty CA. A higher penalty CA (or a
*
lower k) directly increases the incentive for the auditor to exert higher audit effort g H . The
increasing audit effort unambiguously reduces equilibrium earnings management. The relevant
parameter for the equilibrium strategies is the ratio CA/k, which reflects the relative importance of
the enforcement penalty relative to the level of the audit effort cost.
Our first measure defines financial reporting quality as the ex ante probability that the report
accurately informs about the productive outcome x; that is, rH = xH and rL = xL. Given the manager
takes the desired effort aH, this is
6
Francis, Olsson, and Schipper [2006] similarly distinguish between innate and discretionary sources for
earnings quality.
Proposition 4: Financial reporting quality FRQ strictly increases in enforcement intensity f, in the
auditor’s penalty CA, and in the manager’s penalty CM if Q > 0; it strictly decreases in f, CA, and CM if Q
< 0, where
Q ≡ (1 p)(1 ) p . (9)
The proposition shows that FRQ either monotonically increases or decreases in f, in CA and CM,
depending on the sign of the term Q. To see this, we rewrite FRQ by inserting the probabilities,
which yields
FRQ depends on the base accounting quality that is determined by the - and -error and a term
that is affected by a change in equilibrium earnings management and audit effort, both of which are
influenced by enforcement strength. We show in the proof that the term bL (1 g H ) strictly
*
decreases in bL and thus strictly decreases in f, in CA, and in CM.8 The effect on FRQ is therefore
determined by the sign of Q.
prob( yL ) (1 p)(1 ) p ,
which consists of two parts: (1 – p)(1 – ) is the probability that x = xL and y = yL, which is a correct
depiction of the economic outcome, while p is the probability that x = xH but the accounting system
falsely reports y = yL. If the manager manages earnings upon observing yL, she reports mH with
*
probability bL > 0. If x = xL, then earnings management disguises the originally correct signal yL,
which adds to an overstatement of the financial report. We call this bad earnings management.
Conversely, if x = xH, then the accounting signal y = yL is incorrect, and upwards earnings
management corrects the understatement error. We call this good earnings management. If Q < 0,
that is,
7
The cost of an under- or overstatement can vary with the individual decision problem in which the financial
report is used. Applying different weights does not qualitatively affect our results.
8 O
A variation of the owner’s penalty C has no direct effect (see Proposition 3 (i)).
then the probability that earnings management corrects an understatement error is greater than the
probability that it aggravates an overstatement error. According to Propositions 2 and 3,
strengthening enforcement through either increasing f, CA or CM strictly mitigates bL . Therefore, if
earnings management is bad, then less of it improves financial reporting quality; if it is good, then
less of it reduces financial reporting quality because it diminishes the error-correcting function of
earnings management.
Two notes are in order. First, stronger enforcement still fulfils a primary function, which is
deterrence of earnings management. And second, even if earnings management is good, it is not the
manager’s intent to do something good, but managing earnings still follows from the incentive to
increase compensation.
Figure 4 plots the financial reporting quality FRQ on enforcement intensity f ∈ [0, 1] and three
different productive probabilities p. The underlying parameter values are = 0.2, = 0.2, s = 0.6, v =
10, CM = 0.5, and CA/k =1. p = 0.8 is the knife-edge case in which Q = 0, which implies that FRQ is
unaffected by a change in enforcement strength. A lower p yields a Q > 0, so that earnings
management is bad, in which case increasing enforcement intensity monotonically increases FRQ.
p = 0.7 illustrates this. If p > 0.8, in the Figure p = 0.85, Q < 0 and earnings management is good, so
that FRQ decreases in f.
The results follow immediately from the inspection of Q = (1 p)(1 ) p . First, firms that are
more likely to generate high output (high p) make erroneous understatements of accounting signals
more frequently, which are corrected by upward earnings management. Second, the less precise the
base accounting system is—that is, the larger the accounting errors and are—the lower is Q,
because there arise more instances that benefit from a correction.
While less precision tends to generate a negative effect of stronger enforcement on FRQ, accounting
bias has distinct effects depending on the situation. We parameterize conservative accounting by a
parameter ; higher implies higher conservatism. Let , , and denote with
Q the original term Q with and .
Corollary 2: (i) If p ½, greater conservatism strictly decreases FRQ if Q ≤ 0; otherwise the effect is
ambiguous.
(ii) If p < ½, greater conservatism strictly increases FRQ if Q 0; otherwise the effect is ambiguous.
To establish this result, we show in the proof that undetected earnings management bL (1 g H )
strictly decreases in conservatism . To see this, note that a greater increases prob(yL) and
decreases prob(yH) by the same amount, thus increasing prob(yL mH ) and consequently the audit
effort. Intuitively, under higher conservatism, the low accounting signal realizes more frequently and
raises the likelihood that a high preliminary report mH is the result of earnings management. The
auditor responds by increasing the audit effort. There is a countervailing effect because higher audit
effort reduces the manager’s optimal bias, but it is not sufficient to outweigh the positive impact of
.
The probability of understatement and overstatement errors depends on the probability p of the
high outcome xH. If p ½, more conservatism increases the sum of the accounting errors because
Our second measure of financial reporting quality uses the preliminary accounting report y as
benchmark, which is unaffected by earnings management, and it focuses on the effects of auditing
and enforcement on earnings management. Define
Proposition 5: Financial reporting quality FRQEM strictly increases in f, CA, and CM.
This result follows directly from the fact that bL (1 g H ) strictly decreases in enforcement intensity f
* *
and the penalties CA and CM, which we have shown in Proposition 4. Since this definition does not
consider the - and -errors that arise in the accounting system, earnings management cannot
correct such errors, and earnings management is always undesirable. Reducing earnings
management thus unambiguously improves financial reporting quality FRQEM.
Comparing the results for both measures of financial reporting quality reveals different effects of
strengthening enforcement. While the effect on FRQEM, which captures earnings management only,
is consistent with conventional intuition, the effect on the more comprehensive FRQ depends on
circumstances, which are the production technology and the accounting system errors. Thus it is
9
This resembles the result in Gigler, Kanodia, Sapra, and Venugopalan [2009] on conservatism in a debt
contracting setting with no earnings management. They show that debt contract efficiency (which is their
measure of accounting quality) declines with more conservatism because the increase in understatement error
dominates the decrease in overstatement error. This result hinges on the assumption of a relatively high
expected outcome, which is akin to p ½ in Corollary 2 (i).
10
Chen, Hemmer, and Zhang [2007] study an agency model with earnings management, but they do not
consider auditing and enforcement. They show that a small amount of conservatism enhances earnings
informativeness and contracting efficiency. While their model’s specifics differ from ours, the impetus is that
conservative accounting mitigates earnings management, which is similar to Corollary 2.
An enforcement action occurs if the signal yL realizes, the manager succeeds in overstating earnings
and reports mH, and the auditor does not find the overstatement but the enforcer discovers y = yL.
Proposition 6: (i) An increase of the enforcement intensity f strictly increases the likelihood of an
enforcement action if f < f1 and strictly decreases it if f > f1, where f1 s 2C f 0 .
M
(ii) An increase in the penalties C M or C A strictly decreases the likelihood of an enforcement action.
The proposition suggests that using the likelihood of an enforcement action is a reliable measure for
enforcement strength only under specific circumstances. As we show above, bL (1 g H ) declines in
* *
f; hence increasing the enforcement intensity f causes countervailing effects on prob(error). If the
enforcement level is low, increasing f strictly increases the likelihood of an enforcement action.
However, if enforcement intensity increases, it mitigates earnings management and the likelihood of
an enforcement action decreases. We show in the proof that, due to these countervailing effects,
there exists a level f1 < f0 for which the likelihood of an enforcement action reaches a maximum. In
case f1 exceeds 1, the maximum is outside of the relevant range for f [0,1] and prob(error) strictly
increases for all feasible f.
The proposition further states that an increase in either of the penalties CM and CA strictly decreases
the likelihood of an enforcement action. With greater penalty CM, the manager reduces earnings
management, which implies that the basic need for a correction of the published financial
statements declines. The auditor responds by reducing his audit effort, but as the proof shows, this
effect is eventually not large enough to outweigh the impact of reduced earnings management. A
greater penalty CA induces the auditor to extend his audit effort, thus finding mistakes more
frequently. The manager responds by reducing earnings management. Here, both responses work in
the same direction, and both reduce the likelihood of an enforcement action.
Considering the relation between the likelihood of an enforcement action and financial reporting
quality, we note an enforcement action itself is not free of error. An enforcement action arises in
two distinct events. First, the actual outcome is xL, which is correctly reported by the accounting
system as signal yL. The manager succeeds at overstating earnings as mH. The auditor does not
observe y. Hence the report is rH, but the enforcer finds out yL and initiates an enforcement action.
The probability of this event is
Second, the actual outcome is xH, which is incorrectly reported by the signal yL. The manager
succeeds at overstating earnings as mH. The auditor again does not observe y, but the enforcer finds
out yL and initiates an enforcement action. However, given x = xH, the enforcement action wrongly
corrects the (ex post) appropriate financial report. The probability of this event is
p bL* (1 g H* ) f .
Comparing the two probabilities, it is apparent that the error through enforcement is greater than
the corrective function if Q < 0, that is, when earnings management is good.
AQ prob(mL )g L* prob(mH )g H*
0 (12)
prob(yL )b prob(yH ) g .
L
*
H
Proposition 7: (i) An increase of the enforcement intensity strictly increases audit quality AQ if f < f2
and strictly decreases it if f > f2, where f 2 f 0 .
(ii) Audit quality AQ strictly decreases in the penalty CM.
(iii) Audit quality AQ strictly increases in the penalty CA for small CA. It always strictly increases in CA if
prob(yH )
bL 2 f 1 0 , (13)
prob(yL )
Audit quality AQ is a function of a weighted product of g H and bL , but it can be restated as a
function of bL alone, resulting in
CA
AQ prob(yL )b f .
L
k
A
The crucial term is bL fC , whose impact is ambiguous because bL strictly decreases in f and CA.
Proposition 7 (i) states that AQ is an inverse u-shaped function of enforcement intensity f. Intuitively,
this result follows from Proposition 2 that audit effort is inversely u-shaped. The difference is that
the maximum AQ is achieved at a lower f than the maximum of that of audit effort, that is, f2 < f0.
The only monotonic relation is stated in part (ii) of the proposition: AQ strictly decreases in the
manager’s penalty CM. An increase in CM mitigates earnings management, which induces a decrease
of the audit effort and lowers the probability of a high preliminary accounting report mH. Thus ex
ante, the audit effort strictly decreases in CM.
bL 2 f 1 bL 1 ,
implying that the condition (13) holds for a wide set of parameter values. For instance, if prob(yH)
prob(yL), then the condition is always fulfilled. Also, the less conservative the accounting system is,
the higher is prob(yH) and the lower is prob(yL), thus increasing the probability ratio. Intuitively,
audit quality is
AQ prob(mH )g H* .
Hence a less conservative accounting system increases the probability prob(mH) that the high
preliminary accounting signal obtains and the auditor exerts audit effort. This attaches a higher
weight to the positive impact of a greater CA on audit effort and a lower weight on the reduction of
the bias. When the probability ratio is large enough, the positive impact of CA on audit effort always
outweighs the negative impact on earnings management, and audit quality always strictly increases
in CA.
Empirical studies often use audit fees as a measure for (unobservable) audit quality. In our model,
the owner hires the auditor at the beginning of the game, and the audit fee is endogenously
determined in a competitive audit market with homogenous auditors. It is based on the anticipated
audit effort and the expected cost of an enforcement action. In equilibrium, the auditor’s conditional
utility given mH equals
k
U A mH A ( g H* ) 2 prob( yL mH )(1 g H* ) fC A
2
k
A g H* (2 g H* ).
2
If m = mL, the auditor exerts no effort, but because the manager has not managed earnings there is
no cost from enforcement in this case.
k
A prob mH g H* 2 g H*
2 (14)
k
prob yH prob yL b g H* 2 g H* .
L
2
Proposition 8: (i) An increase of the enforcement intensity strictly increases the audit fee A if f < f3
and strictly decreases it if f > f3, where f3 f 2 .
(ii) The audit fee A strictly decreases in the penalty CM.
(iii) The audit fee A strictly increases in the penalty CA for small CA. It always strictly increases in CA if
prob(yH )
bL 3 f 1 0
prob(yL )
The functional form of the audit fee is determined by the induced changes on the equilibrium bL
*
and g H . Earnings management bL decreases in f, whereas the effect of f on g H depends on the
level of enforcement before f increases further. Proposition 2 (ii) establishes an inverse u-shaped
effect and a threshold value f0 that g H decreases in f if f > f0. A decrease, dg H df 0 , is a
*
sufficient condition that the equilibrium audit fee declines. Then increasing f not only reduces
earnings management but also the cost of the audit. Conversely, if dg H df 0 , then the cost of
*
audit effort rises and, with it, the necessary fee, whereas the lower earnings management reduces
audit effort again. We show in the proof that these effects result in the existence of a level f3 at
which the audit fee reaches a maximum and that this level is smaller than the enforcement intensity
that maximizes audit quality AQ (which is, in turn, smaller than the level that maximizes audit effort).
Thus, stronger enforcement increases audit fees only if public enforcement is sufficiently weak.
A higher managerial penalty CM strictly decreases the audit fee. This follows because a greater CM
reduces earnings management, which again reduces the probability of obtaining the high
preliminary report mH and audit effort.
An increase of the auditor’s penalty CA has a similar effect on the audit fee as it has on audit quality
(Proposition 7 (iii)), and the same intuition applies. The proposition states a sufficient condition that
A always strictly increases in CA. The only difference is that the required lower bound for P is slightly
greater than the corresponding threshold for audit quality AQ.
Figure 5 depicts how enforcement intensity affects equilibrium audit quality AQ and audit fee A. The
underlying parameter values are p = 0.3, = 0.05, = 0.05, s = 0.6, v = 10, CM = 0.5, and CA/k =1.
Audit quality and audit fee are inversely u-shaped in f because its primary determinant is the audit
*
effort g H . The audit fee also explicitly captures expected penalties from enforcement.
We begin with recording the effects of the bonus s on earnings management and on audit effort in
equilibrium.
In our analysis so far, we have assumed the bonus s is predetermined, which reflects the fact that, at
the beginning of the game, the owner hires the manager and writes the compensation contract.
Later, there is a change in the strength of enforcement, but the contract is unchanged. We now
derive the optimal contract, under the assumption that the owner can write a new contract or adjust
an existing contract, with the manager after a change in enforcement strength. We assume that the
owner is a Stackelberg leader in the game, that is, she writes the compensation contract ex ante, and
it is observable to the auditor.
It consists of the expected outcome, the manager’s expected compensation, the (equilibrium) audit
fee, and the expected cost of enforcement in case the enforcer detects an error in the financial
report and penalizes the owner at CO.
The manager’s reservation utility is always satisfied because of the limited liability constraint, that is,
s ≥ 0, and the reservation utility is normalized to 0. Because we assume the owner wants to
implement aH, the bonus must be sufficiently high to induce the manager to choose the productive
effort aH over aL. The manager’s incentive compatibility constraint is
E[U M aH ] E[U M aL ] .
v
E[U M aH ] prob(rH )s V prob( yL ) bL*2 prob( yL )bL* (1 g H* ) fC M .
2
E[U M aL ] is the expected utility if the manager chooses the out-of-equilibrium effort aL and also
the out-of-equilibrium earnings management bLL bL ( g H aL ), 11 whereas the auditor still
* *
* *
conjectures (aH , bL ) and does not adjust the equilibrium audit strategy g H :
11 *
We prove in Proposition 9 that equilibrium earnings management is independent of a, i.e., bLL bL* .
V v
s bL2 . (16)
p q 1 2
Strengthening enforcement by increasing f, CM, or CA strictly decreases s*.
The optimal bonus compensates the manager for productive effort and earnings management. The
compensation for productive effort is just high enough to motivate the manager to induce her to
exert effort aH, and it is higher than the disutility V depending on (p – q), which is the difference in
productivity from the two actions and the precision of the accounting system (1 – – ). The bonus
includes full compensation for the induced earnings management.
Proposition 9 also shows that strengthening enforcement through either instrument, enforcement
intensity f or the penalties CM and CA, lowers the optimal bonus s*. Intuitively, the manager’s
expected utility depends on the information content of the financial report rH, which is affected by
* * *
the equilibrium strategies bL and g H . An increase in f or a penalty CM or CA decreases bL ; thus, the
owner can reduce the amount of the bonus and still motivate the manager to choose aH. The
owner’s penalty in case of an enforcement action, CO, does not affect the bonus. A higher CO
increases the owner’s expected cost of inducing high productive effort; hence, if CO grows too large,
the owner is better off paying flat compensation and inducing the manager to exert low effort aL,
which eliminates earnings management and avoids an enforcement action.
A lower optimal bonus also mitigates earnings management (Corollary 3), which has an indirect
effect on the equilibrium, because audit effort may go down. The next proposition assures that our
main results continue to hold when s* is endogenous.
Proposition 10: With an endogenous bonus s*, equilibrium earnings management bL and audit effort
g H behave as follows:
(i) bL strictly decreases in f, CM and CA.
(ii) An increase of the enforcement intensity f strictly increases the audit effort g H if f < f4
and strictly decreases it if f > f4.
(iii) g H strictly decreases in CM.
(iv) g H strictly increases in CA.
As we show in Section 5, financial reporting quality and audit quality ultimately depend on the
behavior of bL for strengthening enforcement, Proposition 10 implies that our earlier results also
carry over to an endogenous contracting setting.
There may be contracts other than our simple bonus contract. For example, the contract can be
augmented with a clawback provision. That is, if the enforcer finds and publishes an error, the owner
can require the manager to repay a bonus. The manager’s expected compensation becomes
prob(rH )s prob(error) s ,
and the expected clawback prob(error)s increases the expected utility of the owner. The
introduction of a clawback provision does not alter our main results.
A more complex contract is a truth-telling contract, in which the owner requests a report yR from the
manager about y and makes compensation contingent on yR and r. Under the conditions underlying
the revelation principle, a payoff-equivalent truth-inducing contract can then be implemented,
which prevents earnings management. The auditor chooses the audit effort sequentially rationally
12
This suggests f4 < f0, i.e., with endogenous bonus, crowding out of auditing occurs already for lower f than
with fixed bonus, but the relation is subtler because in the setting of Proposition 2 (where we establish f0), the
bonus is set arbitrarily subject only to s CM and the manager’s incentive compatibility constraint. However,
we show in the proof of Proposition 10 that for each f and a corresponding bonus s*(f) for which crowding
would occur holding s*(f) constant, there is also crowding out with an endogenous bonus.
7. Extensions
7.1. Auditor generates additional information
We assume that auditing and public enforcement have basically the same audit or investigation
technologies available. That is, the auditor uncovers the raw earnings (signal y) with probability g,
which is determined by costly audit effort, and the enforcer uncovers the signal yj with probability f.
The probabilities g and f are uncorrelated. If g was nonstrategic, we would always have increasing
earnings-management deterrence by increasing either g or f. With a strategic choice of g by the
auditor, we show that earnings management still decreases but g and f are strategic complements
for low f and strategic substitutes for high f.
In reality, the scope and depth of an audit is greater than that of an enforcement investigation.13 An
audit comprises tests of controls and substantive procedures, including analytical procedures and
tests of details, for example, providing audit evidence of physical inventory, bank balances, loan
quality, and the like, to identify material misstatements. Enforcement does usually not perform tests
of internal controls or completeness of the bookkeeping system and inventory but instead focuses
on critical accounting positions, many of which are subject to high judgment. Thus, the scope of
enforcement is much more limited; in other words, the enforcer does not perform another audit.
Recall that our reporting system contains two kinds of errors: one is an intentional error introduced
by earnings management, but another comes from the imprecision of the accounting system to
capture the productive outcome x perfectly, symbolized by the -error and the -error. A key
determinant of - and -errors is the effectiveness of the firm’s internal controls. The effectiveness
of internal controls is subject to an assessment by the auditor, who has to report on control
weaknesses.
13
For the variety of audit activities, see Dye [1993, 1995], Schwartz [1997], Chan and Pae [1998], Pae and Yoo
[2001], Chan and Wong [2002], Laux and Newman [2010], and Ye and Simunic [2013, 2017].
To add structure, assume that, since x is always more informative than y, the auditor corrects the
financial report to include x rather than y whenever they are observed. The audited financial report
is as follows:
xi with probability gi
ri
mi with probability (1 gi ).
If the enforcer obtains yi, the investigation ends without a finding if the report ri equals yi (i = L, H);
otherwise the enforcer alleges an error. However, if the auditor discovered x during his audit, he can
present evidence that ri = xi, and the enforcer accepts this evidence and ends the investigation. If the
enforcer finds out yi ≠ rj and there is no evidence about x available, the enforcer states an error and
imposes penalties to the parties.
The inclusion of superior auditing does not alter the main results of the analysis. In particular,
*
Proposition 2 still holds, which states that increasing enforcement intensity f, bL strictly decreases
* *
and the behavior of g H is inversely u-shaped. That is, enforcement f and g H are complements for
low f and substitutes for high f.
The main difference lies in the effect of increasing enforcement intensity on financial reporting
quality FRQ. The next result describes the effects, where we hold the bonus s constant to be able to
compare it with Proposition 4.
The corollary follows immediately from Proposition 4 and the fact that the information auditing can
*
provide depends on the level of audit effort g H . The probability of an error in the financial reports,
after auditing and enforcement, is the same as stated in (10), but the auditor’s corrective function by
substituting x for r is now valuable and improves FRQ in its own right. As long as a higher
*
enforcement intensity increases g H as well, it reinforces the improvement of FRQ that we describe
k
U A mi A gi2 prob(error mi ) C A ,
2
where the audit fee A is fixed already and avoiding CA is traded off against higher audit effort g. Since
audit fees are noncontingent in most jurisdictions, auditors face several other disciplining
mechanisms that substitute for the lack of continent fees, specifically audit oversight and legal
liability.
Audit oversight institutions, for example, the U.S. Public Company Accounting Oversight Board
(PCAOB), monitor auditors and can penalize them if they do not follow the auditing standards, even
if financial reports turn out to be correct later. We can include audit oversight parsimoniously by
assuming that it ensures that the auditor exerts a minimum audit effort level 0 < g < 1, regardless of
the audit risk of the client and strategic incentives.
Our main results are structurally unaffected by the introduction of g as long as the equilibrium audit
* *
effort g H > g . Otherwise the audit effort g H = g and is not chosen strategically any more. Hence
auditing with g makes it a technology just like our enforcement.
Legal liability arises through litigation by investors who relied on the audited report. Litigation is
typically triggered if an enforcement action was initiated or if it becomes apparent that the audited
report was wrong. If investors sue the manager and the auditor after an enforcement action, the
effect of legal liability is mainly an increase in CM and CA. If it is based on later indicators, litigation is
asymmetric in that investors sue only if the firm reported rH and the actual outcome turned out to
be xL (ignoring issues of hindsight).14 Litigation is another strategic decision that can affect our
results. The interrelation between enforcement and litigation is an interesting topic for further
research.
14
See Chan and Pae [1998] and Ewert [1999].
We show that financial reporting quality does not always improve with greater enforcement
strength. It mainly depends on the proxy for financial reporting quality. One reason why earnings
management can be good is that, by managing earnings upwards, it corrects understatement errors
that arise in the accounting processes. Alleviating earnings management can then have a negative
effect. Another reason is that enforcement is limited in scope, compared to auditing, and therefore
crowding out auditing diminishes financial reporting quality. Summarizing these results, we establish
that strengthening enforcement (before systemic enforcement costs) is desirable but “too much”
enforcement can become detrimental.
We also examine several empirical measures that relate to enforcement effectiveness and derive the
following predictions that can guide empirical tests.
Financial reporting quality (earnings quality): Our results suggest different predictions
dependent on the proxies used for earnings quality.
(i) If the proxy predominantly captures earnings management,15 stronger enforcement
increases financial reporting quality, because it unambiguously mitigates earnings
management.
(ii) If the proxy is comprehensive and captures both unintentional and intentional errors,16 the
predictions are ambiguous. Stronger enforcement tends to increase financial reporting quality
in firms or industries that have a low probability of success, in firms with more effective
internal controls, in firms with less conservative accounting, and vice versa.
15
For example, this is the case with proxies based on the modified Jones model (Dechow, Sloan, and Sweeney
[1995]).
16
An example is the measure of Dechow and Dichev [2002].
Predictions like these can help design empirical tests to gain better insights into the interaction
between auditing and enforcement and their effects on financial reporting quality and audit quality.
They also suggest that the economic effects of strengthening enforcement will differ by industry and
by country.
Our model is intentionally parsimonious to examine and highlight interactions between earnings
management and audit effort in an equilibrium setting. It also makes it easy to define constructs
such as financial reporting quality and audit quality as well as several proxies related to these. There
exist several avenues for relaxing assumptions and extending the model structure to capture other
aspects of financial reporting. Our accounting processes are simple in that we consider random
errors and earnings management. Including effects from complying with accounting standards, real
earnings management, and the like can offer further insights. We model public enforcement as a
fixed technology, but it can select firms for investigation based on other information. Moreover,
enforcers may be guided by individual incentives. In reality, there exist additional institutions whose
purpose is to improve financial reporting quality. Examples are internal controls and governance,
audit committees, audit oversight, whistleblower or media reports, and private litigation. They are
likely to interact with auditing and public enforcement. Finally, while we are interested in financial-
reporting and audit quality, we do not consider the task of a securities regulator who oversees public
enforcement. The regulator may be more interested in social welfare effects that result from the
design of the enforcement system. Social welfare depends on real effects of accounting and the
costs of enforcement broadly.
References
ANTLE, R. “The Auditor as an Economic Agent.” Journal of Accounting Research 20 (1982): 503–527.
BAIMAN, S., J. H. EVANS, AND J. NOEL “Optimal Contracts with a Utility-Maximizing Auditor.” Journal of
Accounting Research 25 (1987): 217–244.
BALL, R., S. P. KOTHARI, AND A. ROBIN “The Effect of International Institutional Factors on Properties of
Accounting Earnings.” Journal of Accounting and Economics 29 (2000): 1–51.
1
E[U M yH ] prob(rH yH ) s vbH2 ,
2
prob(rH yH ) gˆ L 1 < 0
bH
because gˆ L < 1. A necessary condition for bH > 0 is that this derivative is positive (enforcement costs
M
can only reduce E[U yH ] further), which cannot be true. Therefore, bH = 0.
Consider bL next. The manager maximizes her expected utility with respect to bL,
v
E[U M aH , yL ] prob(rH yL ) s bL (1 gˆ H ) fC M V bL2
2
v
bL (1 gˆ H ) s fC M V bL2
2
C
M
v 2
sbL (1 gˆ H ) 1 f V bL .
s 2
CM
E[U aH , yL ] s(1 gˆ H ) 1 f
M
vbL 0 ,
bL s
s CM
which implies bL (1 gˆ H ) 1 f 0.
v s
The inequality holds because fCM < s and gˆ L < 1. To ensure that bL < 1 we assume s/v is sufficiently
small.
Proof of Lemma 2
If the auditor observes mL, given conjecture bˆH 0 , he correctly anticipates that the enforcer will
never find an error because mL = yL. Therefore, prob(error mL ) 0 , and the auditor faces no cost of
enforcement. Consequently, the auditor conducts the minimum audit effort gL = 0.
prob(yL )bˆL
where prob(yL mH ) , which is greater 0 if bˆL 0 . The auditor’s
prob(yL )bˆL +prob(yH )
conditional expected utility is
k 2
U A mH A g H prob( yL mH )(1 g H ) fC A .
2
The first derivative with respect to gH equals
U A mH kg H prob( yL mH ) fC A ,
g H
CA
g H prob( yL mH ) f 0,
k
Proof of Proposition 1
We prove existence and uniqueness of an equilibrium in the feasible range for bL and gH using a
fixed-point argument. From Lemma 1,
s CM
bL (1 gˆ H ) 1 f .
v s
s CM
1 f for gˆ H 0
bL v s
0 for gˆ H 1.
CA
According to Lemma 2, g H prob( yL mH ) f with boundaries
k
g H prob yL prob yH fC A
0
bˆ prob y bˆ +prob y
2
L
k
L L H
prob yL prob yH
2
2 gH fC A
and 2 0.
bˆ 2 prob y bˆ +prob y
3
L
k
L L H
The two reaction functions bL ( gˆ H ) and g H (bˆL ) are monotonic and continuous; hence, the
function g H bL ( gˆ H ) is also continuous and maps gˆ H [0,1] into g H [0,1] due to
bL ( gˆ H ) [0,1] . Therefore, Brouwer’s fixed point theorem implies that a fixed point of
g H bL ( g H ) exists for g H [0,1] . This fixed point constitutes an equilibrium, proving existence.
Uniqueness follows from the negative slope of the linear function bL and increasing and convex
function gH.
* *
Next, we derive the explicit solutions for the equilibrium strategies bL and g H . We start with the
manager’s optimal bias,
s
bL (1 gˆ H )(1 f CM
s ) .
v
s
Defining T f CM
s 1 0 , rewrite bL as bL T ( gˆ H 1) . Solving for gˆ H yields
v
v
bL T
gˆ H s .
T
1 fC A
gH .
1 prob(yL ) k
1+
prob(yL )bˆL
Equating gH = gˆ H yields
T1 T12 4T0T2
The solution of the equation T2bL T1bL T0 0 is bL
2
. The terms Ti depend on
2T2
exogenous parameters, where T2 > 0 and T0 < 0 (due to T < 0) and the sign of T1 is indeterminate.
However, T0 < 0, implies T12 4T0T2 T1 . Therefore, regardless of the sign of T1, bL must be the
positive root because bL < 0 is not feasible.
prob(yL )bˆL fC A
gH
prob(yL )bˆL +prob(yH ) k
fC A
or prob(yL )bˆL g H g H prob(yH ) 0.
k
s fC A
prob yL T ( gˆ H 1) g H g H prob yH 0
v k
prob yH v fC A CA
Tg H2 T 1 g H Tf 0.
prob y L s k k
T3
T4
2T
solution in real numbers requires T4 4TT3 0 , i.e., T4 4TT3 . This must hold as there exists a
2 2
* *
unique equilibrium (bL , g H ) .
To determine the feasible root for gH, we rewrite the function that implicitly defines gH as
Tg H2 T4 g H T3 0 and obtain g H2 (T4 / T ) g H f CA
k 0 . According to Vieta’s rule, the product
, i.e., g H g H f
A A
CA
of the two roots equals f C
k
C
k . This implies that one solution is larger than f k ,
which is the solution g H with the negative root due to T4 > 0 and T < 0 (for the negative root, the
numerator of gH is more negative leading to a larger value for gH by dividing through T < 0). However,
Finally, consider the special case T = 0, where both numerator and denominator of g H are zero and
the quotient is not properly defined. If T = 0 then T3 = 0, T4 > 0, and inserting this into
Tg H2 T4 g H T3 0 yields g H T3 T4 0 .
Finally, a mixed-strategy equilibrium does not exist because the auditor’s expected utility is strictly
concave for any bL.
Consider first a variation of the equilibrium bL. As shown in the proof of Proposition 1, it is implicitly
prob(yH ) fC A v prob(yH )
defined by B T2bL T1bL T0 0 , where T0 T
2
, T1 T 1 ,
prob(yL ) k s prob(yL )
v
T2 , and T f CM
s 1 . The total differential with respect to parameters j = s, f, C M and CA/k is
s
1
B B bL b B B
0 L ,
j bL j j j bL
B T T 2 4T T
where 2T2bL T1 2T2 1 1 0 2
T1 T12 4T0T2 0 . Thus,
bL 2T2
b B
sign L sign for each j.
j j
Variation of f:
B CA fC A C M prob yH C M
bL T 1
f k k s prob yL s
CA CM fC A prob yH C M
bL T 1 0.
k s k prob yL s
0
b
T 0 implies B 0 and L 0 .
f f
B f fC A f prob( yH ) bL
bL 1 0 , implying 0.
C M
s k s prob( yL ) C M
Variation of CA and k:
B bL
fTbL , implying 0 since T 0 .
(C / k )
A
(C A / k )
Variation of s:
B v v prob(yH ) C fC
M A
C M prob(yH )
2 bL2 bL 2 f 1 f 0,
s s s prob(yL ) s2 k s 2 prob(yL )
0
bL
implying 0.
s
Consider next a variation of the equilibrium gH. As shown in the proof of Proposition 1, gH is implicitly
CA prob yH v fC A
defined by G Tg H T4 g H T3 0 , where T3 Tf
2
, T4 T 1 ,
k prob yL s k
and T ( f CM
s 1) . The total differential with respect to parameters j = s, f, and CA/k is
1
G G g H g G G
0 H ,
j g H j j j g H
G T T 2 4TT
where 2Tg H T4 2T 4 4 3
T4 T42 4TT3 0 . Therefore,
g H 2T
g G
sign H sign for each j.
j j
Variation of f:
G g
If f 1
2 (s C M ) , then M < 0, which implies 0 and H 0 . If f 1 2 (s C M ) , the term
f f
2 f C M
s 1
k
CA
0 , and the sign of M is indeterminate. There always exists a range of the
0
parameter f such that M > 0. To see this, ignore for the moment the restriction f ≤ 1, consider the
case f s C M 1 which implies bL 0 and therefore g H 0 . It then follows that
G g
M C A k 0 implying 0 and H 0 . Due to continuity, M must be positive in a range of
f f
f s C M and there exists f0 s C M such that M ( f0 ) 0 , i.e., g H attains a maximum at f 0 .
More specifically, observe that such f 0 requires
CA CM CM 1 s k
M ( f0 ) 2
0 f 1 g
H 0 f 0 f0 M g H f0 A .
k s s 2C C
Define
1 s k
F ( f ) f M gH ( f ) A .
2C C
For f = 0, the audit effort g H ( f 0) 0 and F ( f 0) 1 2 s C M 0 . For f s C M , we
have F f s C M 1
2 s C 0 due to g f s C
M
H
M
0 . Since F is continuous in f, the
intermediate value theorem implies that there exists f 0 such that F f 0 0 . Moreover, f 0 is
unique because
1 2
2 gH 2G 2G g H G G G 2G 2G g H
2
f 2 f f g H f g H f g H g H f g H2 f
f f0
0 at f 0 0 at f 0 0 at f 0
1
2G G
2 0.
f g H
extremum, f0 must be the unique maximum; otherwise, there would exist a minimum in the relevant
range of f, which cannot be the case.
This proves the existence of f 0 such that g H achieves its unique maximum at
1
2 s C f
M
0 s C M . If f 0 1 , then the maximum is outside the feasible range for f and
G
> 0 for any f [0,1].
f
Variation of CM:
G 2 f f fC A f fC A
gH gH 1
C M s s k s k
f f fC A
g H g H 1 1 g H
s s k
f fC A
g H 1 g H 0.
s k
0
0
g H
This implies 0.
C M
G g H
Variation of CA and k: g H fT fT fT 1 g H 0, implying 0.
(C / k )
A
(C A / k )
Variation of s:
G CM v prob yH C M fC A C M fC A
g H2 f 2 g H 2 f 2 1 f 2
s s s prob yL s k s k
CM v prob yH C M fC A
g H g H 1 f 2 g H 2 g H 1 f 2
s s prob yL s k
C M fC A v prob yH
g H 1 f 2
gH gH 2 0.
s k s prob yL
0
0
g H
Therefore, 0.
s
FRQ bL (1 g H )
p (1 p)(1 )
f f
bL (1 g H )
Q.
f
Note that
prob(yL )bL CA
bL 1 g H bL 1
f
prob(yH ) prob(yL )bL k
b
prob(yH ) prob(yL )bL 1 f CA
k
L
prob(yH ) prob(yL )bL
prob(yH ) prob(yL )bL 1 f CA
k .
prob(yH )
prob(yL )
bL
FRQ
bL strictly decreases in f, and so does bL 1 g H . Therefore, > 0 if Q > 0 and vice versa.
f
FRQ FRQ
According to Proposition 3, bL strictly decreases in CA and in CM. Therefore, A > 0 and
C C M
> 0 if Q > 0 and vice versa.
Proof of Corollary 2
As shown above in the proof of Proposition 2 and 3, equilibrium audit effort results from solving
G Tg H2 T4 g H T3 0 , and it follows that
prob yH p 1 1 p
P .
prob yL p 1 p 1
prob yH v fC A
T4 T 1
prob yL s k
G v P
depends on P. Taking the derivative of G with respect to yields gH 0 , implying
s
0
g H
0.
s CM
Since equilibrium earnings management is bL (1 g H ) 1 f , the fact that g H 0
v s
implies
bL s C M g H
1 f 0.
v s
Therefore,
bL 1 g H bL g
1 g H H bL 0.
The statements in the corollary follow directly from
FRQ
bL 1 g H
1 2 p 1 bL 1 g H
Q .
s 2 CM
prob error prob yL bL 1 g H f prob yL 1 g H f f 2 .
v s
prob error
f
s CM 2 C
M
g H
prob yL 1 g H 1 g H 1 2 f
2 f f 0.
v s s f
To ensure this condition holds, the term in brackets must be zero. Consider f 1 s
2 CM , where
1 s
2 CM f 0 . We have
CM g H g H
M
2 C 1 s
H
1 g 1 2 f
s
2 f f
s f
2 CM f
0.
0 due to f f 0
s
Thus, there must exist f1 such that prob(error) attains a maximum. If 1 < f1, this maximum
2C M
is actually not relevant and prob(error) strictly increases for f 0,1 .
Part (ii): As shown in the proof of Proposition 4 that bL 1 g H strictly decreases in C M
and CA,
implying
prob(error) prob( yL ) f
bL 1 g H
0 and
C M C M
0
prob(error) prob( yL ) f
bL 1 g H
0.
C A C A
0
Furthermore, AQ(f) > 0 for all 0 f s C M . Because AQ is differentiable (and thus continuous) in
f, there exists 0 f 2 s C M such that AQ attains its maximum at f 2 . The first-order condition is
AQ f 2 prob mH g
g H prob mH H 0 .
f f f
0 due to bL f 0
The relation f 2 f 0 cannot hold since g H f 0 f 0 at f0 and g H f 0 for all f > f0 (see
Proposition 2 (ii)), implying
AQ
0 for all f f 0 .
f
Thus, it follows that f 2 f 0 . If 1 f 2 , then the maximum is not attained over the relevant range
and AQ strictly increases for all f [0,1].
prob(yL )bL* CA
g
H f
prob(yL )bL* +prob(yH ) k
we rewrite AQ as
CA
AQ prob(yL )bL f .
k
CA
Since CM has only an effect on bL and there are no other effects of CM in AQ prob(yL )bL f ,
k
AQ
we have 0.
C M
Part (iii): First, since AQ = 0 if CA = 0 and AQ > 0 for all 0 C k , it follows that AQ strictly
A
increases in CA if CA is small. Second, the behavior of AQ depends on the behavior of the term bL R
where we use R C A k to save notation. Inserting for bL yields
s CM
b R 1 g H 1 f
L R,
v s
1 g H R 1 g g R 0
H
1 g H
g H
R.
H
R R R
g H prob yL mH
prob yL mH f fR .
R R
0 due to bL R 0
g H prob yL mH 2
Since R g H fR , substituting in the first-order condition yields
R R
g H prob yL mH 2
1 g H R 2 g H fR .
R R
0
Thus, if g H 1
2 , AQ cannot attain a maximum with respect to R. This will clearly be the case if the
largest possible equilibrium audit effort does not exceed ½. According to Proposition 3 (iii), g H
strictly increases in R for each f, thus we only have to consider R = 1 in that calculation. Furthermore,
Proposition 2 (ii) implies that there exists f0 such that g H attains its maximum with respect to f,
implying that the largest possible equilibrium audit effort results from applying
f min 1; f0 ( R 1) where f 0 ( R 1) is the value of f0 which results when inserting R = 1. If
f0 ( R 1) 1 , then the audit effort always strictly increases in f [0, 1] and the maximum is
reached at f = 1.
Let g H and bL denote the equilibrium strategies corresponding to f and R = 1. Then AQ will not
attain a maximum with respect to R if
s 1
f 0 R 1 1 M g H f 0 R 1
2 C 2
due to s CM.
k k
A prob mH g H 2 g H AQ 2 g H .
2 2
The first partial derivative of A with respect to f is
A k prob mH g k AQ g
g H 2 g H prob mH 2 1 g H H 2 g H AQ H .
f 2 f f 2 f f
0
A k prob mH g H
H H H H M
g 2 g prob m 2 1 g 0.
C M 2 C M C
0 0
Part (iii): Since g H 0 if CA = 0 and g H 0 for all 0 C k , it follows that A strictly increases in
A
CA if CA is small. To derive a more specific condition, we apply the procedure as in the proof of
Proposition 7 (iii). Rewrite A by inserting the equilibrium expression for g H to obtain (we use again
R Ck )
A
k
A prob mH g H 2 g H
2
k
prob yL bL fR 2 g H
2
k
prob yL f bL R bL R 1 g H .
2
s CM
b R 1 g 1 f
L
H 1 g H R ,
2
v s
2
implying that it is sufficient to consider 1 g H R . Applying the same steps as in the proof of
Proposition 7 (iii), this term always strictly increases in R if
Since
bL 3 f 1 bL 2 f 1 ,
this condition is stricter than the corresponding condition for bL R . Thus, if the new condition for the
probability ratio holds, both bL (1 g H ) R and bL R strictly increase in R implying that audit fee A
strictly increases in CA.
Proof of Proposition 9
The owner’s utility is
The expected outcome is constant, given a = aH. We first prove that each of the three cost
components increases in the bonus s, implying that the owner prefers the lowest s such that the
manager’s incentive compatibility constraint is satisfied.
Expected compensation is
prob rH s p 1 1 p p 1 p 1 bL 1 g H s .
We have
prob rH s prob rH
prob rH s ,
s s
where
prob rH
p 1 p 1
bL 1 g H
0 . The inequality follows from
s s
Corollary 3.
The derivative is
A prob mH k g
g H 2 g H prob mH k 1 g H H 0 .
s s 2 s
Since both prob mH (through bL ) and g H strictly increase in s (Corollary 3), the audit fee strictly
increases in s as well.
Since bL 1 g H strictly increases in s, the expected enforcement cost to the owner also strictly
increases in s.
Hence, the optimal s is the lowest s that satisfies the incentive compatibility constraint. The
manager’s utility conditional on the actions is
v
E[U M aH ] prob(rH ) s V prob( yL ) bL*2 prob( yL )bL* (1 g H* ) fC M
2
v
prob yH s prob yL bL 1 g H s 1 f Cs bL2 V
M
2
v
prob yH s prob yL bL2 V ,
2
v
E[U M aL ] prob(rH aL ) s prob( yL aL ) bL*2 prob( yL aL )bL* (1 g H* ) fC M
2
v
prob yH aL s prob yL aL bL2 .
2
Note that the optimal earnings management is independent from the productive effort choice and
its effect depends only on the probability of yL with which earnings management is undertaken.
Because the auditor does not observe the manager’s action choice, audit effort is also independent
from the productive effort, the auditor chooses his effort conjecturing aH.
V v
s bL2 .
p q 1 2
Define
v
H p q 1 s bL2 V 0 .
2
1
H H s s H H
j s j
0
j
j f ,C M
,C A k
j s
We have
H bL
1 v p q 1 .
s s
s 1
b 1 g H fC M 1 g H
L
v v
we obtain
bL g H
v 1 g H
fC M s .
s s
Substituting this term in the derivative of H yields
H M g H
s
g
H s
fC p q 1 0
s
s H
because 0 ≤ f ≤ 1, CM < s* and g H s 0 . Thus, sign sign .
j j
Setting j = f yields
bL s
because 0 . Thus, 0.
f f
Given the qualitatively similar relation between bL and CM as well as CA/k, we have
s s
0 and 0.
C M C A k
Proof of Proposition 10
Part (i) follows from
where j = f, CM, or CA/k, using the partial derivatives stated in Propositions 2, 3, 9, and Corollary 3.
Part (ii): To prove the effect of f on audit effort note that g H 0 if f = 0; if f s C M 1 for a
given s 1, g H 0 because bL 0 . Next, since s* strictly decreases in f, there is an fˆ such that
fˆ s ( fˆ ) C M 1, bL ( fˆ ) 0 and g H ( fˆ ) 0 . Due to differentiability and continuity of g H ,
there exists 0 f 4 fˆ where g H achieves its maximum (which is not necessarily within [0, 1], as in
Proposition 2 (ii)). At this point we have
dg H f 4 , s f 4 g H f 4 , s f 4 g H f 4 , s f 4 s
0,
df f s f
0 0
g H f 4 , s f 4
0.
f
g H f , s ( f )
0,
f
i.e., there is crowding out with a fixed bonus s ( f ) . It follows that there is also crowding out with
endogenous bonus s*(f) as
dg H g H g H s
0,
dC M C M s C M
0 0 0
1
bL
v
1 g H s fC M vbL 1 g H s fC M .
vbL
g H 1 .
s fC M
Totally differentiating this term with respect to R ≡ CA/k yields
db s
v s fC M L vbL
dg dR R
H
.
dR
s fC
M 2
dbL s
v s fC
M
vbL .
dR R
Using
bL
v
s R bL s bL
v 1 v
R b R R s
1 v L
s
from the proof of Proposition 9, expanding this expression yields
It follows that
M dbL s
s
dg H v s fC vbL g H s fC M
dR R R
s fC M 2
s fC M
2
dR
s g H
0
R s fC M
because s* R 0 , s C , f ≤ 1 and g H 0 .
M