Professional Documents
Culture Documents
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
Wiley is collaborating with JSTOR to digitize, preserve and extend access to Journal of
Accounting Research
Auditor Independence
RICK ANTLE*
that previous attempts at definitions were not made within the context
of formal, articulated models. The purpose of this paper is to formulate
and explore two plausible definitions of auditor independence within
such a model. The keys to the analysis are the explicit modeling of the
auditor and manager as expected utility maximizers and the characteri-
zation of the auditor's investigative and reporting activities. Since auditor
independence concerns the relationship between the auditor and man-
ager, this modeling of the auditor and manager is necessary to address
meaningfully the concept of auditor independence. The crucial issue is
the extent to which the auditor cooperates with the manager in pursuit
of their self-interests. The extremes of collusion and no cooperation are
delineated, as well as an intermediate form of auditor-manager cooper-
ation. The extent of such cooperation has a dramatic impact on the
optimal compensation scheme for the auditor. The game-theoretic foun-
dations of the model employed here are presented in Antle [1982].
Section 1 introduces the model and the definitions of independence;
section 2 is devoted to discussing these definitions; and section 3 sum-
marizes the paper.
1. The Model
.1 STRUCTURE
' For ease of exposition, I omit a formalization of measurability restrictions. Note that
all of the abstract spaces will be defined with accompanying sigma-algebras, and all
functions introduced will be appropriately measurable with respect to these sigma-algebras.
6 Obviously, other assumptions about sequencing are possible (e.g., the auditor observes
the manager's report before taking an action). I assume that the auditor and manager play
simultaneously because it is reasonable and results in a more tractable formulation. See
Antle [1982] for further discussion.
So, in the parlance of game theory, the strategies of the players in the
normal form of the game consist of: (i) an incentive plan (owner's
strategy); (ii) a function from I to A1 x M (manager's strategy); and (iii)
a function from I to A2 x Y (auditor's strategy). Notice that no mention
has been made of randomized strategies. An important limitation of the
analysis is that I shall restrict attention to "pure" or nonrandomized
strategies. This has some precedent (see, e.g., Harris and Raviv [1981]),
but the reader is referred to Antle [1982] for an example of its restric-
tiveness in the current context.
= f Ui[s(5(y(aj, a2, w)), z(a,, a2, w), Mr(m(al, w))), a,] dA;
= at U2[t('(y(aj, a2, w)), z(al, a2, W), mr(m(al, w))), a2] d1.
Having described the structure of the model, the strategy spaces, and
behavior representation, I now introduce the equilibrium concept. The
notion of auditor independence is immediately encountered. Game-the-
oretic equilibrium concepts are of two distinct types: cooperative and
noncooperative.7 Cooperative equilibrium concepts assume that the play-
ers can make binding commitments before the game is played. It is
obvious that a completely cooperative concept is not appropriate here,
since if the owner and manager could make any binding agreement they
wished, there would be no incentive for hiring an auditor.8 An element
of noncooperative play is therefore essential for my purposes and, indeed,
is usually present in agency analyses. Hence, it is assumed that all
agreements that the owner can make with the auditor and manager are
embodied in the incentive plan, which is consistent with the level of
detail of the model. That is, it would be senseless explicitly to model the
owner as choosing s(.) and t(.) and then use an equilibrium concept that
assumes the possibility of binding preplay commitments.
The behavior of the auditor and manager vis-a-vis one another now
needs to be specified, and this is where independence issues arise. At the
very least, an independent auditor would not collude with the manager
to the detriment of the owner. On the other hand, even an independent
auditor would not be expected to act against his/her own best interests
in carrying out his/her duties for the owner. But this self-interested
behavior must stop short of collusion. Exactly where the line is drawn
These are technical terms. The reader should not infer that the auditor or manager is
not "cooperative" in the colloquial sense.
8 One solution might be to have another party with whom risk may be shared-but the
there is no reason to call this other party an auditor.
For any incentive plan, I, the subgame defined by I is the game between
the auditor and manager whose strategies are A1 x M and A2 x Y and
whose payoffs are defined by I. Given an incentive plan, the above self-
interest assumptions imply that the auditor and manager play a Nash
(best response) equilibrium in the subgame defined by that plan. How-
ever, even if side-payments between the auditor and manager are ignored,
these self-interest assumptions do not completely specify the auditor's
and manager's behavior. It is well known that there are often multiple
Nash equilibria in games, and that these equilibria do not necessarily
provide the players with the same expected utility levels. This means
that for a given incentive plan, there may be several pairs of strategies
for the auditor and manager that meet the self-interest tests, and that
the auditor and/or manager may be worse off with some of these pairs
than others. The auditor and manager thus face the problem of which of
these alternative subgame equilibria they are going to play.
The problem of choosing among alternative equilibria is a familiar one
to most economists. It is often suggested that this selection is accom-
plished by adopting conventions or establishing institutions (see Demski
and Kreps [1982]). The concept and enforcement of auditor independence
may be one such convention. One reasonably stringent definition of
auditor independence would be to require that the auditor be willing to
'The notation "argmax f (v)" means the set of v's that maximize the function, f(.).
Problem 1.1
(a2*, idy) E argmax E U2; (a, *, idm), (a2, Y'( )), II (1.5)
A1xY
Conditions (1.2) and (1.3) require that the incentive plan provides the
manager and auditor with expected utilities at least as great as those
offered by their most favorable (un)employment alternatives. That is,
the parameters, 0, i = 1, 2, are the highest expected utility levels of the
manager and auditor if they do not accept employment with the owner.
Conditions (1.2) and (1.3) then merely reflect the assumption that the
owner employs both an auditor and a manager. Conditions (1.4) and (1.5)
reflect the private nature of the (a,, r (.)) and (a2, y(.)) choices. They
require that the manager and auditor choose actions and (truthful)
reporting strategies that are best responses given the incentive plan that
the owner selects.
Problem 1.1 is the owner's contracting problem with the manager and
a strongly independent auditor. It is essentially a multiperson agency
problem (see Holmstrom [1982a; 1982b]) and has the same economic
nature. The owner's goals are to motivate the auditor and manager to
take the proper actions and to share risk with them efficiently. Typically,
there is a trade-off between efficient risk sharing and motivation (see
Holmstrom [1979]).
V(al, m- ) Al X M.
I refer to the problem of maximizing (1.1) subject to (1.2)-(1.4) and (1.5*)
as problem 1.2.
(1.5*) requires the auditor's strategy to be dominant, and problem 1.2
is the owner's problem when the auditor is independent as defined below.
DEFINITION 2. The auditor is independent if s/he is willing to play any
dominant strategy in the subgame induced by I.
While independence is weaker12 than strong independence, it still
embodies an element of noncooperative behavior of the part of the
auditor. The fact that the independent auditor's strategy is dominant in
the subgame defined by I does not imply that the auditor and manager
could not both be strictly better off by colluding. To sharpen this
distinction, I shall define the lack of independence. An auditor who is
not independent is one who would engage in side-payment schemes with
the manager. The following definitions formalize these ideas.
DEFINITION 3. A feasible scheme of side-payments from the manager to
the auditor is a function p: Y x Z x M -* R such that:
(i) If p(5, z, rm) > 0, then p(5, z, mr) c s(5, z, mi).
(ii) If p( ', z, mr) < 0, then -p(5, z, mr) c t(5, z, m).
In addition, p(., *, 0 is always considered feasible.
" See Mookherjee [1982] for another (apparently less tractable) formulation of con-
straints to eliminate dominated subgame equilibria. Also, in a restricted setting, Sappington
and Demski [1983] have shown that inducing one agent to play a dominant strategy is the
least costly way to eliminate such equilibria.
l2 Independence is weaker in the sense that it does not presume that the auditor will
always play the Nash equilibrium that the owner desires, especially even when there exists
an alternate equilibrium in which the auditor is better off.
Restrictions (i) and (ii) are necessary to ensure that the party from
whom wealth is transferred has sufficient resources to fulfill his/her
obligation.
DEFINITION 4. A nontrivial side-payment scheme is one which is not
identically zero.
DEFINITION 5. The auditor is not independent if the auditor and
manager choose a feasible, nontrivial side-payment scheme, p(., , ),
and strategies, (a&, rh(.)) and (62, rh(.)), that do not solve problem 1.2.
Several authors (e.g., Mautz and Sharaf [1961], Nichols and Price
[1976], Rittenberg [1977], and Watts and Zimmerman [1979]) have
discussed independence in terms of an auditor's resistance to managerial
pressure or interference. This notion of independence (refusing to ac-
quiesce to the manager's attempted influence) is captured here by the
independent auditor's refusal to engage in side-payment schemes with
the manager. It is important to note that in terms of the model, this
refusal is once again a denial of completely self-interested behavior. The
pressures on auditors to maintain their independence are not modeled
here, but I do discuss several possibilities later.
The three definitions of strong independence, independence, and lack
of independence form the basis for the rest of this paper. In the next
section I discuss these definitions and their implications, using examples
to point out the distinctions among them.
2. Discussion of Independence
Now that I have specified the model and defined strong auditor
independence, it is worthwhile to digress for a moment, before probing
auditor independence further, and discuss the role of the auditor. To
simplify the discussion, I assume that the owner is risk-neutral. This
implies that fully efficient risk sharing calls for the owner to bear all
risk.
To begin, consider the owner's problem of contracting with the man-
ager in the absence of an auditor. The owner would like to motivate the
manager to supply an appropriate level of effort, a,. If the effort level is
not mutually observable, the owner must resort to indirect means of
motivating the manager. Usually, this involves basing the manager's
compensation on some possibly noisy and often costly measure of man-
agerial input. To the extent that the measure is noisy, risk is placed on
the manager, for which s/he must be compensated. Thus, inefficient risk
sharing arises for motivational purposes.
In the model presented here, there is more detail than the discussion
above takes into account. The source of one of these "input measures"
has been explicitly modeled. That is, the accounting report, m4, might be
used to motivate the manager. Put aside the problem of getting the
Example 2.1
m 0
a,
0 .75 .25
1 .125 .875
The ij entry in the above matrix is the probability that m is equal to the
first entry in column j when a, is equal to the first entry in row i. For
example, when a1 = 0, m = 0 with probability .75 and m = 1, with
probability .25. Similarly the matrix giving the distribution of z, which
is independent of m and y, is:
z 0 1
a2~~~~~~.
0 .7, .3
l l 4 .6
Notice that the distribution of z depends only on a2. Finally, the distri-
bution of y depends both on m and a2. When a2 = 0, y is independent of
m and equals 0 with probability .125 and 1 with probability .875. In
contrast, when a2 = 1, y depends on m as follows:
y 0 1
m
0 .8 .2
1 .1 .9
The intuition behind this specification is that the signal y is informative
about m only when the auditor works.
To complete the description of the example, assume that the owner is
risk-neutral and the auditor and manager are risk- and work-averse as
follows:
Ui(c, a) = -- V(ai) i = 1, 2,
where V1(0) = V2(0) = 0, V1(I) = 30, V2(1) = 20. The minimum utility
requirements are 230 for the auditor and 200 for the manager, that is, 01
= 200 and 02 = 230.
Optimal contracts can be found using quadratic programming
techniques13 on the appropriate version of problem 1.2. The optimal
contract for the manager does not depend on z and is given below:
(0, 0) $35,344
(0,1) 30,376
(1, 0) 35,344
(1, 1) 58,980
13 Grossman and Hart [1983] analyze the mathematical properties of problems such as
the one contained in this example.
is placed on the manager because it is the most efficient way for the
owner to motivate hard work and truthful reporting.
Similar observations apply to the auditor's contract, which is given
below:
(9 z, rm) t( 5, z, m)
(0, 0, 0) $70,480
(0, 1, 0) 73,436
(0, 0, 1) 44,719
(0, 1, 1) 65,310
(1, 0, 0) 10,219
(1, 1, 0) 50,187
(1, 0, 1) 55,202
(1, 1, 1) 68,785
The expected payment to the auditor with this contract is $62,964, and
the owner's expected utility with this incentive plan is k - 116,487.
Note how the variables 5, z, and mr affect the payments to the auditor.
For any given pair of reports by the manager and auditor, the auditor
receives more if z = 1 than if z = 0. This reflects the fact that observing
z = 1 is more likely when the auditor works than when s/he does not
work. For any given z, the auditor receives more when the audit report
substantiates the manager's report (i.e., mr = ') than when the reports
conflict (5 # m). This is because, given that the manager reports truth-
fully, the audit report is more likely to substantiate the manager's report
when the auditor works than when s/he does not work. The optimal
contract for the strongly independent auditor utilizes these relationships
efficiently to motivate the auditor and induce truthful reporting.
Example 2.2
Assume the same structure as in example 2.1, except that now assume
that problem 1.2, not problem 1.1, is the appropriate formulation. There
is no effect on the optimal contract for the manager, so it is as presented
in example 2.1. However, the optimal auditor's contract now depends
only on z. When z = 0 the payment to the auditor is $44,100, and when
z = 1, s/he receives $76,544. The expected payment to the auditor is
$63,566, and the owner's expected utility is k - 117,089.
The owner's expected payment to the independent auditor is $602
higher than the expected payment to the strongly independent auditor.
This reflects the fact that problem 1.2 is generated by adding (in this
case binding) constraints to problem 1.1.'4
It is intriguing that the auditor's contract given in example 2.1 could
also have been generated by adding to problem 1.1 constraints which
require that the auditor's contract not depend on either the manager's
or the auditor's report.'5 This is consistent with AICPA and SEC rules
which forbid independent auditors from owning shares of their clients'
stock and from possessing any direct financial interest or material
indirect financial interest in their clients. It would appear that these
rules are perverse, because an auditor with a healthy stake in a client
would have an incentive to conduct a thorough, cost-effective audit. This
line of thinking no doubt contributed to the regulations in England circa
1850 that required auditors to hold shares in clients (see Berryman
[1974]). But the logic behind the AICPA and SEC rules, as I understand
it, is that the auditor's financial interest in the client affects his/her
incentives to report truthfully because the audit report affects the value
of the client firm. Example 2.1 demonstates that if the auditor were
strongly independent, efficient auditor's contracts (which induce truthful
14 This observation implies that the owner is always at least as well off with a strongly
independent auditor as with an independent auditor.
15 As of this writing, I am investigating the extent to which this result generalizes.
As a final point about example 2.2, note that the owner can be made
strictly worse off by the auditor's failure to be independent.'8 There are
several ways that auditor-manager cooperation can be detrimental to the
owner. The auditor may ensure the manager by not truthfully reporting
as the owner expects. For example, the manager could pay the auditor to
"smooth" his reports. In example 2.2, the manager would be willing to
pay the auditor up to $18,980 in return for a guarantee that the audit
report is 1. (This is the amount that leaves the manager indifferent
between working and telling the truth when the auditor works and reports
truthfully, and not working and always reporting 1 when the auditor
always reports 1.) If the manager can induce the auditor into accepting
the side-payment, s/he will supply no effort, and the owner's expected
utility would fall to -122,546. This illustrates the consequences to a
naive owner who does not anticipate the auditor's lack of independence.
When choosing the incentive plan, the owner has presumed that the
auditor and manager will play noncooperatively in the subgame induced
by the incentive plan. An intelligent owner would not make such an
assumption unless it were justified.
Conventional wisdom goes a step further. In the context of example
2.2, if the owner realized that the auditor and manager were going to
cooperate, s/he would not hire the auditor. It would be optimal merely
to accept the fact that the manager could not be induced to supply more
16 The contract might depend on the manager's report if it was informative (see Chris-
tensen [1981] and Holmstrom [1982a]) about the auditor's effort or information.
17 The issue of when this use of the audit report is optimal is formally equivalent to that
of when participation in the setting of performance standards is of value (see Christensen
[1982], Baiman and Evans [1983], Dye [1983], and Penno [1984]). For example, in a model
endowed with much continuity, Dye demonstrates results which imply the following in
problem 1.1: if in the presence of y, z and m are not marginally informative with respect to
a2, then there is some gain to including the auditor's report of y in the audit contract.
18 Because the owner can directly implement any scheme that the auditor and manager
can achieve via side-payments, the owner is at least as well off with an independent auditor
as with an auditor that is not independent.
than the minimal effort and just shut the firm down.'9 In this sense,
independence is a necessary attribute of any auditor that the owner would
be willing to hire in the context of this example. However, my conjecture
is that this "necessity" flavor of independence is far from generally true.
If the owner could monitor the transactions between the auditor and
manager, the owner could more effectively orchestrate their interaction.
If the owner had any means other than the auditor to control the
manager, perhaps the gains to the manager from making side-payments
to the auditor could be greatly reduced.20 The consequences of a lack of
independence depend on the owner's opportunities to detect and penalize
the auditor and manager for shirking and/or not telling the truth.
The analysis thus far indicates that the owner would prefer an inde-
pendent auditor to one that is not independent. However, given an
incentive plan, the auditor would be at least as well off by foregoing
independence as s/he would be by maintaining it. The model contains
no forces that induce the auditor to maintain independence. This raises
the issue of the mechanism(s) by which independence might be guaran-
teed.
There are several possibilities, and casual empiricism suggests that all
play a role. First, there may be opportunities to monitor the transactions
between managers and auditors. The concern over management advisory
service contracts may have arisen, at least in part, from the fear that
managements could use these contracts as a vehicle for side-payments.
Also, witness the movement toward audit committees to interface with
external auditors.2' Finally, any auditor caught taking a direct bribe from
management would certainly be punished (e.g., through lawsuits).
Another possibility is the role of multiperiod effects. Two distinct types
of effects may be isolated. First, if owners and auditors could write
multiperiod contracts and if the auditor is independent, one could expect
a distribution of the various audit reports in the long run.22 If the observed
long-run frequency of audit reports indicates that the auditor has prob-
ably not worked and truthfully revealed his/her information, the contract
could call for penalties to be imposed on the auditor (see Radner
[1981]).
"9Another alternative is to sell the firm to the manager and endure the resulting
suboptimal risk sharing. The assumption that x is not observable precludes this in the
context of the example. See Antle [1982] for further discussion.
20 Another possibility is that the owner could purchase information about the auditor's
information. This could be used to impose costs on the auditor for deviating from the
truthful communication of y and, thus, raise the cost of inducing the auditor to lie.
21 This, however, creates another layer of incentive problems.
22 This effect could be used to provide incentives for auditors with multiple clients in a
single period as well. In such a setting, the percentage of the various types of audit reports
issued could provide information about whether the auditor was independent.
3. Summary
23 In fact, reputation effects may help explain the partnership structure of these firms.
See Fama and Jensen [1983a; 1983b] for more discussion of possible determinants of this
structure. Also, Wilson [1983] discusses the market for auditing services and raises several
interesting issues regarding auditor's reputations.
REFERENCES
ANTLE, R. "Moral Hazard and Auditor Contracts: An Approach to Auditor's Legal Liability
and Independence." Ph.D. dissertation, Stanford University, 1980.
- "An Agency Model of Auditing." Journal of Accounting Research 20 (Autumn 1982,
pt. II): 503-27.
BAIMAN, S., AND H. EVANS. "Decentralization and Pre-Decision Information." Journal of
Accounting Research 21 (Autumn 1983): 371-95.
BENSTON, G. J. "Accountants' Integrity and Financial Reporting." Financial Executive
(August 1975).
BERRYMAN, R. G. "Auditor Independence: Its Historical Development and Some Proposals
for Research." Proceedings of the 1974 Arthur Andersen/University of Kansas Symposium
on Auditing Problems. Lawrence: University of Kansas, 1974.
CHRISTENSEN, J. "Communication in Agencies." Bell Journal of Economics 12 (Autumn
1981): 661-74.
. "The Determination of Performance Standards and Participation." Journal of
Accounting Research 20 (Autumn 1982, pt. II): 589-603.
DATAR, S. "A Multiperiod Model of Auditing." Working paper, Stanford University, 1983.
WATTS, R., AND J. ZIMMERMAN. "The Markets for Independence and Independent Audi-
tors." Working paper, Graduate School of Management, University of Rochester, July
1979.
. "Auditors and the Determination of Accounting Standards." Working paper no.
GPB-78-06, University of Rochester, August 1981.
WILSON, R. "Auditing: Perspectives from Multi-Person Decision Theory." The Accounting
Review 58 (April 1983): 305-18.