AVOIDANCE POLICIES A NEW CONCEPTUAL FRAMEWORK

Transcription

1 AVOIDANCE POICIES A NEW CONCEPTUA FRAMEWORK David Ulph OXFORD UNIVERSITY CENTRE FOR BUSINESS TAXATION SAÏD BUSINESS SCOO, PARK END STREET OXFORD OX1 1P WP 09/22

2 Avoidance Policies A New Conceptal Framework 1 David Ulph 2 ABSTRACT This paper develops a general theoretical framework within which a heterogeneos grop taxpayers confront a market that spplies a variety of schemes for redcing tax liability, and ses this framework to explore the impact of a wide range of anti-avoidance policies. Schemes differ in their legal effectiveness and hence in the risks to which they expose taxpayers - risks which go beyond the risk of adit considered in the conventional literatre on evasion. Given the individal taxpayer s circmstances, the prices charged for the schemes and the policy environment, the model predicts (i) whether or not any given taxpayer will acqire a scheme, and (ii) if they do so, which type of scheme they will acqire. The paper then analyses how these decisions, and hence the tax gap, are inflenced by for generic types of policy: Disclosre earlier information leading to faster closre of loopholes; Penalties introdction of penalties for failed avoidance; Policy Design fndamental policy changes that design ot opportnities for avoidance; Prodct Register - the introdction of GAARs or mini-gaars that give greater clarity abot how different types of scheme will be treated. The paper shows that when considering the indirect/behavioral effects of policies on the tax gap it is important to recognise that these operate on two different margins. First policies will have deterrence effects their impact on the qantm of taxpayers choosing to acqire different types schemes as distinct to acqiring no scheme at all. There will be a range of sch deterrence effects reflecting the range of schemes available in the market. Bt secondly, since different schemes generate different tax gaps, policies will also have switching effects as they indce taxpayers who previosly acqired one type of scheme to acqire another. The first three types of policy generate positive deterrence effects bt differ in the switching effects they prodce. The forth type of policy prodces mixed deterrence effects. 1 Previos versions of the paper have been presented at seminars at the Universities of Oxford, Edinbrgh eicester, St Andrews and at the College of William & Mary, Virginia. I am gratefl to participants at these conferences for helpfl comments, bt especially to John Moore, Mike Deverex, Clemens Fest and Jdith Freedman. 2 Professor of Economics, University of St Andrews and Senior External Research Fellow, Centre for Bsiness Taxation, University of Oxford. Contact details: School of Economics & Finance, University of St Andrews, St Andrews, Fife KY16 9A, Scotland. d1@st-andrews.ac.k 1

3 Avoidance Policies A New Conceptal Framework Introdction Tax avoidance is a major policy challenge to fiscal athorities in virtally all advanced economies. Not only does it accont for a significant amont of revene loss bt it has a nmber of other problematic featres. A considerable amont of economic resorces are tied p creating, selling, implementing, contering and investigating tax avoidance schemes that are essentially artificial paper transactions that serve no economic prpose other than redcing tax liability. Becase avoidance schemes are often very sophisticated, they are also expensive and so not available to all taxpayers. This leads to a sitation where different taxpayers are paying tax at different rates on the same economic activity. This creates an inefficient allocation of resorces, since taxpayers gain a competitive advantage not throgh sperior technology or prodcts bt throgh greater access to and/or willingness to se avoidance schemes. This also prodces a manifest sense of nfairness, prodcing both horizontal and vertical ineqality. A second reason why avoidance presents policy challenges is that there is a very active market in prodcing and devising avoidance schemes, so fiscal athorities need to consider the effects of their actions on not jst the demand for schemes bt also the spply. Finally tax avoidance poses a significant policy challenge becase avoidance schemes at least those that are well devised and well implemented are legal and therefore will not always be contered throgh the conventional methods of investigations that lead to the repayment of tax pls interest pls penalties. In addition to the possible se of penalties/investigations, fiscal athorities therefore have to consider other ways of contering avoidance: Policy design; GAARS; the coherent principles approach to tax law design 3 ; Information powers sch as disclosre and retrospection. 3 See Pinder (2006) 2

4 To make sensible decisions abot how best to conter avoidance, there are a nmber of things fiscal athorities need to be able to nderstand. What effects do these different policies have on the level of non-compliance? What effects do these different policies have on the type of non-compliance? Do measres to conter avoidance drive taxpayers towards tax planning or tax evasion? What effects do they have both individally and in combination on the tax gap and, more widely, on economic welfare? Yet following Allingham and Sandmo (1972) the focs of virtally all the economic analysis of tax non-compliance has essentially been on evasion 4. While these models prodce sefl insights they sffer from a nmber of limitations when it comes to thinking systematically abot the above qestions: The focs is on a single decision how mch of a redction in tax liability a taxpayer wold seek. It does not address the choice of how to obtain this redction in liability. The focs is on a single taxpayer and does not consider the wider market for schemes. A limited nmber of instrments are considered typically penalties and effectiveness of investigations. The aim of this paper is to provide a new framework for thinking abot avoidance and policies to conter it that overcomes the above limitations. In particlar the framework has the following properties. Taxpayers confront a market that prodces a range of possible avoidance schemes that will redce their tax liability. These schemes can vary from very effective schemes that in some ways come close to tax planning throgh to less effective schemes that, in the limit, might be close to evasion. Taxpayers have to decide not only whether to acqire a scheme and hence their qantm of tax non-compliance bt also which scheme to acqire To characterise the differences in the different schemes it is necessary to consider a mltiplicity of different risks that taxpayers face and not jst a single risk of detection 4 Andreoni, Erard and Feinstein (1998), Cowell (1990), Slemrod and Yitzhaki (2002), provide overviews of mch of the developments on behavior. The paper by Feldstein (1999) has initiated a stream of work on measring the welfare costs of non-compliance. 3

5 and penalties. Taxpayers can mitigate their risks throgh the choice of scheme to acqire. There is a mltitde of taxpayers who differ in both the size of their tax base and hence in their incentives to be non-compliant - as well as in their attitdes/vales towards compliance specifically the reptational concern they have abot being identified as doing anything that might be deemed illegal and so possibly incr a penalty. The market can therefore be thoght of as offering a range of prodcts differing in a nmber of dimensions. If there were no differences in taxpayer attitdes these prodcts cold be ranked by a single qality index, and we wold have a market which was vertically differentiated with higher qality prodcts selling at a higher price. owever, differences in taxpayer attitdes generate an element of horizontal differentiation, so the market is differentiated both vertically and horizontally. Althogh de facto the market for schemes is dominated by a nmber of large players, for the prposes of this paper I will take it that there is enogh competition that prices are effectively determined by simply costs and so will not be affected by policies 5. The framework explains which types of taxpayers acqire which types of scheme. There are many different types of policy instrment, and the framework prodces comparative static predictions of the effects of different type of policy on both the amont and the natre of non-compliance, and hence on the size of the tax gap. An important implication of the framework is that when considering the impact of policies on the tax gap then in addition to the sal direct effects the direct impact of policies on the magnitde of the tax gap associated with any particlar type of scheme - there are now two different types of indirect/behavioral effects that need to be taken into accont (i) The first is the deterrence effect - taxpayers stop acqiring schemes that lower their tax liability. owever in a framework that allows for different types of scheme there will be different deterrence effects on different schemes. So it may sometimes be the case that the deterrence effect of a particlar policy is positive for some schemes bt negative for others. (ii) The second is the switching effect taxpayers switch the type of scheme they acqire. The sign of the switching effect will depend on (a) the direction in 5 In a companion paper, Damjanovic and Ulph (2009) we explore the implications of having an imperfectly competitive market where prices are affected by policies, bt this is for the simpler case where there is jst a single type of scheme available. 4

6 which taxpayers switch; (b) the size of the tax gap for different types of scheme. As we will see, the tax gap for high qality avoidance schemes may be higher or lower than that for lower qality schemes. Since the framework allows for there being a range of possible avoidance schemes it cold, nder some interpretations, be thoght of as encompassing both pre tax planning and pre evasion as extreme cases. owever this is not the interpretation that is proposed in this paper 6. Therefore a fll analysis that encompasses tax planning, tax avoidance and tax evasion wold reqire a somewhat different framework than that employed here, one which treats both planning and evasion as somewhat discretely different from avoidance. 6 The approach nderlying this paper is that tax avoidance arises whenever for a given allocation of real resorces by a taxpayer (prodction/consmption plans) the taxpayer resorts to a set of artificial/paper transactions that converts the streams of income that wold have arisen natrally from these real resorces into different income streams which wold have arisen natrally from an alternative real allocation of resorces and so enables the taxpayer to obtain a lower tax rate. While there is an element of artificiality there is no concealment or misrepresentation of their affairs. By contrast pre tax planning involves taxpayers in rearranging their real affairs (prodction /consmption plans) to obtain the best otcome in terms of profits (firms) or tility (individals) while paying tax at the stattory rates on the incomes that arise natrally from these real plans. To the extent that this rearrangement lowers the rate of tax they pay, this comes at a cost of having to choose a less attractive real allocation than wold be desirable if there were no differences in tax rates. It is this real cost that taxpayers are avoiding by the se of avoidance schemes. Of corse taxpayers sing tax planning may need to pay for good advice in order to choose the best arrangement of their affairs and this be interpreted as sing a scheme. Bt the crcial difference between tax planning and tax avoidance is that tax planning involves a real cost over and above that of acqiring the scheme/advice. On the other hand, tax evasion involves deliberate concealment or misrepresentation of the taxpayers affairs and the certain knowledge that this is illegal, and the type of taxpayers willing to do this may not be the same as those willing to engage in tax avoidance. By way of illstration consider a firm facing different rates of corporation tax in different contries. It faces the choice of locating some prodction facility either in contry A or contry B. Gross profits wold be higher in contry A than in contry B bt if the tax rate in contry B was sfficiently far below that in contry A it might relocate to contry B tax planning. Alternatively it might seek to retain its prodction in contry A bt find some artificial way of shifting its profits to contry B albeit at a cost of setting p a complex set of artificial transactions tax avoidance. Finally it cold keep its prodction in contry A bt, in completing its tax retrn simply declare its profits as arising in contry B tax evasion. 5

7 The paper is in for sections. The first section sets ot the basic model; the second derives the eqilibrim predictions abot which types of taxpayer acqire which type of scheme and hence the size of the tax gap. The third derives the comparative static predictions abot the effects of different types of policies on taxpayer behavior and the implications of this for the tax gap, while the forth concldes. Section 1 The Model Consider a taxpayer with tax base Y. The taxpayer considers acqiring a tax avoidance scheme that will lower the rate of tax on this base by the amont Δ t > 0, giving a potential fll tax saving ΔT = Δt. Y 7. This avoidance scheme is one of a class of avoidance schemes that exploit a particlar differential in tax rates between very similar activities. These different rates cold reflect: the internal tax policies of a given jrisdiction -for example the fact that in the UK National Insrance contribtions are imposed on earned income bt not nearned income; differences in tax policies in different jrisdictions; opportnities to shift taxable income across time and so get a lower disconted tax rate. In what follows the differential tax rate Δ t will be referred to as the tax wedge. There may be many different ways of constrcting schemes that exploit any given tax rate differential. So the taxpayer faces a market in which there are a nmber of different schemes prodced and sold by varios companies 8. They all deliver the same redction Δt in the tax rate bt differ in both the risks with which they face the taxpayer and their prices. To describe these risks and the prices consider for the moment a single generic scheme. 7 Implicit in this is the idea that this all refers to a specific period of time say a tax year over which tax is de. 8 Since, as indicated above, all schemes are going to be sold at a price eqal to their marginal cost, it wold be possible to allow for in-hose schemes prodced by taxpayers that can match the least-cost technology in the market. 6

8 1.1 A Generic Scheme Each scheme confronts the taxpayer with the following five risks. Risk 1 egal Effectivenes The first risk is whether or not the scheme works in law is legally effective. The significance of a scheme s being effective is that in this case the tax athorities have no powers to recover any tax saved let alone impose any penalties for the se of sch a scheme. owever if a scheme is ineffective, and if a taxpayer is detected and sccessflly challenged for sing sch a scheme then the tax athorities can recover the tax saved (pls interest) and may be able to impose a penalty - to the extent that the tax athority has a policy of penalising failed avoidance schemes 9. Whether or not a scheme works is ltimately determined by the corts. For a scheme to work in law it mst be that case that every step in what is often a very complex seqence of transactions mst work and mst also be properly implemented. Assme that, for some schemes at least, at the time that the taxpayer acqires the scheme, there cold be some scope for dobt as to whether or not it really does work. So let p, 0 p 1 e e e be the probability that the taxpayer attaches at the time of acqisition to the scheme s being effective. This probability will be based on advice that the taxpayer will have received when the scheme is spplied. It is assmed that this is good advice and that p is the tre probability of the scheme s effectiveness and is the same for all taxpayers. The vale of p e will vary across schemes. Risk 2. egislative Change Even if a scheme works in law, there is still a risk that the tax athority 10 decides that the arrangements on which the scheme depends for generating the tax saving involve the exploitation of some loophole in the legislation and takes legislative action to close the 9 Sch a policy wold of corse be controversial 10 Strictly speaking it is ltimately Parliament that decides to pass legislation closing loopholes, this legislation being introdced by Ministers, this decision drawing on advice by tax athority. Bit for prposes of this paper these distinctions do not matter all that matters is the probability of legislation being introdced. So, for simplicity, I will contine to talk abot the tax athority as being the decision maker on this isse. 7

9 loophole ths making the scheme incapable of delivering the saving. et p be the probability that the legislation is nchanged so ( p ) 1, 0 p 1` is the probability that the loophole will be closed. Again these are the probabilities at the time of acqisition. Whether or not the tax athority closes the loophole will depend on a nmber of factors amongst which is the nmber of taxpayers sing the scheme or anticipated to se it. For the prposes of this paper it is assmed that the likelihood of the loophole being closed depends solely on the natre of the loophole e.g. the extent to which it is thoght by the tax athority to significantly breach the spirit of existing legislation. As above it is assmed that the taxpayer obtains good advice abot this at the time of acqisition and so knows the tre vale of p The probability p will vary across schemes. Althogh the athority s decision to close a loophole is assmed to depend solely on the natre of the loophole, the precise vales of and the way these vales vary across schemes will depend on policies implemented by the tax athority and in particlar its willingness to give clear gidance. At one extreme the tax athority may have a very clear view of what types of scheme/loophole it will try to close and what types of scheme it will not close. So whether this clarity of viewpoint is commnicated (effectively) to taxpayers, or whether they jst come to learn it over time, we can characterise this sitation as one in which there will be a broad range of schemes for which p = 1, another broad range of schemes for which for which p = 0, and only a very narrow range of schemes 0 < p < 1. This is the case where the athority offers clear gidance or creates what are often called bright lines. Alternatively the athority may be either nable or nwilling to give clear gidance on what types of loopholes it will seek to close and decide everything on a case-by-case basis, with perhaps jst a few illstrations of the types of things that wold definitely lead to legislative action and the types of things that wold not. We cold captre this sitation by assming that there is a very narrow range of schemes for which p =1 (so athority will definitely take no legislative action) and another narrow p range where p = 0 (so the athority will definitely take legislative action) there is a wide range of schemes for which it is nclear whether or not the tax athority will act, and so 0< p <1. Indeed in the case of very considerable ncertainty, it cold be that for most of 8

10 1 the schemes in this latter range p =. We can therefore characterise policies sch as 2 prposive drafting that aim at giving greater clarity/ gidance to taxpayers as a policy that raises p for high qality schemes bt lowers it for low qality schemes. Risk 3 Speed of egislative Action If the scheme works in law, bt will be closed throgh legislative action, the taxpayer has to think abot the speed with which the loophole on which it depends will be closed down, and, if it is closed, whether there is any element of retrospection whereby any tax savings already made when the loophole is closed will be lost. This can all be smmarised in the parameter ϕ, 0 ϕ 1 that measres the fraction of the possible tax saving Δ T that the taxpayer expects to retain in the event of the tax athority s taking legislative action to close the scheme. The vale of ϕ will be inflenced by things like disclosre powers which will affect the speed by which the tax athority becomes aware of the existence of certain types of scheme and whether or not some element of retrospection applies. For simplicity it is assmed that ϕ does not vary with the scheme that is sed, and that the taxpayer has a pretty good idea of the vale of ϕ based on past experience of how qickly and freqently loopholes get sht. Risk 4. Sccessfl Challenge to egally Ineffective Schemes If the scheme is ineffective (fails in law) the taxpayer has to consider the possibility that the tax athority sccessflly challenges it and recovers all the tax pls interest, pls, possibly, a penalty. et p, 0 p 1 be probability that the tax athority sccessflly challenge the c c scheme. This is the prodct of 3 nderlying probabilities: that the tax athority investigates the taxpayers; if it investigates, that it discovers the scheme has been sed; if it discovers, it sccessflly demonstrates that the scheme fails and collects all the tax pls interest pls penalties. For the prposes of this analysis, we do not need to keep track of these separate nderlying probabilities. 9

11 For simplicity it is assmed that pc does not vary across schemes or taxpayers. It wold be relatively straightforward to introdce a more general treatment, bt this is not central to the analysis. Finally it is assmed that throgh experience, interaction with other taxpayers, media coverage etc, taxpayers have a good nderstanding of the vale of p c. Risk 5 Imposition of Penalty The final risk that the taxpayer faces is that if the scheme is ineffective and is sccessflly challenged then the taxpayer will not only have to pay back all the tax (pls interest), bt may in addition have to pay a penalty, which is typically a fraction of the tax saved by the scheme. In practice there is some discretion as to the extent of the penalty imposed, so let f 0 be the expected fraction of the tax savings that will be imposed as a penalty. While in principle this cold vary across schemes for simplicity it will be assmed that this is the same for all schemes. It is important to recognise that some taxpayers may also sffer some reptational damage if they se an ineffective scheme that is effectively challenged. While there may be a nmber of factor s affecting the extent of this damage, for simplicity I will assme that it is proportional to the size of the taxpayer s tax base, Y. et ρ 0 be the factor of proportionality and call this the reptational concern of the taxpayer. This is a parameter that varies across taxpayers. Finally let C > 0 be the cost of acqiring this particlar scheme 11. This will either be the cost of devising the scheme in hose or else the price of bying the scheme in the market. In the latter case it is assmed that the price takes the form of an pfront fee. In general these costs will vary across schemes. In particlar there are a nmber of reasons for thinking that schemes that are more likely to be legally effective will be more expensive: more or more expensive - resorces may be reqired to check that all the steps in a given scheme really do work in law; schemes that are to work in law will have to be installed more careflly; 11 The cost of acqiring a scheme incldes the cost of implementing it. In principle these costs can encompass not jst the costs of prchasing/devising and implementing the scheme bt also any real economic costs the taxpayer faces in aligning its bsiness with the tax system. The more artificial is a tax avoidance scheme, the lower these latter costs will be. 10

12 schemes that are close to tax planning may involve the taxpayer in having to re-deploy real economic resorces rather than sing artificial paper transactions to lower taxes. Bringing this all together, we can see that the net gain to a taxpayer with tax base Y and reptational concern ρ from acqiring any particlar scheme is: ( ) ( ) π( Y, ρ) =Δ ty pe p + (1 p). ϕ + (1 pe).(1 pc) (1 pe). pc. f 1 pe pcρy C (1) The first term on the RS of (1) is the expected net financial benefit to the taxpayer from acqiring the scheme. The first term in sqare brackets is the expected fraction of the fll tax saving, ΔtY, that the taxpayer will obtain from acqiring the scheme if it trns ot to be effective The second is the expected fraction of the fll tax saving if the scheme is ineffective bt the tax athority fails to sccessflly challenge it. The third terms is the expected loss to the taxpayer (as a fraction of the fll tax saving) if the scheme trns ot to be ineffective, the tax athority monts a sccessfl challenge and not only recovers all the tax (pls interest) bt imposes a penalty. The second term on the RS of (1) is the expected cost to the taxpayer s reptation from acqiring a scheme that trns ot to be legally ineffective and is sccessflly shown to be so by the tax athority. The final term on the RS of (1) is the cost of acqiring (and implementing) the scheme 12, which, as noted, takes the form of an pfront charge The expression in (1) can be re-written as: { [ ] [ ]} π( Y, ρ) =Δ ty p p + (1 p ) ϕ + (1 p ) 1 p (1 + f) (1 p ) p ρy C. (2) e e c e c et s make the standard and qite realistic 13 assmption that the probability of sccessfl challenge of an ineffective scheme is sfficiently low that 12 Notice that what this brings ot is that there is always at least one downside to avoidance the cost of acqiring the scheme. For taxpayers with concerns abot their reptation there will additionally be a downside associated with avoidance schemes that trn ot to be ineffective and are sccessflly challenged. 13 If this assmption failed to hold nobody wold ever acqire a scheme that was close to pre evasion, i.e. pe 0, and yet evasion clearly takes place. 11

13 1 p (1 ) c + f >0. (3) Given (3), the term in crly brackets on RS of (2) is positive and less than 1 and represents the fraction of the fll tax savings, Δ ty that the taxpayer can expect from the scheme. We can think of this as measring the qality of the scheme denoted by q, so [ (1 ) ϕ] (1 )[ 1 (1 )] q = p p + p + p p + f. (4) e e c In smmary a scheme is characterised by: the 5 parameters ( p, p, p, ϕ, f ) which captre the 5 risks faced by the taxpayer; e c the size of the tax wedge, Δ t that this class of avoidance schemes is trying to exploit; the costs C of acqiring the scheme. Given or assmptions two of the risk parameters - p e and p - will vary across schemes as will the costs of acqiring a scheme. So the taxpayer can affect the risks by the decision abot which scheme to acqire albeit at a price. The probability of a scheme s being legally effective, government/tax athority. While p e, is prely a featre of the scheme and cannot be affected by the p - the risk that the legislation on which a scheme depends will remain nchanged will also vary across schemes depending on the details of the how they are constrcted, the precise vales of p attaching to varios schemes will also be inflenced by the tax athority throgh the gidance that it offers. The remaining risk parameters p, ϕ f as well as the tax wedge Δ t being exploited by this particlar class of c, schemes are all constant across schemes and will all also be affected by policies prsed by the tax athority and/or government. The five risk parameters affect the qality, q, of a scheme. Notice that qality is monotonically increasing in p and ϕ ; monotonically decreasing in p and f. owever it it will be monotonically increasing, constant or decreasing in p e according as p + ( 1 p) ϕ > 1 pc (1+ f). (5) < c 12

14 So a taxpayer will perceive a less legally effective scheme to be of higher qality if the risk of being effectively challenged and the conseqent penalty is very low, while the risk of having the loophole changed qickly is qite high. The final aspect of any given scheme that we need to nderstand is the tax-gap to which its acqisition wold give rise. The tax gap is a measre of the amont of tax that the tax athority ltimately fails to collect throgh varios types of non-compliance. There are definitional isses srronding the tax gap and in particlar the qestion of how avoidance gets treated. It wold be generally agreed that if a taxpayer acqires a scheme that is illegal/ineffective, then the taxpayer has been non-compliant. owever this will only give rise to a tax gap if the athority fails to sccessflly challenge the scheme and recover the tax thogh it is important to note that the income raised from any penalties that are imposed is typically not conted as helping to redce the tax gap. If the taxpayer acqires a scheme that is effective and if the athority decides not to change the legislation on which the scheme depends for its effectiveness then this sggests that the scheme complies with both the letter and the spirit of the legislation and so acqiring sch a scheme wold not cont as non-compliance. owever if the athority decides to close the loophole then the taxpayer has not complied with the spirit of the legislation, and the prchase of this scheme wold constitte non-compliance. If a scheme is effective the tax athority has no powers to recover tax so the extent to which this non-compliance gives rise to a tax-gap depends on the speed with which the legislation is closed. et g denote the fraction of the potential tax saving ΔtY. that will ltimately fail to be collected if a taxpayer prchases a generic scheme. Given or above discssion this is defined as ( ) ϕ ( ) ( ) g = p.1 p. + 1 p.1 p 0. (6) e e c Notice that g = 0 if an only if a scheme is flly compliant with the letter and the spirit of the legislation that forms the basis of its effectiveness, and which conseqently will not be changed. 13

15 Now other things being eqal the tax athority prefers schemes with lower tax gaps to those with higher tax gaps. The interesting qestion is therefore how its ranking of schemes compares to the qality ranking by taxpayers and whether their views are always opposed to one another or might sometimes be aligned. From (4) and (6) we see while an increase in p raises the qality of a scheme it lowers the tax gap. owever the tax gap will be monotonically increasing, constant or decreasing in So if ( p ) p e according as 1 ϕ > 1 pc. (7) < the chances of sccessflly challenging a scheme if it is ineffective is qite high, schemes in this area are typically not compliant with the spirit of the law so the tax athority is very likely to want to close the loophole, bt it jst takes a long time to spot and change the legislation, then the tax gap will be higher the more legally effective is the scheme. The following proposition smmarises the circmstances nder which the perception of qality by the taxpayer is aligned with that of the tax athority. Proposition 1 Schemes that are perceived to be of higher qality by the taxpayer will have lower tax gaps if: (i) they have the same degree of legal effectiveness; (ii) they have the same probability, p, that the legislation on which they depend will be changed and if ( 1 ) ( ) ( 1 ) pc p + p f < c p ϕ < 1 p c. 1.3 The Market for Schemes As mentioned above we want to think of there being a differentiated prodcts market that prodces a variety of schemes within a particlar class that all bring abot the same redction Δt on a given tax base Y. The schemes work in different ways and so offer taxpayers different exposres to the 5 risks identified above. Schemes differ in: p e - the probability that they work in law; p - the perceived probability that the tax athority will leave nchanged the legislation on which the scheme depends; 14

16 C - the costs of acqiring the scheme. In relation to the costs to the taxpayer of acqiring a scheme, C, in this paper it is assmed that: (i) the market for schemes is perfectly competitive; (ii) each type of scheme is prodced at a constant marginal cost. Given these assmptions the cost to the taxpayer of acqiring the varios schemes are constant and naffected by the varios policies prsed by the tax athority 14. While there cold be very many schemes in the market with different types of taxpayer acqiring different schemes, it will simplify the analysis greatly if we assme that there jst two schemes. The first is a pre avoidance scheme for which p = 1, p 1 while the e second is a scheme that lies in the grey area where p e < 1, p is lower for this scheme that got the first. p < 1 and, moreover the vale of As noted above there is no garantee that jst becase the first scheme is certainly legally effective it is of a higher qality than the second scheme. Bt in fact the most interesting analysis follows by assming that this is the case and this is the assmption that will be maintained throghot the analysis in rest of the paper. So call the first scheme the high qality scheme and the second the low qality scheme and characterise then by 0< pe < pe = 1; 0< p < p 1;. (8) 0< q = p e p ( 1 p ) ϕ + + ( 1 pe )[ 1 pc(1 + f) ] < q = p + ( 1 p ) ϕ 1 If the costs of acqiring the two schemes are denoted by C, C respectively, then the net gains from acqiring these two schemes are: π ( Y, ρ ) =ΔtY.. q C (9) ( e ) π ( Y, ρ) =ΔtY.. q 1 p p ρ. Y C (10) c 14 A related paper, Damjanovic and Ulph (1999), analyses the effect of tax policies when there is also a market for schemes bt when the price is endogenos. owever in that paper there is a single homogenos prodct. 15

17 Notice that if C C then all taxpayers wold prefer the high qality scheme to the low qality scheme, and the low qality scheme wold not be spplied in the eqilibrim.. So to have an interesting case where both schemes exist in eqilibrim it is necessary to assme that the high qality scheme sells at a higher price than the low qality scheme. Indeed, as will become apparent, it is necessary to make the stronger assmption that the cost differential is greater than the qality differential: C C q > > 1. (11) q Finally notice that the tax gaps associated with these two schemes are: ( 1 ) ϕ ; e ( 1 ) ϕ ( 1 e )( 1 c) g = p < q g p p p p q = + > (12) so the high qality scheme may have a higher or lower tax gap than the low qality scheme. This completes the basic description of the model. The next section sets ot the predictions concerning the natre of the eqilibrim. Section 2: Eqilibrim Predictions In order to nderstand the impact of policy changes introdced by tax athority, we first need to determine which taxpayers acqire a scheme and which one this is. A taxpayer will acqire a particlar scheme if and only if: (i) it gives at least as great an expected net gain as the other scheme; (ii) the expected net gain is positive. 16

18 2. 1 Who acqires what Consider first the conditions nder which each of the two schemes makes a positive retrn for the taxpayer. If we start with the high qality scheme, then since it works for sre in law, we see from (6) that reptational risk is not a factor affecting this consideration, and that the high qality scheme will be profitable as long as the taxpayer has a sfficient large tax base. Specifically the high qality scheme is profitable so long as Y C tq. 0 > Y =. (13) Δ 0 ere Y is the critical size of the tax base at which a taxpayer is jst indifferent between acqiring the high qality scheme and having no tax scheme. Now consider the low qality scheme. et ρ = Δtq. ( pe ) 1. p c > 0, (14) and notice that, from (10), it follows that for all ρ ρ the net gain from the low-qality scheme is negative no matter what the vales of Y and C. In other words, there is a grop of taxpayers with a sfficiently high reptational concern who will never acqire the low qality scheme however cheap it is and however large is their tax base. Those who have a lower reptational risk factor will be prepared to acqire the low qality scheme if their tax base is sfficiently high. So, from (10) the low qality scheme generates a positive expected net gain so long as C Y > Y 0 ( ρ) = Δ tq. 1 p. p. ρ ρ < ρ ( e ) c (15) 17

19 ere Y 0 ( ρ ) is the critical size of the tax base at which a taxpayer with low reptational concern is jst indifferent between acqiring the low qality scheme and having no tax scheme. Notice that Y dy (0) > 0; > 0; Y ( ρ) as ρ ρ. (16) d ρ It is also easy to see that, given (11), Y 0 (0) < Y 0. (17) Now consider which of the two schemes gives the highest net gain for any given taxpayer. et Y ( C C ) e ( ) ( ) ( ρ) = > 0. (18) Δt. q q + 1 p. p. ρ c This defines the critical level of income at which a taxpayer with reptational risk factor ρ 0 wold be jst indifferent between the high qality scheme and the low qality scheme. A taxpayer with a larger tax base than Y wold strictly prefer the high qality scheme, becase the retrn on the higher base wold better offset the higher cost of the better qality scheme. A taxpayer with a smaller tax base than Y wold strictly prefer the low qality scheme, becase the lower tax base cold not jstify the higher cost of the high qality scheme. Notice also that Y is a strictly decreasing fnction of ρ. Greater reptational risk makes the low qality scheme less attractive and so taxpayers will be prepared to switch to the high qality scheme at a lower tax base. It is easy to see that it follows from (11) that Y (0) 0 > Y. (19) 18

20 ( ) ( µ ) Finally let µ ρ be the critical vale of ρ for which µ 0 0 Y ρ = Y = Y ρ. This is defined by Δtq q C. (20) 1 p. p q C µ ρ =.1. ( ) e c From (11) and (14) it follows that 0 < µ ρ < ρ. (21) From the above analysis we can work ot how any given taxpayer ranks the three options: by neither scheme; by scheme 1; by scheme 2, and hence which of these decisions they make. All these ideas can be smmarised in the two Figres shown below in the Annex. In Figre 1 all the varios crves discssed above are illstrated. Figre 2 in the Annex shows which taxpayers make which decision. This is a more schematic version of Figre 1 in which sections of crvilinear fnctions have been represented by straight lines. ere Points A, B 0 0 and E correspond to the levels Y (0), Y, and Y (0), respectively of the tax base. Point F corresponds to the critical reptational factor µ ρ at which the three crves 0 0 Y ( ρ), Y ( ρ) and Y intersect. The line AD is a linear representation of a section of the crve Y (ρ), while ED is a linear representation of a section of the crve Y 0 ( ρ). Taxpayers with characteristics that lie above the crve ADC acqire the high qality scheme. Those with characteristics in the area ADE acqire the low qality scheme, while those with characteristics that lie below 0EDC acqire neither scheme. Now taxpayers reptational risk factors are not readily observable - by either tax athorities or social researchers thogh there is some interesting research to be done to investigate this 19

21 frther. owever there are 4 key predictions abot how behavior relates to the size of a taxpayer s tax base. (a) The poorest taxpayers those for whom Y Y 0 (0) - will acqire neither scheme. (b) (c) (d) This is essentially becase their tax base is too small to jstify the p-front costs of acqiring a scheme. There is a grop of lower tax base taxpayers those for whom Y who will acqire either the low qality scheme or nothing at all. There is a grop of middle tax base taxpayers - those for whom 0 0 Y Y 0-0 Y Y Y 0 - who will be observed to acqire either the low qality scheme or the high qality scheme. The richest taxpayers those with Y Y0 - will all acqire the high qality scheme. Finally notice that in the theory developed here there are for key drivers of behavior: Opportnities for avoidance. These are given by Δ t which affects the benefit/demand for avoidance and C, determine the ease of getting advice/schemes. C which captre spply side factors that Incentives for avoidance In this model this is captred by the parameter Y. Attitdes towards avoidance. In this model this is captred by the parameter ρ Risks from avoidance These are captred by the 5 parameters ( p, p, p, ϕ, f ) e c aving determined who acqires which scheme, it is now possible to calclate the tax gap. 2.2 The Tax Gap Assme that the two parameters ( Y, ρ ) that characterise taxpayers are jointly distribted across 2 R + according to the density fnction f( Y, ρ ) > 0 which, for simplicity, is assmed to be everywhere positive. Then it follows from the above analysis of who bys what that the aggregate tax gap, G, expressed in absolte terms is given by ˆ ρ Y ( ρ) g Yf( Y, ρ) dydρ Y ( ρ ) G =Δt ˆ. (22) ρ g Yf( Y, ρ) dydρ + Yf( Y, ρ) dydρ 0 0 Y ( ρ) ˆ ρ Y 20

22 We can see that any policy change will have three different type of effects: Direct Effects - these are defined as the effects that arise throgh the impact of policy changes Δt, g, g holding taxpayer behavior constant. Indirect Effects: these are defined as the effects that throgh the impact of policy changes on taxpayer behavior whether or not they acqire a scheme, and, if so which one they acqire holding Δt, g, g constant. Notice that, to first order, changes in ˆρ have no impact on G so these indirect or behavioral effects can broken into Deterrence Effects. These arise when taxpayers who previosly acqired a scheme no longer do so 15. These can be frther classified and defined as: o Deterrence Effects on ow Qality Schemes. These arise when policy changes raise Y 0 ( ρ ) on [ 0, ˆ ρ ]. Formally o ( ) ( ) ˆ ρ dg = Δt. g Y ρ f Y ρ, ρ dy ( ρ) d ρ. (23) 0 Deterrecet Effects on igh Qality Schemes. These arise when policy changes raise 0 Y and will affect taxpayers for whom ρ ˆ ( ρ) dg t g Y f Y d dy 0 0. = Δ, ρ 0 ˆ ρ ρ. Formally (24) Indirect Switching Effects These will arise to the extent that the policy changes affect Y ( ρ ) on [ 0, ˆ ρ ]. We have ( ) ˆ S ρ dg =Δt. g g Y ( ρ). f Y ( ρ), ρ dy ( ρ) dρ. (25) 0 The sign of this depends on not jst the direction in which Y ( ρ ) is shifted by policy bt also on whether the tax gap is greater for the low qality scheme than for the high qality scheme. Since at this level of generality there is no restriction on the latter, all that we will be able to establish in the next section is the direction in which Y ( ρ ) is shifted by policy. We will say that there has been a Switching Effect Towards ow (resp. igh) Qality according as ( ) ( dy ρ > 0 res p < 0). 15 Obviosly this can happen in reverse and taxpayers who previosly did not acqire a scheme now choose to do so. In this case the deterrence effect will be negative and cold be referred to as an acqisition effect. 21

23 Section 3: Comparative Static Predictions This section examines the impact of varios types of policy on taxpayer behavior and hence, from (23) (25) the indirect effects of these policies on the tax gap. Changes in all six of the policy parameters p, p, p, ϕ, f and Δ t will be examined, bt it will be sefl to grop e c them nder for different types of policy. 3.1 Penalties The first policy is that of increasing the penalty on failed avoidance, which in this model is captred by an increase in f. From (8), (13), (15) and (18) it is easy to see that this has no effect on Y ( ) q and hence on ρ and redce µ ρ. Formally 0 Y, bt will lower and hence increase 0 q Y ( ) ρ, lower 0 0 dq dq dy dy ( ρ) dy ( ρ) = 0, < 0 = 0; > 0; < 0. (26) df df df df df The intition behind this is very straightforward and we have: Proposition 2 An increase in the penalty for failed avoidance will prodce: a positive Deterrence Effect on the ow Qality Scheme; a zero Deterrence Effect on the igh Qality Scheme; a Switching Effect Towards igh Qality. This is illstrated in Figre 3 in the Annex where the new bondaries are represented by dashed lines. What this shows is that taxpayers in the area GDE who previosly acqired the low qality scheme now acqire no scheme the ow Qality Deterrence Effect - while those in the area ADGF now switch from acqiring the low qality scheme to acqiring the high qality scheme the igh Qality Switching Effect. 3.2 Disclosre/Information The second policy is that of disclosre or of greater information powers for tax athorities to learn what schemes are in the market and who is sing them. This has two parts. First it reqires companies that market schemes to reveal their existence and how they work as soon 22

24 as they are broght to market. Each scheme is then given a nmber by the tax athority and any taxpayer who acqires and ses a scheme is reqired to pt the nmber of the scheme on their tax retrn. Corresponding to these two different featres of the policy we can think of disclosre having two effects. (a) For schemes that are effective bt where the tax athority wold have wanted to close the loophole on which they depend for their effectiveness, the tax athority discovers the schemes are in operation more qickly and closes them down faster. This can be captred in the model as a redction in ϕ 16. (b) For ineffective schemes it raises the probability of detection and, probably, the effectiveness of investigation, which can be captred in the model as an increase in Consider these effects in trn p c Earlier Closre of oopholes Earlier closre of loopholes will obviosly lower the potential retrn on both types of scheme making them less attractive to taxpayers. This shows p in the formal model in the fact that, as is easily seen from (8), a redction in ϕ lowers both Y 0 ( ρ ) and 0 (15) raises both. Formally Y q and q and hence, from (13) and 0 dq dq dy dy 0 ( ρ) < 0, < 0 > 0; > 0 (27) dϕ dϕ dϕ dϕ owever the impact of this policy on the difference in qality between the two schemes and hence, from (18), on Y ( ρ ) and so the Switching Effect is ambigos. The reason is that the high qality scheme is more likely to be effective - and so it is more likely that loophole closre is a relevant risk to be considered bt the risk of having the loophole closed is lower so it is not clear that closing loopholes faster matters more for high qality schemes than for low qality schemes. Formally ( ) ( ρ ) d q q > dy > = pe ( 1 p ) ( 1 p ) 0 0 dϕ < dϕ <. (28) 16 A special case of the policy of greater information powers is that of retrospection. Under this provision taxpayers are told (in advance) they will reqired to repay all tax savings made on any scheme that is closed. This can be captred in the model as a special case in which ϕ is driven to zer. 23

25 These reslts are smmarised in: Proposition 3 Earlier closre of loopholes will prodce: a positive Deterrence Effect on the ow Qality Scheme; a positive Deterrence Effect on the igh Qality Scheme; a Switching Effect that can go in either direction.. The effect of earlier closre of loopholes is illstrated in Figre 4 for the case where ( ρ ) dy < 0 and so the Switching Effect is towards the igh Qality scheme and in Figre dϕ 5 for the case where ( ρ ) dy > 0 and so the Switching Effect is in the other direction. In dϕ Figre 4 the ow Qality Deterrence Effect is represented by the area GJDE while the igh Qality Deterrence Effect is represented by the area JKD. The Switching Effect Towards the igh Qality Scheme is represented by the area AJGF.. In Figre 5 the ow Qality Deterrence Effect is represented by the area GDE while the igh Qality Deterrence Effect is represented by the area JKDG. The Switching Effect Towards the igh Qality Scheme is represented by the area AGJF igher Risk of Sccessfl Challenge It is it is intitively obvios and easy to see more formally from (8) that an increase in the risk of sccessfl challenge has exactly the same qalitative effect as an increase in the expected penalty that will be imposed conditional on being challenged. So the predictions are as in (26) and as illstrated in Figre 3. Proposition 4 A greater risk of sccessfl challenge has exactly the same behavioral effects as an increased penalty on failed avoidance The latter are stated in Proposition Designing Ot Avoidance The third type of policy that we can consider is that of redcing the incentive for avoidance throgh re-designing tax policy in a way that redces the wedges in the tax system nder which very similar activities get taxed at the different rates. Within the framework adopted here we can think of as this as a redction in Δ t. From (8) this has no impact on either 24

26 0 0 q or q thogh from (13), (15) and (18) this directly increases Y, Y ( ρ ) and Y ( ρ ). Formally 0 0 dy dy ( ρ) dy ( ρ) > 0; > 0; > 0. (29) dδt dδt dδt Proposition 5 Policy reforms which lower the opportnity for avoidance will prodce: a positive Deterrence Effect on the ow Qality Scheme; a positive Deterrence Effect on the igh Qality Scheme; a Switching Effect Towards ow Qality.. The intition is clear. owering the tax wedge redces the gain from acqiring an avoidance scheme, while costs remain naltered, so creating a deterrence effect on both types of scheme. On the other hand precisely becase one scheme is of higher qality than another the redction in the tax wedge will have a bigger redction in its net retrn than in that of the lower qality scheme, casing taxpayers to switch towards the low-qality scheme. The comparative static predictions are illstrated in Figre 5, where the interpretation is jst as above in section Better Gidance The final type of policy that will be analysed is that of giving better gidance as to how well varios tax schemes fit with the spirit of the legislation and so the likelihood of the legislation on which schemes depend for their effectiveness being changed. As indicated in Section 1 this policy can be captred in the crrent framework as an increase in redction in p and a p. This makes high qality schemes more attractive and low qality schemes less attractive. The way this shows p is that, from (8), the effect of these changes in p will be to increase q and lower q. From (13), (15) and (18) these changes will in trn lower both 0 Y and Y ( ρ ) and raise Y 0 ( ρ ). Formally: 0 0 dq dq dy dy ( ρ) dy ( ρ ) > 0, < 0 < 0; > 0; > 0. (30) dp df dp dp dp 25