The Optimal Size of a Tax Collection Agency Author(s): Joel Slemrod and Shlomo Yitzhaki Source: The Scandinavian

Journal of Economics, Vol. 89, No. 2 (Jun., 1987), pp. 183-192 Published by: Blackwell Publishing on behalf of The Scandinavian Journal of Economics Stable URL: http://www.jstor.org/stable/3440063 Accessed: 02/12/2010 05:07
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=black. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. 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.

Blackwell Publishing and The Scandinavian Journal of Economics are collaborating with JSTOR to digitize, preserve and extend access to The Scandinavian Journal of Economics.

http://www.jstor.org

This researchwas completedwhile the first authorwas a NationalFellow at the Hoover Institution.Severalalternative tax stratehavebeen suggested deal withthis problem. subcommittee for every that recentlytestifiedbefore a Congressional additionaldollarallocatedto his agency'sbudget.increasing for punishment evasion.Minneapolis. J. I.The increasedrevenuegained from stricter enforcementdoes not enter the expressionbecause it merelyrepresentsa transferamong the economy'sidenticalagents. and Scand. Introduction In recentyearstherehas been considerable concernaboutthe erosion of tax revenuesdue to increased evasion. The appropriatecondition is that. One proponent of the third strategy.including to gies reducing tax the marginal ratesso as to reducethe incentiveto evade.1987 The OptimalSize of a TaxCollectionAgency* Joel Slemrod MN and NBER.the Commissionerof the US Internal Revenue Service. of Economics 89(2).the resource cost of increasingenforcementshould equal the saving of excess burdendue to the decline in exposureto risk. 183-192.' Although the Commissioner *The authorswould like to thankthe US-IsraelBinationalScience Foundation financial for support. Helpfulcommentson an earlierdraftwere received fromJoramMayshar AgnarSandmo. Israel Jerusalem.and expandingthe resourcesof the Internal RevenueServicein order to increasethe likelihoodof detectingand prosecutinginstances of tax evasion. Shlomo Yitzhaki HebrewUniversity.The paper clarifiesan earlier findingof Sandmowhich suggeststhat the optimaldegree of enforcementis higherthan a simple cost-benefit calculationwould indicate.Scand. at the margin.it could returnten dollars in additionaltax collections. MA. J. USA Universityof Minnesota. Abstract The conditionthat characterizes optimalsize of a tax collection agencyis derivedand the in then given a simple interpretation terms of excess burden. of Economics 1987 . Cambridge. StanfordUniversity.

Postal Service and General Government Appropriation. For example. as any other government agency. Their implicit condition for the optimal degree of enforcement is apparently that marginaltax revenue be equal to the marginal cost of collection. that is. 433). Jr. Yitzhaki undoubtedly interpreted these figures as evidence for an expanded IRS budget. where excess burden is defined as the dollar value of the difference in the expected utility between the case of paying taxes with certaintyand the alternativeof paying taxes as a lottery. taxes. not real output. the marginalresource cost of enforcement should equal the reduction in the excess burden of tax evasion. Slemrod S. of Economics 1987 . and those that have dealt with the issue do not speak with one voice. even governments starved for funds refuse to provide enough to collect taxes under existing law. Shoup (1969) notes that although "it might at first appear that money should be appropriated for tax administrationuntil the incremental dollar of cost yielded just one dollar's increase in revenue ..."(p. before the House Subcommittee on the Treasury. 283). in their textbook. J. 217). who apparently espoused the view that the optimal size of the tax collection agency is even larger than simple cost-benefit analysis would indicate. Scand.The tax collection agency. February 27. See the testimony of Comissioner Roscoe L. At the optimum level of enforcement... 1984." (p. but its value is not necessarily to be reckoned in terms of the dollars taken in . That implementationis a valuable commodity.. in the absence of further analysis it is not clear whether it is desirable to expand the tax collection agency in this circumstance. Egger. "marginal The principal goal of our paper is to clarify this issue by deriving the condition that characterizes the optimal size of a tax collection agency in a simple model and then providing a simple interpretation of this condition.and 184 J. and that at the optimum cost [should be] higher than marginaltax revenue. and no further... and its size should be expanded to the point where its marginal cost equals its properly defined marginalsocial benefit. .."(p. Schultz and Harriss (1959) lamented the fact that ". Public finance economists have not paid much attention to the issue of the optimal degree of enforcement of the tax law.. However. although ample evidence indicates the additional funds spent on administrationwould yield many times their amount in added collections. The first explicit analytical treatment of this issue is in the seminal paper by Sandmo (1981). this would be to overlook the fact that in tax administration real resources are being used up to implement a system of transferpayments. uses real resources to operate.

of Economics 1987 . Note that expected tax payments equal twL*-(1 -pF)g* where F= 1 + r.As modeled by Allingham and Sandmo (1972). (3) where g* and L* stand for the optimum level of tax understatement and labor supply. respectively. L*). Optimal Enforcement Policy The problem of tax evasion is inherently one of choice under uncertainty.g. L).Note that it is assumed that L must be chosen before the state of the world (audited or not audited) is known. g the amount of tax understatement. a representative risk-averse household faces a given probabilitythat an understatement of true tax liabilitywill be detected and punished. assessed as a fraction of the evaded tax liability.g. The government chooses p and t and is constrained to raise a given amount Scand. We offer some concluding comments about promising directions for future research in Section IV. In Section III we compare our results to those obtained by Sandmo (1981). It is convenient.Optimalsize of tax collection agency 185 The paper is arrangedas follows. L labor supply. L). Now we consider the problem faced by the government. to restate expected utility in terms of "true"after-tax income (c). A simple model of optimal government enforcement policy is presented in Section II along with an interpretation in terms of the marginal excess burden of tax evasion.g .L)+pU(wL-twL L. defined as (1l-t)wL. and reconcile the apparentlyconflicting results obtained in the two papers. following Christiansen (1980). (1) where w is the wage rate. A simple formulation of the household's optimization problem is Max E(U)=(1 -p)U( wL-(twL-g). J. t the statutorytax rate. II. The household maximizes expected utility by balancing at the marginthe expected utility of an undetected evasion of tax liability with the expected utility of the penalty for evasion. and J the rate of penalty for tax evasion. (2) The first-ordercondition with respect to g is (1-p) U(c*+g*. Equation (1) then becomes Max E(U)=(1 -p)U(c+g.g L)+pU(c . L.p the probabilitythat the evasion will be detected and penalized. L*)=prUi(c*g*.

are * * * U(c +g L*) . In general.(c *-Tg* L*)] =r(Rla/t).Models which include this dimensionof choice inevitablyconclude that an optimalpolicy featuresa large penaltyand a small probabilityof conviction such that no offenses are committed.s. is clearly positive. Thus one conclusion we can drawnfrom (5) is that at a maximum a R/ p > a'(p). in choosing p and t. or by increasing p the tax collection agency could increase revenue by more than the cost of doing so. becausefines are transferswith no resource cost. An excellent discussion of this issue is contained in Stern (1978).t subject to twL*. Yitzhaki G of revenue to spend on. (6) where R is the sum of expected revenues and penalties collected and r equals.2 The probability that the tax collection agency can detect an act of evasion depends on the amount of resources devoted to it. Thus. p. a higher p causes expected revenue to be paid as a less risky lottery than otherwise. Consider equation (5).(1 -pF)g* = G + a(p). we denote its cost function as a(p). L*) +pU(c*p.s.This kindof consideration outto lies side of our modeling effort. public goods. Scand. the objective of the government is to maximize the expected utility of the representativehousehold. J. while maintaining tax collection agency does entail a resource costs. where society is not indifferentto pairs of t and p that raise the same expected revenue.h. The l.U(c* . of Economics 1987 . We assume that. See SectionIII of this paperfor a reconciliation Sandmo'sfindingsand the resultspresented of here.Jrg L*) = r[(aRl/a) .t and p. g*. The expression in brackets on the r. say. the government'sproblem can be stated as Max W=(1 -p) U(c*+g*.a'(p)] (5) wL*[(1-p) U1(c*+g* L*) +pU. 278. respectively.h. 3Note that Sandmo (1981) comes to the same conclusionin his expression(48). and it can be shown that r is also positive. the marginal expected utility of relaxing the government revenue requirementby one dollar.3 2Note that we also assume that the penaltyfor tax evasion is not an instrument subjectto choice by the government. The first-order conditions of this problem with respect to p and t. Slemrodand S. What limits the penalty for tax evasion is a presumptionthat the penalty shouldbe commensurate the seriousnessof the crime. must then also be positive. althoughit entails the expenditureof real resources.186 J.This strategyimpliesa zero social loss. L*) (4) The government'sproblem then is to choose between two revenueraising instruments. at a maximum.

Following Yitzhaki (1987). In this model the excess burden of evasion.g * with probability (1 . it is not meaningfulto talk of the excess burden 4 See Yitzhaki (1979) for a similar interpretation of the optimal coverage of excise taxation. Expression (6) implies that at a maximum a R/l t is positive.. to be the difference in expected utility between the case of paying the revenue requirement with certainty and the alternative of paying it with uncertainty. or else no taxes would be paid).p) and T+ :rg * with probabilityp. In the tax-farmer case. The total social cost of evasion is conveniently thought of as the sum of the excess burden of tax evasion. i. (7) which is greater than zero as long as g* is positive (which it will be if evasion is better than a fair gamble. EB. which is the effective contingent tax schedule when evasion is considered. ratherthan as a certainty. aR/l p would equal a'(p) and lead to a socially excessive amount of resources devoted to tax collection. J. a(p). In the absence of any other uncertainty. the representativehousehold is clearly worse off in the end.e.pF) is positive). Scand. each individual believes himself to be better off by engaging in evasion.This contrasts to the classic case of the tax farmer interested only in maximizing profits. is simply the utility loss due to the fact of paying taxes as a lottery. Consider first the case where the tax is lump sum with value T.or EB= U(c+(1 -pF)g*)-[( -p) U(c+g*)+pU(c-jrg*)]. as defined above. that the government should be on the increasingpart of the "Laffercurve". Furtherinsight into the interpretationof equation (5) can be gained by defining what we refer to as the excess burden of tax evasion. in utility terms.4 When there is a distortionary labor income tax in addition to the presence of tax evasion. of Economics 1987 . T-( 1 -pF )g*. we define the excess burden of tax evasion. However. Then the social optimization problem can be interpreted as minimizing the social cost of evasion. Of course.Optimalsize agency 187 oftax collection This follows immediately from the fact that increasing p decreases the representativeindividual'swelfare.it is always the case that the representative risk-aversehousehold would prefer to pay its expected tax liability. ceterisparibus.Note that this measure of excess burden is independent of the cost of administering the tax system (which is presumably positive in this model. with certaintycompared to the alternativeof paying T. because of the tax rates will be adjusted to raise the same amount of expected revenue. or if (1 . and the administrative cost of limiting the amount of evasion.

of Economics 1987 .Because the termin parentheses mustbe less thanone for evasionto be betterthana fairbet. or thatwelfaredeclinesas p is increased. It is. The increased revenue to be gained from increasing the enforcement of the law does not enter the expression because this is merely a transfer among the economy's identical agents. This solution.He showsthat .p= 1/F). Note that at marginalreductionin excess burdenequals the marginaladministrative the optimump.k. he concludesthata W/ap is less thanzero. however. at the margin. the optimal p would not be so high as to completely discourageevasion.5 5Cowell (1985) arguesin a model with fixed labor supplythat.a W/ap strictlypositive.Whathe fails to recognizeis that for his utilityfunction W= -e. at the point where a cost. Thus. This is true even though the first-ordercondition (5) may be satisfied at p= 1/F. Yitzhaki of either tax evasion or labor income taxation separately. In this case the inducedlaborsupplyfroma reductionin p outweighsthe disutilityof the riskiness of tax payment.but only the total excess burden of the system as a whole.It is that. in the appendixis incorrect. the marginal savingin excess burdenfrom a furtherincreasein p is positive. consider an experiment where the government increases p but returnsthe additional revenue thus gained by adjustingt enough so that total net revenue (includingboth taxes and penalties) is unchanged.his derivation it. even thoughit maybe negativefor highervaluesof p. meaningful to talk of the marginalexcess burden from changinga particularinstrument of the government.so that although(1/W) ( W/lp) is is negative. w ap (t -pt) where Wis social welfareand 0 is the penaltyrate for evasion. In the presence of costly enforcement.the optimalp is even lower. Slemrodand S. there is a simple rule for the optimal size of the tax collection agency.. though.the cost of increasingp should equal the saving of excess burden due to the decline in the exposure to risk.excess burdenactuallyincreaseswithp.even if enforcementof the tax laws was costless.e. thusimplying the optithat malvalueof p is zero if thereis any resourcecost to increasing However. In order to derive the marginal excess burden with respect to a change in p.(d(EB)/dp) = ra'(p). J. Wis alwaysnegative. In the Appendix we show that this expression allows us to express (5) simply as (8) where -(d(EB)/dp) is the reduction in the excess burden of tax evasion due to a marginalincrease in p. Scand.188 J.for a constantabsoluterisk aversionutilityfunction. Weiss(1976) and Stiglitz(1982) suggestthatthe excess burdenmightbe increasing the in neighborhoodof a value of p just high enough so that there is no evasion (i.This impliesthat. correspondsto a local minimum.

b +weE. marginal revenueis calculated of The tax net the moneynecessaryto keep the evaderat the samelevel of expected utility. by This can be demonstratedby first forming Ve/Ve from Sandmo's(22) and (24). is apparat odds withthe conclusiondrawnby Sandmo(1981) in the conently textof an essentially similar. but not exactly. of Economics 1987 . marginal the size of the taxcollectionagencyexceedsthe marginal cost.and is equal to the increasedexpected penaltypaid.The firstdifferginaltax revenueandthe usualuncompensated ence is the usual purgingof the income effects on behaviorof the for changein p. holdingE and Le constant.a payment of somewhat less than -b+ 0wE. is required. Le labor supply reported to the tax authorities.The reductionin the requiredpaymentis approximated (U.that"themarginal additionalevadershould at the optimalbe equal to the marginal tax the compensated a term revenue.E unreportedlabor supply. 6 the penalty rate for evasion.at the optimum.The term . These claims are justifiedby a conditionof the optimum whichsays.computedalong the compensatedsupply curves"is then twe[(8Le/ where We is the wage ap) -(OLe/ap)] lb comp +powe[(aE/p) lb-comp -(aE/ap)] .payablein the "detected" state. is higher than a simple cost-benefitanalysiswould seem to indicate"." Sandmo'susage of this term is not obvious and could lend itselfto misinterpretation. The apparent conflictbetweenhis resultand ours however.and hinges on the precise meaningof the term tax revenue.is only semanticin nature. The terms in brackets reflect the difference in behavioural responsedue to incomeeffects. He claimsthat "marginal shouldbe higherthan marginaltax revenue".in Sandmo's cost of catchingan words.. (p.b+ OweE is the amountof compensation paidin the "detected" stateneeded to keep the expectedincomeof an evaderconstant.Thus. in Sandmo'smodel and notation. and b a paymentfrom the governmentto the individualpayable only in the detected state.. Reconciliationwith Sandmo the revenuefromincreasing The findingthat.b + OweE +pOwe( E/lp).6 6To see this more clearly.Optimalsize of tax collection agency 189 III. Because the evaderis riskaverse. The secondis the nettingout of the fundsrequired the compensation.note that dR/dp is.equalto the amountof funds. computed along the compensated supply "marginal curves.needed to fully compensatean evaderfor a unit increasein p." 283). which is Scand.computedalong supplycurves. received in the state with higher marginalutility.modelthanthe one cost presentedhere./2 U)(C -C e)2.equal to twe(aLe/ap) . there are two differencesbetweenthis concept of marnotion. rate. J. thoughmorecomplex. The difference between this and what he refers to as tax "marginal revenue.plus the of reflecting curvature the utilityfunctionor the degreeof riskaversion.It is also approximately..and that "the optimal probabilityof detection .

one with the opportunity to evade and the other without this opportunity. penaltyfor [(U(C . 2U.1 L" + E). Conclusion The rulefor the optimalsize of a tax collectionagencyderivedhere is based on simplemodel of the individual's decision problemand the Beforethisrulecanbe useddirectly a guide as government's objectives.(C. Theoptimization problemshouldalso be expandedto includemany with to typesof individuals. to policy. Yitzhaki and Sandmo'sconcept of marginalrevenue is thus cruciallydifferent thanthe standard that of usage. Le+E). tax subthe to tracting fundsthe taxcollectionagencymustdisburse ensurethat the evadingpopulationis no worse off when the probability audit of increases. assumption the that the probability detectionis fixed must be generalized. L"+E) .A numericalexamplepresentedin Slemrodand Yitzhaki (1985) shows clearly that conclusions based on an inappropriate notion or marginalrevenue can lead to large errorsin the optimal settingof p..Increased enforcementcould conceivablyincreasethe degree to which individualstryto avoiddetection.andhis statement the probability detection oughtto be set higherthana simplecost-benefit calculation would indicatemust be interpreted with great caution.This featurewould allow study of the horizontaland vertical equityaspectsof taxevasion. Slemrod S. Scand.Alternatively.U( C.190 J. V..How a recognition these additional of costs wouldaffectthe optimal size of the tax collectionagencyis not clear.= (C. the of In of and of probability detectiondependson the character magnitide the the evasion.bothaspectsof the problem mustbe expanded. Le+E)]/TE.This "simple" costbenefitcalculation must measuremarginal revenuecarefully. different opportunities evadeanddifferent tastes. 2 u.U(C'. Then substitute pU. IV. U(C'.(C.however. of Economics 1987 . L+E) 7 Sandmo 1981 considers two ( ) groups of individuals. Le + E) (Ce _ Ce)2 . These resourcesare anothersocialcost of evasion. Finally.(C. Withrespectto the individual decisionproblem.2' + Ow"E) 1 U. J. fact.7 the Varying size of the tax collectionagencyis not the only possible the responseto the problemof tax evasion. L"+E) (2b+ Le+E) for Tr from (14). using the Taylor expansion following (59).and on the resourcesexpendedby the evaderin reducing likelihoodof detection. we obtain v.

w(l -t).s.=a V/at. F. which has evasion. =R/t. w(1 -t). w(l -t) is the after-tax wage rate.h. F. w. and which raises the same revenue is given by V(0. We assume that evasion is prevented by an effective and costless enforcement structure. where twL*-(1 -pF)g*=G and pF < 1. Thus. F. if negative.t).t). and p is the probability of detection.V. ==R/ap. where Z represents lump-sum income (or. respectively.V(O. w.Vpand .RP/R. One potentially valuable direction for research is to develop an integrated approach to the optimal government policy toward tax evasion. P)|IR=. J.G. Appendix. Then the excess burden is defined as EB= V( . . F. the first-order condiScand.Optimalsize of tax collection agency 191 tax evasion may be increased. p) be the indirect expected utility function. so as to distinguish the excess burden of evasion from the total social cost of evasion.and V =a V/ap. P)/PI =G w(l Let R.G. the first term of (Al) is constant with respect to either or t. of equations (5) and (6) in the text are . Derivationof the marginalexcess burden We define the excess burden of a tax system which raises a given amount of revenue to be the increase in utility that could be achieved if the same revenue could be raised using lump-sum taxes and in the absence of evasion. Another possible response is to change the tax law itself so that the opportunities and incentives are decreased. - RpR. F is the rate of penalty on detected evasion. p'). (A3) and d(EB)/dp = -[ V-V. p') where p'F' > 1. of Economics 1987 . Hence p d( EB)/dp= -a V(O. For a given G. Let V(Z. The utility level achieved by an efficient tax system is V( . The utilitylevel achieved by a tax system which must use income taxation. (A4) It is straightforwardto show that the l. p). and G is the required revenue. F'. Rp Then (a tap)lR == (A2) V. a lump-sum tax). w( . (Al) where R is tax revenue. F'.

June 1980.C.: Incometax evasion. Gaertner (eds. Heineke(ed. Cowell. tradeoff. at the optimum.: Public policy and tax evasion:Some problems.A.C. Clotfelter.1969.Prentice-Hall. Review of Economics and Statistics LXV. Co.: A note on optimum taxationand administrative Review LXIX. Slemrodand S.: Utilitarianism horizontalequity:The case for randomtaxation. 1759.W.J.Working PaperNo. 201-2.N. of Economics1987 . 265-88. laboursupply. References Allingham. Yitzhaki.In J. November1985. the saving in excess burden due to an increase in p. The Economics of the Shadow Economy.).p-a'(p)] and Vt= . & Sandmo.In A. should be equal to the marginalresource cost of increasing p. Wenigand W. FirstversionsubmittedNovember1985. S.: Income tax evation:A theoreticalanalysis.1343-52. (A6) (A5) Some substitutionand rearrangementof (A4). Journalof PoliticalEconomyLXXXIV. June 1982.192 J.: AmericanPublicFinance. and (A6) yields (A7) This condition says that. V. Slemrod.S. Stern. 1959. S. XV. Yitzhaki. 1985.NationalBureauof EconomicResearch.:On the economictheoryof policy towardscrime.). June 1979.S. (A5).& Harriss. L. Two commentson tax evasion.L. F. 389-93.Journal of Public Economics III. M. and Stiglitz.:PublicFinance. 363-73.J. J. -[dEB/dp] = a'(p). valued in dollars.1978.Journalof Public Economics XI.: Tax evasion and tax rates: An analysis of individualreturns. Scand. 323-38.. EnglewoodCliffs. 1-33. finalversionreceivedJanuary1987. Schultz. Yitzhaki tions for optimal p and t in the presence of administrativecost can be written as Vp= -rR.Economic Modelsof Criminal North-Holland Behavior.A. Co.and the equity-efficiency of Public Economics XVI.:The optimalsize of a tax collection agency. 475-80. February 1984. December 1981.C. May 1974. December 1976.. costs.Journal Sandmo..A.: A note on "income tax evasion: a theoretical analysis". & Yitzhaki. of in Weiss.Aldine Publishing Chicago.. Shoup.Journalof PublicEconomicsXII.N. Publishing Amsterdam.PublicFinanceQuarterly April 1987. AmericanEconomic Yitzhaki.J.R.Journalof Public EconomicsI.:On the excess burdenof tax evasion.: Christiansen. or that the marginalsocial cost with respect to p is zero. November 1972.:The desirability cheatingincentivesand randomness the optimalincometax. Berlin. Springer-Verlag.