Internal Control versus External Manipulation: A Model of Corporate Income Tax Evasion

Author(s): Kong-Pin Chen and C. Y. Cyrus Chu

Source: The RAND Journal of Economics, Vol. 36, No. 1 (Spring, 2005), pp. 151-164
Published by: Wiley on behalf of RAND Corporation
Stable URL: http://www.jstor.org/stable/1593759
Accessed: 08-02-2016 11:00 UTC

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp

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.

RAND Corporation and Wiley are collaborating with JSTOR to digitize, preserve and extend access to The RAND Journal of
Economics.

http://www.jstor.org

This content downloaded from 14.139.211.75 on Mon, 08 Feb 2016 11:00:36 UTC
All use subject to JSTOR Terms and Conditions
 RANDJournalof Economics
Vol.36, No. 1, Spring2005
pp. 151-164

Internal control versus external

a model
manipulation: of corporate
income tax evasion

Kong-Pin Chen*
and

C. Y. Cyrus Chu**

Weofferaformal modelof corporateincometaxevasion. Whileindividualtaxevasionis essentially

a portfolio-selectionproblem, corporate income tax evasion is much more complicated. When
the owner of a firm decides to evade taxes, not only does she risk being detected by the tax
authorities, more importantly,the optimal compensationscheme offered to the employees will
also be altered. Specifically,due to the illegal nature of tax evasion, the contract offeredto the
manager is necessarily incomplete.This creates a distortionin the manager'seffortand reduces
the efficiencyof the contract.Taxevasion thus increases theprofitretainedby thefirm not only at
the risk of being detected,but also at the cost of efficiencyloss in internalcontrol.

1. Introduction
* The classic article of Allingham and Sandmo (1972) resultedin much theoreticaland em-
piricalanalysis on why and to what extent an individualwould evade taxes.1However,corporate
income tax evasion has to date attractedrelativelyscanttheoreticalinvestigation.2One reasonfor
such an unbalanceddevelopmentin the researchof tax evasion is that essentially no corporate
income compliance microdatahave been previously availableto researchers.Another and per-
haps more importantreason is thatthe conceptualdifferencein the evasion decision between an
individualand a corporationis hardto captureanalytically.The standardanalyticalframeworkof
individualtax evasionessentiallytreatsthe decision as a portfolio-selectionproblem.The amount
of tax evaded is the risky asset that yields a higher payoff if evasion is not detected by the tax
authority,but a lower payoff if it is. The individualselects the optimalamountof the risky asset,

* Academia Sinica and NationalTaiwanUniversity;kongpin@gate.sinica.edu.tw.

** Academia Sinica; cyruschu@sinica.edu.tw.
We thank Gregory Besharov, Sheng-Cheng Hu, Jeffrey MacKie-Mason, Debraj Ray, Agnar Sandmo, Tomas
Sjostrom, Henry Wan, Lixin Ye, and seminarparticipantsat Duke, NYU, Ohio State, and Penn State for their useful
comments. We are especially gratefulto the EditorRaymondDeneckere,Peter Katuscak,JenniferReinganum,and two
referees, who providedespecially insightfulcriticisms and constructivesuggestions.
1 See, for
instance,Cowell (1990) and Andreoni,Erard,and Feinstein (1998) and the referencestherein.
2 Kreutzerand Lee
(1986), Wangand Conant(1988), and Chu (1990) are partof the scant literatureon corporate
income tax evasion.

Copyright? 2005, RAND. 151

This content downloaded from 14.139.211.75 on Mon, 08 Feb 2016 11:00:36 UTC
All use subject to JSTOR Terms and Conditions
 OFECONOMICS
152 / THERANDJOURNAL

given the probabilityof being detected and the penalty imposed by the tax authority.3A natural
question to ask is: Can the evasion behaviorof corporatemanagersbe satisfactorilyexplained
by the traditionaltax evasion model for individuals?Our answer to this question is "no."Tax
evasion by a corporationis much more complicatedbecause it involves the strategicbehaviorof
more thanone person,therebychangingthe relationbetween the firmand its managerand, in the
process, distortingthe incentivesof the latter.
Our argumentis based on a critical insight: a contractthat is based on illegal actions will
not be honoredby the court and is thus not enforceable.More specifically, suppose a principal
hires an agent to engage in an illegal activity.Dependingon the effort level of the agent's illegal
action, there are severalpossible outcomes. This means that in orderto induce effort, the payoff
of the agent must be contingenton outcomes. Since the agent participatesin an illegal activity,
he bearsthe risk of being detectedand penalized.To induce him to participate,the contractneeds
to compensatehim not just for his effort, but also for this risk. Efficientcontractingcalls for the
principalto sharethe risk with the agent, by paying the agent a higherwage when the illegal act
is detected (and the agent punished).That is, the agent's pay needs to be contingenton whether
or not the illegal action is detected. However, this is practicallyimpossible. The principalcan
always renege on the contractby refusingto pay the agenta higherwage when the illegal actionis
detected.The courtwill not honorthe termsof the contracteven if the agentfiles a lawsuit,because
it is based on illegal activity.Knowing this, the agent will insist thatthis risk be compensatedex
ante throughthe laborcontract.But if this is so, the contractwill be incompletein the sense that
the agent's pay will be the same regardlessof whetherthe illegal action is detected or not. The
contractthus plays the double role of rewardingthe agent for his effort and for his bearingthe
risk of tax evasion. Consequently,thereis distortionin the agent's incentivesin effort.
In the context of tax evasion, suppose a firm consists of a risk-neutralprincipaland a risk-
averse agent. The latter,being an agent of the former,is responsiblefor the operationof the firm,
includingfiling tax returns.As such, tax evasion requiresthe cooperationof the manager.By the
same argument,the labor contractoffered to the agent will fail to have the principaland agent
shareevasion risk efficiently,essentially because compensationto the lattercannotbe contingent
upon whether evasion is detected or not. This incompleteness in contractwill thus distort the
effort of the agent.
The argumentabove implies thatthe firmintendingto evadetaxes has to balancethe tradeoff
between two considerations.On the one hand, tax evasion can increase its expected after-tax
income. On the other hand, beside the risk of detection consideredin the traditionalindividual
tax evasion literature,the firmalso needs to bearthe cost of efficiency loss in internalcontrol.In
a word,beside the traditionalincome-versus-risktradeoff,thereis an additionaltradeoffbetween
internalcontrol and expected gain in evasion. The latteraspect of evasion cost is one that is not
and cannot be formulatedin the standardindividualtax evasion model. We need to emphasize
that the distortionof effort does not come from the risk of being detected by the tax authority
per se, but from the fact that the structureof the compensationscheme for the managerhas to
be changed if the principaldecides to evade taxes. In essence, this is because individualincome
tax evasion is a single-persondecision problem,whereascorporateincome tax evasion inevitably
involves the interactionof many persons. It is in this sense that corporateincome tax evasion is
much more complicatedthanindividualincome tax evasion.
Ourmodel implies that althougha risk-neutralindividualwill evade taxes if and only if the
expectedprofitfrom evasion is greaterthanthatfromreportinghonestly,a risk-neutralownerof a
firmwill evade tax only when the expectedprofitfrom evasion is greaterthanthatfrom reporting
honestly by a substantialmargin.Even if the expected tax savings from evasion is positive and
the principalis risk neutral,she will not necessarily choose to evade taxes, i.e., the condition
for profitabletax evasion is more stringentfor firms than for individuals,because the internal
efficiency loss may outweigh the expected gain from evasion.

3 Laterworks have also incorporatedlaborsupplydecision into the model. In these cases an individual'sdecision
to evade taxes is more than just a portfolio-selectionproblem. See, for example, Cowell (1981), Sandmo (1981), and
Andreoni,Erard,and Feinstein (1998).
? RAND 2005.

This content downloaded from 14.139.211.75 on Mon, 08 Feb 2016 11:00:36 UTC
All use subject to JSTOR Terms and Conditions
 CHENANDCHU / 153

Althoughthe source of inefficiencywe discuss above is based on the idea thatillegal action
results in incompleteness of contract and thus loss of efficiency in internalcontrol, there are
potentially other reasons that can also incur internalcontrol inefficiency when the firm evades
tax. We shall explore later in the article, somewhatinformally,two of them that we believe are
equally important.
The remainderof the articleis organizedas follows. In Section 2 we set up the formalmodel
andidentifythe sourceof inefficiencyin internalcontrol.In Section 3 we discuss some extensions
of the basic model and make qualificationsfor our results againstthese extensions. In particular,
we discuss the cases when the managercan extortpaymentfrom the owner,and when they have
repeatedinteraction.We also exploreotherpossible sourcesof inefficiencyas the firmevades tax.
Section 4 concludes.

2. The model
* Consider a standardprincipal-agentmodel in which a risk-neutralprincipal(owner) of a
firmhires a risk-averseagent (manager)to producean outputy.4 For simplicity we assume y to
be the profit(gross of wage cost) of the firm.The realizationof y is stochasticanddependson the
effort level of the manager,denoted e. The value of profit,however,is verifiableand observable
to both principaland agent. Let f(y I e) be the density functionof the realizationof y when the
effort level of the manageris e, with supporton IW+. We make the following assumption.
Assumption1. f(y I e) first-orderstochasticallydominatesf(y Ie') if e > e'.
This assumptionis standardin the literatureof principal-agentanalysis(see, e.g., Holmstrom
(1979)). It implies that when the managerexerts more effort, the expected profitof the firmwill
be higher.The net profitof the principalis y - w, where w is the compensationfor the manager.
There is also a profittax (with rate t) thatthe principalhas to pay, so her after-taxprofit(utility)
is (1 - t)(y - w). The utility of the manageris assumed to be u(w, e) = u(w) - v(e), where
u(.) is the agent's utility in income and v(.) the disutility of effort. Assume thatu' > 0, u" < 0,
v' > 0, v" > 0, and v'(O)= 0. The last assumptionessentially guaranteesan interiorsolution for
the agent's effort. This facilitatesour presentationbut is not importantfor the main argument.

o The honest principal. If the owner does not evade tax, her optimizationproblemis
ProblemH.

max(l - t)E[y - w(y)]

w(y),e

subjectto e e argmax
el
E[u(w(y)) - v(e')],
E[u(w(y))- v(e)] > U,

where U is the reservationutility of the manager.The first constraintabove is the incentive-

compatibility(IC) constraint,and the second is the individual-rationality(IR) constraint.
Assuming thatthe first-orderconditionapproachis valid, problemH becomes

max(l - t) -
w(y ))f (y Ie)dy
w(y),e

subjectto J u(w(y))fe(y I e)dy - v'(e) = 0,

u(w(y))f(y e)dy- v(e) > U.

4 From now on we will use

agent and managerinterchangeably,and likewise for owner and principal.We also
referto the principalas "she"and the agent as "he.
? RAND 2005.

This content downloaded from 14.139.211.75 on Mon, 08 Feb 2016 11:00:36 UTC
All use subject to JSTOR Terms and Conditions
154 / THERANDJOURNAL
OFECONOMICS

The first-orderconditionsfor problemH are thus

l-t . fe(Y Ie)

u'(w(y)) ff(y e)

(1- t) f(y- w(y))fe( I e)dy + t u((yW))fee(y I e)dy - v"(e) = 0,

where AL> 0 is a multiplierfor the IC constraint,andX > 0 for the IR constraint.Furtherassume

the following.

Assumption2. f(y I e) satisfies the monotonelikelihood condition:fe/f is increasingin y.

The solution to problemH is the second-bestoutcome for internalcontrol, and we denote
the optimal compensationof problemH by wl(y). By Assumption2 we know that wl(y) is an
increasing function of y. Moreover,since v'(0) = 0, we also know that the optimal effort level
e* >0.

[ The evading principal. If the principalintends to evade tax, then after the value of y is
realized, she decides upon the value of y to be declaredto the tax authority.Let r be the value
of y declared.Assume that p is the probabilitythatthe firmis auditedby the tax authority,5and
that once the firm is audited, evasion will be detected for sure. Let q(y - r) and x(y - r) be
the penalties to the owner and manager,respectively,for the amountof income evaded, y - r.
Assume the following.

Assumption3. q(O) = O,q'(z) > O,q"(z) > 0, and limzoo q'(z) = oo for all z > 0. 6 Similarly
for x(z).
The key observationis that a contractis meaningfulonly if it is legally enforceable.The
wage contract wl(y) discussed in the previous subsection is enforceable because if the owner
refuses to pay the amountw (y) to the manageraftery is realized,the lattercan take the contract
to courtand demandpayment.Since y is verifiable,the courtwill honorand enforce the contract.
In contrastto the case of the honest principal,in the case of tax evasion, even if the truevalue of
profity is verifiableto all, the wage has to be a functionof reportedprofit,r, ratherthanthe real
profity. That is because if r is the official profitdeclaredto the tax authority,then a wage thatis
contingenton y will have to be implicit and thus has no legal power.This implicit contractwill
then be renegedon by the principal:since w1(y) increasesin y, and since the reportedprofitr will
always be smaller than real profity if evasion is to be profitableat all, the firm is always better
off paying the managerwl(r) than wl(y). But given thatthe implicit contractis not enforceable,
the owner always has incentiveto renege on the contractafterr is reportedand pay accordingto
r ratherthan y. This implies thatthe wage contracthas to be contingenton r ratherthany.
The firm's optimizationproblem, if it intends to evade tax, consists of two stages. In the
first stage, it offers a contractto the manager.In the second stage, after y is realized, it decides
how much tax to declare.We will solve the problemby backwardinductionand thus discuss the
second-stagetax-reportingproblemfirst.It turnsout thatour resultcriticallydependson whether
the manageris liable for tax evasion if it is detected.We will thusdiscuss the two cases separately.
Themanageris not liablefor evasion. In this subsectionwe discuss the case when the manageris
not legally liable for evasion.It can, however,be arguedthatsince the manageris by definitionthe

5 In this articlewe have focused on the

evadingbehaviorof taxpayers,but we do not go into the government'sor
the tax authority'spolicy responseagainstevasion.The auditingprobabilityis thusindependentof firmcharacteristics.For
discussionof this aspect,see, for example,CremerandGahvari(1996), Dubin,Graetz,andWilde (1992), andMookherjee
and Png (1992).
6
If q(.) is not assumedto be an increasingfunction,then the detectionprobabilityp has to be a functionof r- y
in orderto avoid a corer solution(i.e., for the second-orderconditionof (1), which is to be derivedshortly,to hold, which
involves more tedious algebrabut offers the same insight).
? RAND 2005.

This content downloaded from 14.139.211.75 on Mon, 08 Feb 2016 11:00:36 UTC
All use subject to JSTOR Terms and Conditions
 CHENANDCHU / 155

agent of the firm who runs the business on behalf of the owner,he cannotdistancehimself from
evasion.Thus the nonliablecase discussedhere is more for the purposeof providinga benchmark
(and a framework)in facilitatinglatercomparisonthanfor the purposeof its result.
In the case when the manageris not liable for tax evasion, x(y - r) = 0 for all r and y. The
managerthus suffers no loss of utility when evasion is detected. Consequently,the owner does
not have to compensatehim for any risk involvedin evasion, and needs to providehim with only
enough incentive to declarethe amountof income thatmaximizes expected after-taxprofit.First
we discuss the second-stageproblem.

Second stage. Suppose w(r) is the wage function offered to the managerin the first stage.
Since y > r if evasion is to be profitable,by Assumption2 we know that w(y) > w(r). This
means that even if the manageris not liable for evasion, he has to be providedwith incentive to
underreportprofit,because he is paid less underthe reportedincome r thanunderthe real income
y. Since the loss of the managerfor reportingr (instead of real profit y) is w(y) - w(r), the
owner needs to compensatethe managerby this amountin orderto induce him to underreport.7
The owner and managerwill thus need a new contractto transferthis payment.They can sign a
"service"contractaskingthe managerto reportr, andin returnfor this service the ownerwill pay
him an amountw(y) - w(r). Forexample,let s = w(y) - w(r). Then the termsof the contractcan
be as follows: "Herebythe owner of the firmasks the managerto preparethe tax returnon behalf
of the firm.The latteracknowledgesthat the profitfor the firm is r. In returnfor this service the
firm agrees to compensatehim by an amounts." This is an enforceablecontractbecause if the
firmfails to carryout the termsof the contractafterr is reported,the managercan take it to court,
show the reportedprofitr on the tax return,and demandpayment.
Note that in general the court will not delve into the issue of whetherr is the true profit
when asked to enforce this service contract.As such, tax evasion will not be exposed because
of this contract.The owner, however,might threatenthat if the agent goes to the court, she will
expose the evasion so as to invalidatethe contract.This threatis not credible,since in thatcase the
ownerwill also be penalizedand will owe morein penaltythanthe amountshe owes the manager,
w(y) - w(r). On the other hand, if the managerfails to reportr as profit, the firm is underno
obligationto pay for it.8 Given the service contract,the manageris fully compensatedfor the loss
of wage income w(y) - w(r) and is thus willing to underreportprofit.Denote this scenarioby the
contract{w(r); w(y) - w(r)}, where w(r) is the wage contractin the firststage, and w(y) - w(r)
the service contractin the second. Note thatunderthe extendedcontract,the wage of the agent is
w(r) + (w(y) - w(r)) = w(y), which is exactly the same as when the agent is paid accordingto
the real profity. That is, under {w(r); w(y) - w(r)}, the agent is actuallypaid accordingto real
profity althoughit is r thatis declared.
It is importantto emphasize that the service contractdiscussed above is needed only if the
wage of the managermust depend on reportedprofit r ratherthan real profit y (as is claimed
above). If, on the other hand, the principalcan manageto pay the agent accordingto y directly,
then there is no need for the service contractbecause, as can be seen from the argument,the
function of the service contractis to make sure that the agent is paid accordingto real profity,
even when he reportsr. Put differently,it is actuallynot importantfor our argumentwhetherthe
agent's wage contractdependson r or y. They both lead to the principal'smaximizationproblem
thatwe discuss below. This fact also implies thatthe efficiency loss in controlthatwe claim does
not come fromthe fact thatthe wage contractneeds to dependon reportedprofit.The second-stage
optimizationproblemof the principalis thus, given the value of y and the fact thatthe agent is de

7
In fact, to induce the agent to reportr as profit,it might not be enough to pay him only w(y) - w(r). He might
ask for more than that amountfor participatingin collusion. That is, he might extort more paymentthan w(y) - w(r)
from the principal.This case is discussed in Section 3.
8 In fact there is no incentive for the agent to renege, since he is fully compensatedon the one hand, and is
paid
after r is reportedon the other.
? RAND 2005.

This content downloaded from 14.139.211.75 on Mon, 08 Feb 2016 11:00:36 UTC
All use subject to JSTOR Terms and Conditions
156 / THE RAND JOURNALOF ECONOMICS

facto paid accordingto w(y), to maximize expected after-taxprofit:

max(l - p)[y - w(y) - t(r - w(y))] + p[(l - t)(y - w(y)) - q(y - r)].
r

The first-orderconditionis
)t
q'(y - r) = (1 (1)
p
One can easily verify thatthe second-orderconditionholds. From(1) we have y = r + q'/- [(1 -
p)t/p] . Thatis, the firmwill not evadetax if therealizedgrossprofitis lowerthanq- 1[( 1- p)t/p];
it will evadean amountq'-1 [(1 - p)t/p) if otherwise.The optimalreportedprofitis thusa function
of real profit:

y /q-l(q -Pt) ify>q'-l(1 -Pt);

(2)
0, ify < q'1(-Pt)

By Assumption3 we know thaty > r. It can easily be seen from (1) that ar/ay = 1; thatis,
the amountevaded is independentof the realizationof y. We note thatthe second-stageproblem
for the principalis exactly the individualincome tax evasionproblemwith a risk-neutraltaxpayer,
as consideredin, for example, Srinivasan(1973).
First stage. Given the second-stageoutcome, the principaldesigns a contractto maximize
her after-taxincome.
Underthe contract{w(r); w(y) - w(r)}, the manager'sutility is E[w(r) + (w(y) - w(r))] -
v(e) E[w(y)] - v(e), which is exactly the same as that in problemH. Similarly,the expected
=
profitof the owneris the gross profitfrom which are subtractedthe two paymentsto the manager,
which is

(1 - p) [y- w(r) + (w(y) - w(r)) - t(r - w(y))]f(y e)dy

+ p [y - w(r) + (w(y) - w(r)) - t(y - w(y)) - q(y - r)]f(y Ie)dy

= (1 - t) (y - w(y))f(y I e)dy

+ ( - p)t (y -r)f(y Ie)dy - p q(y -r)f(y | e)dy. (3)

Although (3) is somewhat different from the principal's expected profit in problem H, their
correspondingsolutions are obviously the same.9This means that the optimalwage contractfor
the evasion problemsatisfies w(r) = w(r(y)) = wl(y). Consequently,the second-besteffortfrom
the managerwill be induced.This is essentially because the second-stagetax-reportingproblem
can be made independentof the first-stagetraditionalprincipal-agentproblem.That is, they are
actuallytwo independentdecision-makingproblems.The ownercan thuschoose the optimallevel
of r in the second stage without affecting the optimal solution for the labor contractin the first
stage. We thereforehave the following.
Proposition 1. If the manageris not liable for tax evasion, then there will be no efficiency loss
in internalcontrol for evading tax. The principaloffers the contract{w(r); w(y) - w(r)} to the
agent, and the amountof income evaded is q'-l[(1 - p)t/p] if y > q'-l[(l - p)t/p], and zero
otherwise.

9 This is because the optimalamountevaded, y - r, is a constantby (2), and the second and thirdintegrationsin
(3) are both constants.
? RAND 2005.

This content downloaded from 14.139.211.75 on Mon, 08 Feb 2016 11:00:36 UTC
All use subject to JSTOR Terms and Conditions
 CHENANDCHU / 157

The scenariowe discuss correspondsclosely to the so-called two-accounting-bookpractice

that in the real world is very commonly used by tax-evadingfirms,especially in less-developed
economies. In this practice,the firmkeeps two sets of accountingbooks, one that recordsall the
real values of revenueand cost that are used for internalcontrol, and a second that has the false
accountinginformationused for tax purposes.Interpretedin termsof our model, the owner uses
the book thatcontainstrueaccountinginformationto controlthe managerby offeringhim contract
wl(y), and she uses the book thatcontainsfalse information(r) andthe paymentw(y) - w(r) for
the agent to evade tax. By doing so the firmessentially separatestax evasion and internalcontrol
into two independentdecision-makingproblems.Consequently,there will be no efficiency loss
in internalcontrol. As is proved in Proposition 1, this is indeed the case when the agent is not
liable for evasion. But when the agent is liable for evasion, we show in the next subsectionthat
therewill be efficiency loss in internalcontrol.
The manager is liable for evasion. Although tax evasion will not incur efficiency loss when the
manageris not liable, in general the manager,being the one who preparesthe tax return,10can
hardlydistance himself once it is detected. In that case the owner needs to compensatehim for
the risk involved in orderto induce him to participatein evasion.
Since the manageris risk averse and the owner risk neutral,the efficient allocationof risks
calls for the latterto bearall the risksfor the manager.If this can be achieved,thenthe managerwill
act as if he is not liable, andthe resultin Proposition1 is attained.At firstblush, it might seem that
this is indeedpossible. Forexample,the ownercan offer the managera contractthatis contingent
on whetherevasionis detected.Thatis, the manageris still paid accordingto {w(r); w(y) - w(r)}
if evasion is not detected. If, however,evasion is detected, then the manageris paid according
to {w(r); w(y) - w(r) + x(y - r)}. Since the manageris compensatedby exactly the amountof
penalty when evasion is detected,he is fully insuredagainstthe risks from evasion and will thus
act as if he is not liable for evasion.11The objective function and constraintof the managerare
thus exactly the same as in problemH, and the objective function of the owner differs from (3)
only by a termp f x(y - r)f(y)dy (the expectedvalue of the penaltyfor the agentthatshe needs
to pay), which has no consequenceon the solutionfor optimalwage w(y). In otherwords, w (y)
is still the contractthat induces the second-besteffort in the first stage, and there should be no
efficiency loss in internalcontrol.This argument,however,is not correct.
The reason for this is as follows. A contractthatis contingenton illegal activity will not be
honoredby the court.As such, a term of contractthat asks the principalto pay a certainamount
(in our case x(y - r)) to the managerafterthe illegal action of tax evasion is detectedwill not be
honored.Therefore,if evasion were to be detected, the principalwould certainlyhave incentive
to renege on the contract.Knowingthis, the managerwill insist thatthe compensationfor risk of
evasion be impoundedinto eitherthe wage contractor the service contractex ante. Consequently,
the pay for the managercannotbe contingenton whetherevasionis detectedor not. This createsan
incompletenessin the contract.We will show thatthis incompletenesspreventsefficient sharing
of risks, and results in distortionof the manager'seffort.12The argumentabove can actually
be put into a more general context. A contract,to be enforceable,cannot partitionthe states of
natureaccordingto differentoutcomes caused by an illegal action. This restrictionon an "illegal
contract"will create an incompletenessin the sense that the differentstates of naturethat result
from the illegal action can only be treatedidenticallyby a contract.In a word, a contractmust be
incomplete regardingdifferentstates that are caused by an illegal action. The argumenthere is
simply an applicationof this principlewhen the illegal action is tax evasion.13

10For
example, in the United States the corporateincome tax return(Form 1120) requiresthe signatureof the
officer.
l As in the case when the agent is nonliable, the agent can also extort more payment from the principal. See
Section 3.
12We are
gratefulto a refereefor suggesting this approachto us.
13 It can be
arguedthat one of the main reasons why violence is so prevalentin organizedcrime is that since
"contracts"regardingillegal actionscannotbe honoredby the courts,in orderto enforcean agreement,privateenforcement
in the form of violence must be used.
? RAND 2005.

This content downloaded from 14.139.211.75 on Mon, 08 Feb 2016 11:00:36 UTC
All use subject to JSTOR Terms and Conditions
158 / THERANDJOURNAL
OFECONOMICS

Define nr(y)to be such thatpu(w(y) + r(y) -x(y -r))+ (1 - p)u(w(y) + r(y)) = u(w(y)).
That is, nr(y) is the risk premiumnecessary for the managerto be indifferentbetween receiving
w(y) for sure and participatingin evasion, which is actually a lotterythat pays out w(y) + 7r(y)
with probability1 - p and w(y) + 7r(y) - x(y - r) with probabilityp. Since by (2) the amount
evaded, y - r, is fixed regardlessof realizationy, the penalty function x(y - r) is a constant.
The risk premium7r(y) thus dependsonly on the degree of risk aversionof the manager,but not
directly on y. The more risk averse the manager,the greaterthe value of Jr(y) needed to induce
him to take the risk.
Since the risk of evasion can only be compensatedex ante, there are two ways the principal
can compensatethe managerfor this risk. The first is throughthe original wage contractw(r).
In this case, the rewardfor bearing the risk of evasion is impounded into the original wage
contractso that the manageris fully compensatedfor the risk involved. The contractoffered to
the agent is thus {w(r) + 7r(y);w(y) - w(r)}. That is, the agent is paid a wage w(r) + Jr(y) if
the realizationof profitis y. The otheris throughthe service contract,so thatthe contractoffered
is {w(r); w(y) - w(r) + Jr(y)}. That is, the agent is paid w(y) - w(r) + 7r(y) for his service
in reportingr as profit. Again, the first scheme cannot work because once 7r(y) is paid upfront
throughthe wage contract,thenin the second stagethereis no moreincentivefor the managerto be
willing to reportr insteadof the trueprofity. Thus, all the rewardfor the manager'sparticipating
in evasion will be paid throughthe second scheme, i.e., throughthe service contract.In this case,
the expected utility of the managerwhen he evades tax is

f pu(w(y) + Jr(y)- x(y - r))f(y e)dy + (1- p)u(w(y) +n(y))f(y e)dy- v(e)

= f u(w((y))f e)dy - v(e),

where the equalityfollows from the definitionof 7r(y). The expected profitof the owner is

p [(1 - t)(y - w(y) - r(y)) - q(y - r)]f(y e)dy

+ (1 - p) y- w(y) - t(r - w(y) - r(y))]f(y e)dy

= (1- t) (y - w(y)- 7r(y)) f | e)dy+

)d+ t()(y p) r)(y e)dy

- p q(y - r)f(y | e)dy.

The maximizationproblemof the firmis thus

ProblemE.

max(l-t) f[y-w(y)-r(y)]f(y
w(y),e
| e)dy+t(1-p) (y-r)f(y I e)dy-p fq(y-r)(y (y e)dy

subjectto e E argmax u(w(y)) f (y I e')dy - v(e), (4)

u(w(y))f(y e)dy- v(e) > U.

The first-orderconditioncorrespondingto the IC constraint(4) is

1-t 1 + a7(y)= f
+ (5)
u'(w(y)) aw(y)
.J i f

? RAND 2005.

This content downloaded from 14.139.211.75 on Mon, 08 Feb 2016 11:00:36 UTC
All use subject to JSTOR Terms and Conditions
 CHENANDCHU / 159

Denote the solution to problemE by wE(y). It can be seen clearly that wE(y) = w1(y) if
and only if Trr(y)/aw(y)= 0, i.e., only when the need to compensatethe risk of evasion for the
managerin the second stage does not interferewith the wage policy in the first.In other words,
tax evasion will not incur efficiency loss in internalcontrol only if the tax evasion decision in
the second stage is independentof the provision of effort (throughthe incentive contract)in the
first stage. In general, however, a7r(y)/aw(y) is not zero. For example, if the managerexhibits
increasing risk aversion,14then he needs to be compensated by more for the risk of evasion
when he has more income (i.e., when w(y) is larger).As a result, 7r(y) increases in w(y) and
a7r(y)/aw(y) > 0. In this case, by (5) we know that wE(y) is less than w'(y). That is, he is paid
less under wE(y) than underthe second-bestcontract.Consequently,there is underprovisionof
effort on the partof the manager.The intuitionis clear: if the agent's aversionto risk increases
with income, then since the principalneeds to compensatehim more for the risk of evasion at
higherrealizationsof y, she is less willing to encouragehigheroutputthanunderthe second-best
contract.As a result, the agent is undermotivated.Similarly,if the managerexhibits decreasing
risk aversion(i.e., a'(y) < 0), then he is overcompensatedin wE (y) relativeto the second-best
contract.Finally, if the managerhas a constant degree of risk aversion, then 7r(y) is fixed and
a7r(y)/aw(y) = 0. Thereis thus no loss of efficiency in the contract.We thus have the following
proposition.
Proposition 2. If the manager is liable for tax evasion, then unless he exhibits constant risk
aversion,the firm will incurloss of efficiency in internalcontrol.Moreover,the manageris over
(under)compensated,relative to the second-best,when he exhibits decreasing(increasing)risk
aversion.
We have thus shown that there is a tradeoff between the gain from evasion and the loss
of internalcontrol efficiency incurredby evasion. It is importantto note that the efficiency loss
in question is not the amountthat the principalhas to compensatethe managerfor the risk of
evasion involved. It is the loss of efficiency in controlling him in addition to that amount of
compensation.Unlike the case of a single individual,a risk-neutralowner does not necessarily
evade tax when the tax evasion "lottery"yields positive expected profit,as she needs to bear the
additionalcost of efficiencyloss in control.This, of course,does not meanthatthe two-accounting-
book scheme mentioned above is impossible. In fact, companies in many developing countries
have been successful in using this scheme in evading tax. It only means that if the manageris
liable, then the firmcan scarcelymakeuse of the two-accounting-bookscheme withoutincurring
loss of efficiency in internalcontrol. In addition,whethera firm will evade tax depends on the
relative size of the expected profitsunderthe optimal solutions for problemsH and E.
A comparisonof resultsin the subsectionsabove thatdescribedthe cases when the manager
is and is not liable for evasion suggeststhatwhethertax evasion involves efficiency loss in control
solely depends on whetherthe decision of evasion can be separatedfrom that of control of the
agent. If so, then they areessentially two decision problemsthatcan be optimizedindependently,
andthereis no loss of controlas the firmevadestax. On the otherhand,if they cannotbe separated,
then the optimalreportingstrategywill interferewith the solutionof the principal-agentproblem,
and the agent's incentives will be distortedthereby.
Our model predicts the common sense result that if the cost of evasion is smaller-for
example, the audit probabilityp or the penalties of evasion, q(.) and x(-), are smaller-then
it is more likely that the firm will evade tax. More interestingis the case concerning internal
control. It implies thatmutualtrustbetween the principaland agent might be an importantfactor
in determininginternalcontrol efficiency as the firm evades tax. For example, if the manager
can trustthatthe principalwill (againsther short-runinterest)compensatehim for the amountof
penalty x(y - r) when evasion is detected, then he is fully compensatedfor evasion risk. In that
case, thereis efficient sharingof risk, and the second-besteffortcan be induced.This will in turn

14
That is, the degree of risk aversionof the manager,a(y) = -u"(y)/u'(y), is increasingin y.
? RAND 2005.

This content downloaded from 14.139.211.75 on Mon, 08 Feb 2016 11:00:36 UTC
All use subject to JSTOR Terms and Conditions
160 / THERANDJOURNAL
OFECONOMICS

predictthat tax evasion is more prevalentin smaller firms or family firms, in which interaction
is closer and trusttighter.Since less-developedeconomies tend to have more smallerand family
firms, our model also predicts that, other things being equal, tax evasion is more common in
less-developed economies.
Finally, one might wonder, although there is efficiency loss in control, whether this loss
is significantenough to deserve attentionat all. According to Jensen and Murphy(1990), each
$1,000 change in shareholderwealth correspondsto an annualincreaseof the CEO's salaryby
only 2 cents. This implies thatthe potentialloss to the firmcan be substantialeven if the manager's
compensationdeviates from the second-bestlevel by only a small amount.

3. Discussion
* In this section we considerseveralissues thatare not discussed in the main model, together
with the influences of these additionalconsiderationson our main result, and the qualifications
thatneed to be made for our resultsto hold againstthese considerations.

L Agent's extortion. In Section 2 we implicitly assumed that as long as the agent is fully
compensatedfor his risk,he is willing to cooperatewith the principalin evasion.This mightnot be
truein reality.The agentcan takeadvantageof his role in the evasion anddemanda portionof the
gain. In a word, he can extortthe owner duringthe evasion. This case is more difficultto analyze
because at presentthere is no standardtheory of extortion,so it is problematicto model how an
extortionpaymentis determined.Examplesof extortion,however,abound.For instance,whistle
blowers are generally treatedleniently in the legal process. As such, this gives the managerthe
powerto blackmailthe principal.To preventthe managerfromcarryingout the threatof informing
the tax authority,the principalhas to rewardhim with a compensationhigherthan wE(y) + 7r(y),
i.e., the wage contractoffered to the managerhas to satisfy the "no-whistle-blow"constraint.15
Given the lack of a standardmodel of extortion, we will abstractfrom the discussion of how
an extortionpayment is determinedand simply assume that z(y) is the amountof payment, in
additionto total wage w(y) and compensationfor the risk of evasion 7r(y),thatis receivedby the
agent. In this case the first-orderconditioncorrespondingto (5) is
-
1- t o(rr(y)+ z(y)) fe
u'(w(y)) a w(y) f
It can easily be seen that as long as the extortionpayment is independentof wage policy
in the first stage (so that az(y)/lw(y) = 0), then there will not be an additional efficiency
concern that enters into our model. In other words, as long as az(y)/aw(y) = 0, the distortion
mentionedin Section 2 is the only efficiency loss incurred.Generally,this might not be the case.
Forexample,we can imagine thatwhen the profitof the firmis high (so that w(y) is high as well),
the managermight believe thatthe firmcan affordto be extortedby a greateramount.In thatcase,
z(y) is higher as well. Consequently,az(y)/aw(y) > 0, and extortioncreates anothersource of
distortion.However,extortionitself does not necessarily worsen the distortionof the incentive
contract.From(6) we can see thatas long as Jr(y) and z(y) have oppositerelationswith respectto
changesin w(y) (i.e., a7r(y)/ w(y) and az(y)/aw(y) havedifferentsigns), thenextortionactually
amelioratesthe extentof the contract'sdistortion.Putdifferently,althoughthe principal'sexpected
gain of evasion is necessarilyreducedwith extortion,the efficiencyof the incentivecontractmight
actuallyimprove.The reasonfor this is quiteintuitive.Forexample,if the agentexhibitsincreasing
risk aversion,then our previousreasoningshows thathe is undermotivatedrelativeto the second-
best result. However, if the agent is able to extort more for higher realizationof y (i.e., if z(y)
increaseswith wage), thenthereis an additionalmotivationfor the agentto provideeffort,because
it correspondsto higher output and, thus, higher extortionrent. The problemof underprovision

15We thankJennifer
Reinganumfor suggesting this to us.
? RAND 2005.

This content downloaded from 14.139.211.75 on Mon, 08 Feb 2016 11:00:36 UTC
All use subject to JSTOR Terms and Conditions
 CHENANDCHU / 161

of effort is thus assuaged in the presence of extortion.Naturally,if nr(y) and z(y) have opposite
relationswith respectto changesin w(y), thenthe abilityof the agentto extortcan only exacerbate
the problemof effort distortion.

[ Repeated interaction. It mightbe arguedthatsince a firmis a long-livedentity,theprincipal

and agent should have a long-termrelationship.If this is so, then a contract(or collusion) that is
not legally enforceablecan still be enforced as an implicit contractby way of reputation.This is
actually an applicationof the folk theoremin repeatedgames (see, for example, Fudenbergand
Maskin, 1986). By this argument,one party who has been cheated can retaliateby refusing to
cooperatewith the deviant in the future.If the loss caused by retaliationis large enough, and if
the parties do not discount the futureheavily, then the threatof retaliationcan deter them from
deviationagainstimplicitly agreedactions.
In our context, this implicit contractmight work as follows. The principalpays the agent
accordingto wl(y), and the agent declaresr as profit.In the case when evasion is detected, the
principalpromises to compensatethe agent the full amountof penalty,x(y - r). The manager,
if betrayed,can refuse to collude with the principalin the future.If the owner does not discount
futureincomes by much,thennoncooperativeoutcomein the futureis enoughto deterthe principal
from cheating.This argumenthas some merit,and perhapsworks in some environments,but the
folk theorem, and thus the validity of this argument,actually holds only under very stringent
assumptions.Above all, it requiresthatthe principaland agent are boundto interactindefinitely,
otherwisethe end-periodargumentwill unravelto destroyall possibility of successful collusion.
In our context, once tax evasion is detected and both are penalized, it is very likely that the
relationsbetween the managerand the principalwill be severed,16i.e., they will not have a long-
termrelationship(as is requiredby the assumptionof the folk theorem)afterevasion is detected.
Consequently,the abilityof the managerto "retaliate"againstthe principalwhen the latterreneges
is in doubt.

D Other sources of inefficiency. In our model, the source of inefficiency in tax evasion
comes from the fact that the contractis deemed incomplete due to its illegal natureand thus is
difficultto enforce.This incompletenessin turninducesdistortionon the agent'seffort.Thereare,
however,potentiallymanyotherreasonswhy corporateincome tax evasioncan createinefficiency
in internalcontrol.Herewe discuss two of the reasonsthatwe believe to be of the greatestinterest.

Informationalvalue of signals and detection probability.The basic tenet of contracttheory is

that in orderto make the contractbetween the owner and manageroperationaland enforceable,
the signals on which the wage contractis based must be verifiable(see Hart, 1995). According
to Holmstrom (1979), the more firm-relatedvariables that are of informationalvalue for the
agent's effort level are written into the contract,the better the principalcan control the agent,
and thereforethe more efficient is the contract.But by doing so, more verifiableinformationon
the internalgovernanceof the firm will be made public, makingit easier for the tax authorityto
detect tax evasion. If the probabilityof being detected increases significantlywhen the contract
is more detailed and open to the public, then the principalwill be unwilling to incorporateas
much informationas the optimum would requirewhen she evades tax.17We can thus have a
model in which the degree of contractcompletenessis a choice variableof the principal.A more
complete contractenhances efficiency in internalcontrol, but it also increases the probability
thattax evasion will be detected.To evade tax, the owner chooses the optimaldegree of contract
completeness to balance the tradeoffbetween the two considerations.
More specifically, suppose Yi .... Ynare n profit-related(verifiable) signals that are of
informationalvalue in inferringthe effort level of the agent. If the firm does not evade tax, then

16For example, the managermay be forced to resign.

17For
example,the knowledgethatthe hiringandpayingof black-marketlaborcannotbe basedon writtencontracts
makes one more cautious in hiring such laborthanwhat the optimumwould otherwisecall for.
? RAND 2005.

This content downloaded from 14.139.211.75 on Mon, 08 Feb 2016 11:00:36 UTC
All use subject to JSTOR Terms and Conditions
 OFECONOMICS
162 / THERANDJOURNAL

by Holmstrom(1979) the wage contractshouldbe a functionof all the yi's. If it evades tax, then
it must manipulatethe values of (at least a subsetof) the yi's thatarereportedto the tax authority.
However,this will increasethe probabilitythatevasionis detectedif these yi 's arealso includedin
the laborcontract.Supposethe expectedgain of evasion in manipulatingthe value of yi is Xi, and
the probabilityof being detectedwhen Yiis (is not) used in the wage contractas a determinantof
wage is pi + APi (Pi). The efficiency loss (in monetaryterms)when Yiis not incorporatedin the
wage contractis assumedto be ci. The expectedgain of evasionby manipulatingthe value of yi is
thus Mi - - piq - ci if Yiis not writteninto the contract,and Ni - Xi - (pi + Api)q if it is.
Then if max{Mi, Ni } < 0, the firmwill not use Yias an instrumentto evade.If max{Mi, Ni } > 0
and Mi > Ni, then yi will be used as an instrumentto evade and it is not writteninto the contract.
If max{Mi, Ni} > 0 and Mi < Ni, then Yi is sufficientlyimportantfor internalcontrol despite
the fact that it increases detection probabilitywhen it is writteninto the contract.Consequently,
yi will be used as an instrumentto evade and it is also writteninto the contract.The firm then
does the same exercise for all the yi's and decides whetherto evade tax and,if so, which subsetof
{yi }=l to use as instrumentsto evade. Finally,it can also decide whetherthose yi's thatare used
to evade tax should be writteninto the contract.We thus not only have a model in which the firm
can choose which signal(s) to use as instrument(s)to evade, but also in which the completeness
of contractis endogenously chosen to balance the tradeoff between detection probabilityand
internalcontrol.
Double use of control instruments:Anotherpossible loss of efficiency,when the firmevades tax,
is that the same instrumentused to control the manageris also used as an instrumentto evade
tax. Let us consider the following example, which reflects the real situationin many countries.
Suppose there is a restaurant,the profit of which is sales revenue minus wages and the cost of
inputs.The wage of the restaurant'smanageris based on revenuesandcosts, which areimperfect
signals for his effort level. To evade taxes, the restaurantowner can either ask the managernot
to issue receipts to customers (especially cash-transactioncustomers) so that the sales record
can be suppressed,or ask the managerto exaggeratethe amountof inputs so that costs can be
overstated.Supposethatthe manageris askedto evade taxes by suppressingsales records.In that
case he might be temptedto treathis friends or family in the restaurantwithoutpaying for it, or
he might simply put partof the revenue(for which he did not issue receipts)into his own pocket.
As a result, it will be hard for the owner to evaluate the performanceof the managervia the
informationrevealedby revenue.In otherwords, thereis an efficiency loss of internalcontrolfor
the restaurant.The case is the same for overreportingcosts. If the manageris askedto overreport
costs in orderto evade taxes, then the managerwill be proneto waste and to increasethe costs of
the firm(perhapsfor his own perquisites),and the informationvalue of costs as an instrumentto
controlthe manageris partlylost.18
More specifically,suppose thatin orderto evade tax, the manageris askedto reporta lower
value of outputy. Given this discretion,he might be able to divert some outputfor his own use
and claim lower output.The informationvalue of outputas a signal of effort level will therefore
be reduced.That is, the informationon the manager'seffort level providedthroughthe function
f(y I e) will then be noisier when the manageris asked to help evade tax by manipulatingthe
value of y. Therecan be many ways to model this. For example,the conditional-densityfunction
of outputcan be assumedto be a mean-preservingspreadof the original-densityfunctionf(y I e)
when the firmevadestax. Since thereis informationloss in the signal, the efficiencyof the optimal
contractwill be reduced.

4. Conclusions
* This articleproposes what we believe to be the first theoreticalmodel of corporateincome
tax evasion. We explore the link between internalcontrol and the externalevasion decisions of

18Note thatthis does not

necessarilymean thatthe managercolludes with the owner to evade taxes. He might not
be in chargeof tax matters,and is only asked not to issue receiptsfor some consumers.
? RAND 2005.

This content downloaded from 14.139.211.75 on Mon, 08 Feb 2016 11:00:36 UTC
All use subject to JSTOR Terms and Conditions
 CHENANDCHU / 163

businesses. Since the attemptof the owner to evade tax requiresthe collusion of the manager,it
bringsadditionalrisk to the managerif he is liable for evasion. The fact thattax evasion is illegal
preventsthe possibility of compensatingfor this risk ex post, meaningthat the managermust be
compensatedthroughthe wage contractex ante. This not only creates an incompletenessin the
contract,but also forces the laborcontractto play the double role of rewardingthe agent's effort
and compensatingfor the risk of evasion. Consequently,it incursefficiency loss in controllingthe
manager'seffort.
We also exploreotherpossible sourcesof inefficiencyin tax evasion.The firstis thatin order
to cheat on tax, the businessownerneeds to createvague informationto misleadthe tax authority.
As long as this informationis also needed for internalcontrol, such vagueness will reduce the
informationalvalue of the contractin controlling the manager.The second possible source of
efficiency loss is thatin orderto evade tax, the firmneeds to eitheroverreportcost or underreport
revenue. But once it delegates the power of manipulatingthe value of cost or revenue to the
manager,the latter can abuse this power for his own benefit. Consequently,the informational
value of cost or revenueas an instrumentof controlis reduced.Although we have not modelled
the two resourcesof inefficiencyformally,the intuitionis clear.
The idea thatif a contractinvolves actionsthatareillegal (thusit is deemedto be incomplete,
which in turnincurs efficiency loss) appearsto us much more generalthan in the context of tax
evasion. In particular,this might help to explain why violence is prevalentin organizedcrime, as
it uses violence as a way to enforce agreementsthatare otherwisenot honoredby the courts.
Finally, since corporateincome tax evasion involves the cooperationof membersin a hier-
archy, we have reason to believe that a deeper understandingof the recent developmentin the
theory of collusion in hierarchiescan greatlyenrichthe study of corporateincome tax evasion.19

References
ALLINGHAM, M.G. AND SANDMO,A. "IncomeTax Evasion:A TheoreticalAnalysis."Journalof Public Economics,Vol.
1(1972), pp. 323-338.
ANDREONI,J.R., ERARD,B., AND FEINSTEIN,J. "TaxCompliance."Journal of Economic Literature,Vol. 36 (1998), pp.
818-860.
CHU,C.Y.C."PleaBargainingwith the IRS."Journalof Public Economics,Vol. 41 (1990), pp. 319-333.
COWELL, F.A. "Taxationand LabourSupply with Risky Activities."Economica, Vol. 48 (1981), pp. 365-379.
.Cheating the Government:The Economicsof Evasion. Cambridge,Mass.: MIT Press, 1990.
CREMER, H. ANDGAHVARI, F. "TaxEvasion and the OptimumGeneralIncome Tax."Journalof Public Economics,Vol.
60 (1996), pp. 235-249.
DUBIN,J.A., GRAETZ, M.J., ANDWILDE,L.L. "StateIncome Tax Amnesties:Causes."QuarterlyJournalof Economics,
Vol. 107 (1992), pp. 1057-1070.
FELLI,L. ANDVILLAS-BOAS, J.M."RenegotiationandCollusionin Organizations"JournalofEconomicsandManagement
Strategy, Vol. 9 (2000), pp. 453483.
FUDENBERG, D. ANDMASKIN, E. "TheFolk Theoremin RepeatedGames with Discountingor with IncompleteInforma-
tion":Econometrica, Vol. 54 (1986), pp. 533-554.
HART,O. Firms, Contracts,and Financial Structure.New York:OxfordUniversityPress, 1995.
HOLMSTROM, B. "MoralHazardand Observability"Bell Journalof Economics,Vol. 10 (1979), pp. 74-91.
JENSEN, M. ANDMURPHY, K.J. "PerformancePay and Top ManagementIncentives."Journalof Political Economy,Vol.
98 (1990), pp. 225-264.
KOFMAN, E ANDLAWARREE, J. "Collusionin HierarchicalAgency."Econometrica,(1993), pp. 629-656.
KREUTZER, D. ANDLEE,D.R. "On Taxationand UnderstatedMonopoly Profits."National TaxJournal, Vol. 39 (1986),
pp.241-243.
MOOKHERJEE, D. ANDPNG,I.P.L. "Monitoringvis-a-vis Investigationin Enforcementof Law."American Economic
Review,Vol. 82 (1992), pp. 556-565.
SANDMO, A. "IncomeTax Evasion, LabourSupply,and the Equity-EquityTradeoff."Journal of Public Economics, Vol.
16 (1981), pp. 265-288.
SRINIVASAN, T.N. "TaxEvasion:A Model."Journalof Public Economics,Vol. 2 (1973), pp. 339-346.
TIROLE, J. "Hierarchiesand Bureaucracies:On the Role of Collusion in Organizations."Journalof Law,Economicsand
Organization,Vol. 2 (1986), pp. 181-214.

19The
investigationof collusion in hierarchiesis still in its preliminarystage. See Tirole(1986) for a seminalstudy,
Tirole (1992) for survey,andKofmanand Lawarree(1993) andFelli and Villas-Boas (2000) for recenttheoreticalresults.
? RAND 2005.

This content downloaded from 14.139.211.75 on Mon, 08 Feb 2016 11:00:36 UTC
All use subject to JSTOR Terms and Conditions
164 / THE RAND JOURNALOF ECONOMICS

. "Collusion
andtheTheoryof Organizations."
In J.-J.Laffont,ed.,Advancesin EconomicTheory,SixthWorld
Congress,Vol.2. NewYork:Cambridge Press,1992.
University
WANG,L.F.S. AND CONANT,J. "CorporateTax Evasion and OutputDecisions of the UncertainMonopolist" National
TaxJournal,Vol. 41 (1988), pp. 579-581.

? RAND 2005.

This content downloaded from 14.139.211.75 on Mon, 08 Feb 2016 11:00:36 UTC
All use subject to JSTOR Terms and Conditions