Optimal tax compliance and penalties when the law is uncertain.
This article examines the optimal level of tax compliance and the optimal penalty for noncompliance in circumstances in which the substance of the tax law is uncertain--that is, when the precise application of the Internal Revenue Code to a particular situation is not clear. In such situations,
With these questions in mind, this article shows that, applying standard assumptions from the economic literature on deterrence, the tax penalty regime that would induce the optimal reliance (or non-reliance) on uncertain tax laws depending on the circumstances would involve (1) a rule of strict liability with respect to taxes owed as well as to the penalty and (2) a penalty that roughly approximates the famous Bentham-Becker punitive fine, calculated by dividing the harm (the underpaid tax) by the ex ante probability that the harm would be detected. This article also explains why a fault-based approach to tax penalties, under the standard assumptions of the classical deterrence model, would not work as well as the strict liability approach. Reasons for the inferiority of the fault-based approach include its comparatively high administrative costs, its inability to properly regulate "activity levels," and its relatively unattractive distributional consequences. This article concludes, however, that if Bentham-Becker level penalties or wide-spread use of tax liability insurance are not feasible, a second-best case can be made for using a fault-based penalty regime similar to the one currently in force. The framework used in this article may have implications for any area of law where the substantive law is uncertain.
I. INTRODUCTION
This article examines the optimal level of tax compliance and the optimal penalty for noncompliance in circumstances in which the tax law is substantively uncertain--that is, when the precise application of the Internal Revenue Code (Code) to a particular situation is not clear. These circumstances arise more often than one might think. There are many clear-cut cases in tax. For example, when tax protestors say that the U.S. tax laws do not require them to pay any tax on their U.S. income, they are wrong; and if they take such a position on their returns or, on the basis of this position, opt not to file returns at all, they will, if caught, face a substantial fine or even spend some time in jail. On the other hand, there are many close cases in tax.
Say you take a trip to Miami, during which you attend a job-related conference for one day and lounge on the beach sipping margaritas for two. Is the trip "primarily for business purposes" or not? Good question.
(1) Or say you are a taxpayer engaging in a transaction primarily for the purpose of reducing your income tax liability and the transaction entails some, but very little, economic substance. Will a court respect the form of the transaction and allow the tax treatment you have chosen? Another good question. (2) These sorts of questions pervade the tax law, producing interesting issues for tax lawyers as well as good test questions for the basic income tax class in law school. This article addresses this sort of legal uncertainty.
A number of interesting questions arise in these ambiguous situations. I will focus on two of them. First, as a normative matter, what degree of substantive legal certainty should taxpayers insist on before they rely on a particular interpretation of a tax rule? That is, if a given transaction is not clearly prohibited, but neither is it clearly allowed, what is the appropriate threshold of confidence that taxpayers ought to have before engaging in the transaction? Take the mixed business/personal Miami trip mentioned above. How sure should you be about the deductibility of those expenses before taking such a position on your return? Or how much economic substance must a transaction have--how likely must a pre-tax profit be--to justify actually going forward? If we can answer those questions, the next question follows: what penalty regime would give taxpayers the right incentive with respect to relying on substantively uncertain tax law?
To address these and related questions, I will use the following hypothetical: Joe Taxpayer (who can be thought of either as an individual investor, a business owner, or a manager of a corporation) is trying to decide whether to invest, or have his company invest, in a particular business transaction. In making this decision, Joe takes into account a range of issues, all of which boil down to one obvious question: how much money will the transaction make net of costs? As part of this analysis, Joe considers the legal consequences of the investment, including the likelihood that the investment might lead to some sort of civil litigation or government enforcement action. Joe evaluates these legal risks then weighs them against the expected benefits of the deal. Among the legal risks he contemplates are the possible tax consequences of the transaction.
Now, focusing the analysis on the tax planning question, assume that from Joe's perspective (or that of his company) the investment is worth making only if it qualifies for a particular tax treatment. That is, assume the deal makes sense--its overall expected benefits exceed the overall expected costs--only if it qualifies as a "nontaxable transaction" or, alternatively, only if it generates a special tax loss or tax credit that can be used to offset taxes on other income. Thus, the after-tax profitability of the deal turns on the answer to the tax question. Now here is the problem: If Joe's expert tax advisor tells him that the special tax treatment he seeks for the transaction is neither clearly forbidden nor clearly legal under the existing tax laws, how should Joe proceed? In other words, if the law in question, at least as applied to Joe's particular transaction, is uncertain (in terms of how it will be applied ex post by the Internal Revenue Service (Service) or courts to particular transactions), what incentive does society want Joe to have in this situation? What is the optimal degree of tax compliance and what is the optimal tax penalty regime?
As it turns out, the answers to these questions depend on a number of factors. To see this point, let us simplify the analysis further by assuming that the only thing Joe cares about with respect to tax planning is the expected value of the sum of the possible back-taxes (plus interest) and the potential penalty. Joe, in other words, is a rational actor in the traditional economic sense of the term, a true homo economicus; more pejoratively, Joe is a quintessential example of Holmes's "bad man." (3) Assume further that not only is Joe without a conscience but he faces no informal external sanctions either, such as social norms against tax noncompliance. Either his friends, neighbors, co-workers, and fellow corporate managers are utterly indifferent to Joe's reputation for paying his taxes, or he is indifferent to their opinions.
Given all of these simplifying assumptions, Joe's decision regarding whether to undertake the particular transaction in question, and whether to report the transaction on his tax return in the desired manner, will depend on his ex ante assessment of (1) the probability that the particular tax position in question will be discovered and scrutinized by the Service, (2) the probability that, if detected, the position would be rejected by the Service and ultimately by a court, and (3) the size of the penalty in the event of both detection and rejection. (4) Obviously Joe would not be able to estimate these variables with great precision, but presumably he would give it his best shot or pay a tax advisor to do so. It also seems sensible to assume that Joe would invest in additional information, up to the point at which the marginal cost of the additional information equals the marginal benefit gained from the information. Again, assuming there is some residual uncertainty even after these investments in information are made, then the question of whether the deal is profitable to Joe will depend on this evaluation of uncertain tax law and uncertain tax law enforcement.
That is Joe's perspective. What about society's perspective? What does society want Joe to do when the substantive content of the law can only be estimated? (5) Start with the two obvious and extreme positions. First, one could hold the view that Joe should go through with the transaction in question only if the tax position that makes the deal attractive--the tax-deferral or the special tax loss or special tax credit or whatever--is certain to be upheld by the Service and the courts. That is, Joe should adopt the particular interpretation only if he is certain that Congress intended for the preference to apply to the particular type of transaction that he is considering and to the particular class of taxpayers of which he is a member. Alternatively, one could hold the view that, so long as the tax position in question is not clearly and indisputably forbidden by the Code, the taxpayer should feel free to go through with it even if extraneous evidence, or common sense, makes clear that Congress did not have the taxpayer or his type of transaction in mind for this particular tax benefit. Obviously, neither of these extreme positions is the right answer. Rather, the right approach will depend on the circumstances. Indeed, this article contends that, when the substantive tax rules' meanings are uncertain, as applied to a taxpayer's particular situation, the taxpayer should (and inevitably will) make his decision based on his, or his legal expert's, probabilistic assessment of what the law actually is--or what a court would say that it is. I argue further that the formal penalties for tax underpayment should incorporate and enforce this concept of probabilistic compliance. Interestingly, as the discussion below explains, the existing tax penalties to some extent already take this approach.
Part I of this article explains the primary sources of uncertainty in the income tax laws. Part II operationalizes the concept of substantive tax law uncertainty by adopting a probabilistic understanding of substantive law and by describing what I call the "tax compliance continuum." Part III adopts the assumption of "detection certainty," the idea that every tax position actually is scrutinized by the Service and demonstrates the deterrence benefits of a strict liability tax penalty regime, especially in terms of its ability to induce taxpayers to behave optimally with respect to ex ante legal uncertainty. Part IV introduces the problem of "detection uncertainty," known in the tax context colloquially as the "audit lottery," and shows how, under certain assumptions, the classic Bentham-Becker punitive penalty can induce taxpayers to internalize the expected tax liability associated with substantively uncertain tax positions. Part V explores the role that tax transaction liability insurance, whether privately or publicly provided, could play in such a strict liability tax penalty regime with a punitive Bentham-Becker penalty. Part VI explains (1) why, under traditional deterrence assumptions and assuming a fully deterring Bentham-Becker penalty, a fault-based tax penalty regime is inferior to a strict liability regime, but (2) why, assuming the Bentham-Becker penalty is unrealistic, a fault-based regime might be the second-most optimal.
II. SOURCES OF TAX LAW UNCERTAINTY
Before the analysis can get under way, an initial question is: why is there substantive tax law uncertainty in the first place? Those with only a passing familiarity with the U.S. tax laws might question the plausibility of the claim that the tax law is rife with uncertainty. After all, the U.S. federal income tax system is among the most detailed and comprehensive legal regimes in the world. Given the thousands of pages of the Code and umpteen-thousands of pages of Treasury Regulations, (6) the nonexpert might be tempted to conclude that precise tax treatment of every conceivable transaction should be derivable from the existing tax laws, so long as one has the time to read and the expertise to understand the Code and regulations--or has the resources to hire someone else to do it. Everyone realizes, of course, that the Code is inscrutable to the common man but surely the tax cognoscenti, whose opinions can be bought for a price, can find the answers in all of those pages. For two general reasons, however, there is, and likely will always be, considerable uncertainty in the tax laws. The first has to do with the complexity of the laws. The second has to do with unintended gaps or loopholes in the law.
By most accounts, the U.S. federal income tax is "the paradigmatic system of rules" rather than standards. (7) A taxpayer's tax liability is determined by applying a precise rate schedule to that taxpayer's taxable income. This sounds simple enough but, as already mentioned, the number of words in the Code that must be consulted to determine one's income tax liability is staggering. Furthermore, the level of specificity, and hence the complexity, with which the Code defines terms and explains procedures is legendary. Some of this complexity results from Congress's habitual attempts to enact social policy into the Code, whether it be subsidizing a particular form of investment (such as research and development) or a particular class of taxpayers (such as families or the poor). Another source of complexity is the attempts by Congress and the Treasury Department to close unintended loopholes in the tax laws, discussed further below. Whatever the source, the complexity of the tax rules is a primary source of substantive legal uncertainty. This sort of uncertainty is the primary reason why so many individual taxpayers either have their returns prepared by professionals or rely on computer programs such as Turbotax for assistance. Every year, as the Code increases in length and complexity, more taxpayers find it useful to seek expert help to reduce the uncertainty associated with filing their returns. (8) Complexity-induced uncertainty is a problem that plagues not only unsophisticated individuals but also wealthy individuals and large corporations who can afford expert legal advice.
A second source of substantive legal uncertainty in the tax law is somewhat less familiar to nonexperts, although it is well known among tax lawyers, accountants, and even beginning tax students. Although there are obvious reasons to have numerous detailed tax rules (for example, to set out clearly the tax treatment of the most frequent types of transactions), there is simply no way for Congress or the Treasury Department to anticipate every contingency and provide in advance the precise tax treatment of every conceivable transaction or investment. (9) The world is just too complex. Even if it were conceivable to fully specify the Code in this extreme sense, doing so would be unreasonable. At some point, the increased degree of ex ante precision in the law is outweighed by the cost of figuring out such details in advance and by the loss of flexibility that accompanies ex ante rulemaking. This is the source of the unintended loopholes mentioned above. (10)
The problem of unintended loopholes appears especially bad when one considers taxpayers' incentives to find (or some would say, to create) these unintended gaps and exploit them to their advantage. Once an unintended loophole is found to work for one taxpayer, there is a natural tendency for others to use it as well. Thus, what starts as a small gap in the tax laws can, under the right conditions, become a yawning chasm and the ultimate result can be both inefficiency (because taxpayers alter their investment decisions in the effort to minimize their taxes) and maldistribution of resources (because the ability to exploit tax loopholes is not evenly or otherwise fairly distributed across taxpayers). (11) When this process of unintended-loophole discovery and exploitation occurs, it is a virtual necessity that the Service and the courts be empowered to apply some general anti-abuse gap filler, some statutory interpretive standard (in the rules-standards sense of the word) that limits these opportunities. Examples of these anti-avoidance standards in tax law are the economic substance and business purposes doctrines. (12) The use of such anti-avoidance standards, however, comes at a cost to the system. Employing these ex post anti-avoidance standards increases ex ante substantive legal uncertainty, as taxpayers cannot know for sure in advance where the Service or a court will draw the economic substance or sham transaction line after the fact--just as a driver cannot know for certain where the negligent-driver line will be drawn. The uncertainty created by the existence of anti-avoidance doctrines is generally a good thing when compared with the alternative of allowing taxpayers to exploit all unintended loopholes with absolute impunity.
One conclusion that follows from the preceding discussion is that some level of substantive tax law uncertainty is inevitable. This is not to say of course that Congress and the Treasury Department have no control over the amount, degree, or even the type of legal uncertainty that exists. Obviously they do. They can invest more or less time in specifying the rules in advance, more or less effort in avoiding unnecessary and confusing complexity. Indeed, there is a large political science literature that explores the question why and under what circumstances legislatures would intentionally write vague or ambiguous statutes, pointing out legislatures' desires to shift the responsibility of unpopular decisions to enforcement agencies, to courts, (13) or to a legislature's inability to reach a stable consensus on legislative language. (14) Moreover, particularly in the tax context, it is possible to conceive of Congress, perhaps with the cooperation of the Treasury Department, actually using the level of legal uncertainty as another tool in their tax enforcement toolbox. That is, if taxpayers are thought to be risk-averse, it is not difficult to imagine how strategically increasing tax law uncertainty, and hence the variance of possible tax outcomes, in some contexts could serve the same function as increasing noncompliance penalties directly. (15) Having said all of that, this article focuses on legal uncertainty that is unavoidable by lawmakers. Therefore, for most of this article, I assume that the choice of the optimal tax penalty and tax liability rule should ignore the effects of the penalty and rule on Congress's decision to increase or decrease the certainty of the laws.
III. A CONCEPTION OF UNCERTAIN (OR PROBABILISTIC) TAX LAW: THE TAX COMPLIANCE CONTINUUM
The next step in the analysis is to provide some functional content to the idea of substantive legal uncertainty. Uncertainty itself is a vague term. It could mean that the law is so vague that the taxpayer cannot begin to guess what it prohibits or allows. Some laws may elicit such an extreme reaction. The analysis of this article, however, will not provide much help in addressing that sort of profound legal uncertainty. Rather, it will focus on situations in which the law is not certain, but educated guesses can still be made about what the law means and how it will be applied to a given situation.
One way to operationalize this conception of substantive legal uncertainty is to array the possible range of tax positions (or interpretations of the tax laws) along a continuum according to their probability of success on the merits assuming they are reviewed by a court. (16) On one end of this continuum lie tax positions that are indisputably illegal. The probability that the Service and a court would uphold such positions if asked to do so is zero. Taking such a position on one's tax return would accurately be characterized as outright tax evasion. On the other end of the continuum are tax positions that are clearly legal, in the sense that the probability that the Service and the courts would sustain them on the merits, if presented with the question, is equal to one. In between these two extreme points are an infinite number of possible tax positions with varying ex ante probabilities of success on the merits. Figure 1 captures the idea of this tax compliance continuum.
This continuum is useful not only because the two endpoints illustrate the most extreme positions, but also because the continuum allows a whole range of tax positions to be graded on the basis of their relative aggressiveness according to their relative position on the continuum. Thus, as we move from right to left along the continuum in Figure 1, the various tax positions represented by the continuum become increasingly aggressive in the sense that their probability of being rejected on the merits by a court, if detected, increases. (17)
[FIGURE 1 OMITTED]
It is easy to conceive of tax positions that would fall on either end of this continuum. On the clearly illegal side are those taxpayers who simply decline to report cash income or intentionally take deductions for expenses that were never incurred. Such behavior constitutes obvious illegal tax evasion. Obvious tax evasion might also include hiding income in illegal foreign accounts. Given the zero or near-zero probability that such positions would be upheld on the merits if detected, the only motivation for taking them is the hope of going undetected. On the other end of the spectrum, there are many tax positions that are clearly legal: a simple business expense deduction or the exclusion of an item that is clearly a gift. There is little dispute that the tax treatment of many transactions is clear and that the answer to many particular tax questions can be known with a high degree of certainty.
More interesting are the many tax positions that fall between zero and one in terms of probability of success on the merits. As any tax practitioner or any student in an introductory federal income tax class can tell you, almost all of the interesting tax questions fall in this range. This holds true regardless of whether the area is corporate, partnership, or individual income taxation. In fact, a reasonable case can be made that the vast majority of the tax issues that end up being decided by a court had, at the time the transaction was planned and carried out, a positive-but-less-than-one probability of success on the merits. Even the most infamous tax-shelter cases, the largely extinct 1970s real-estate-limited-partnership kind as well as the more recent corporate-shelter variety, would doubtless fall somewhere to the right of "clearly illegal" on the continuum. (18) Of course, for most tax-shelter transactions, there would be disagreement about exactly where on the continuum the various shelter transactions fall. (19) But for almost all such transactions, one could at least imagine using something like the continuum above to assess the relative aggressiveness of the taxpayers' positions.
Conceiving of the tax compliance decision in such probabilistic, predictive terms will often not comport with reality. Although we may be accustomed to viewing some taxpayers as quintessential rational actors--I have in mind those taxpayers who spend a great deal of time and money to find loopholes in the law--it is quite a different matter to imagine the average individual taxpayer making a probabilistic calculation to determine what the substantive tax law in fact is. Hence this analysis may not apply to average taxpayers filing out their 1040EZ. At least for sophisticated taxpayers, however, (by which I mean taxpayers with sufficient resources and incentives to hire expert legal advisors) such probabilistic estimates are a part of the game. (20) In fact, the Code and regulations make the application of tax penalties for noncompliance depend on such probabilistic predictive assessments. In that sense, the existing tax penalty structure already incorporates something like the tax compliance continuum described above. That is to say, if the Service and the courts determine after the fact that a particular position taken by a taxpayer is wrong--and thus that the taxpayer owes additional taxes--then the determination of whether the taxpayer must pay additional penalties depends on the Service's (and, if the case winds up in court, the court's) ex post assessment of the taxpayer's ex ante probability of success on the merits.
To get a sense of how a probabilistic predictive analysis of the meaning of an uncertain legal rule might work, consider a current tax compliance penalty provision. Under existing law, if a taxpayer understates her tax liability (that is, the Service and courts determine that she took an impermissible tax position), she will generally have to pay, in addition to the back-taxes and interest, a penalty of 20% of the understated tax unless she can persuade the Service or a court ex post that the position in question ex ante had approximately a 40% chance or better of prevailing on the merits assuming the issue were reviewed by a court. (21) Such uncertain legal positions are said to have "substantial authority." (22) Thus, the current taxpayer penalty regime for tax underpayments incorporates a version of the sort of probabilistic predictive assessment that I describe above; implicit in the whole analysis is the assumption that the substantive tax law itself is uncertain. This definition of substantial authority is not the only example of probabilistically assessed tax penalties. Tax penalties can also be avoided if the taxpayer can show that (1) the tax position at issue was disclosed to the Service, and (2) the position had a "reasonable basis" in the law (i.e., a 20% chance of prevailing on the merits, assuming detection). (23) Thus, if you bring your uncertain tax position to the Service's attention, you are allowed to be somewhat more aggressive, in the sense of taking a position that is a little further to the left on the tax compliance continuum. (24)
There are other examples of the probabilistic reasoning of this sort in the area of tax enforcement. For certain categories of transactions that the Service has reason to believe are of questionable legitimacy, because of the nature of the transactions or because of the Service's experience with similar transactions in the past, special rules apply. For example, these transactions must be reported to the Service; hence the term "reportable" transactions. (25) In addition, to avoid underpayment penalties for such transactions, the taxpayer must be able to show not only that the tax position in question had substantial authority (as defined above) but also that she reasonably believed the position was "more likely than not" correct. (26) That is, she must have reasonably had the view that the position would fall to the right of the mid-point on the tax compliance continuum of Figure 1. (27)
As if all these probabilistic standards were not enough, there are also the rules governing tax preparer penalties. Thus, a tax preparer can be penalized for signing a return or endorsing a tax position that does not have at least a "realistic possibility of success," which is sometimes quantified to mean at least a 33% chance of winning on the merits, assuming detection is certain. (28) Alternatively, if the position in question is disclosed to the Service, the preparer can still sign the return and avoid penalties so long as the position is at least not "frivolous," which some have quantified as something greater than a 10% chance of winning on the merits. (29) Taxpayers themselves also are subject to special penalties for filing frivolous returns and an even greater penalty for taking positions that are clearly illegal.
If one wanted to array all of these various points along the relative continuum of uncertain tax positions, it would look like this:
[FIGURE 2 OMITTED]
All of these various penalties depend not on whether the particular tax position at issue is legal or illegal, but on the position's ex ante probability of being legal or illegal as determined either by the Service, a court, or the taxpayer. Later in this article I will return to the question of whether the above described penalty scheme, or some modified version of it, can be justified on deterrence grounds. But first let us logically consider the prior questions, mentioned in the introduction: what is the socially optimal ex ante probability-of-winning-on-the-merits threshold that taxpayers ought to apply in making tax planning decisions? And what tax penalty regime creates the incentives most likely to achieve that result?
At this point, I need to be clear about what I mean by "optimal" and "efficient" in this context. In one sense, the efficient result in this example would be if the tax could be designed so that Joe Taxpayer could utterly ignore it and thus, make his investment decisions entirely on the basis of his pre-tax calculations. Put differently, the most efficient, or least distortive, tax is not an income tax at all but some form of lump sum tax, perhaps a head tax. I am assuming that society, through Congress, has decided that the overall social-welfare-maximizing form of taxation is an income tax, one that allocates tax burdens in a particular way depending on taxpayers' individual levels of income and that, despite its distortive effects, such an income tax is overall socially optimal. Thus, when I speak of designing the "optimal" tax penalty, I mean the penalty that induces compliance with the law, taking the level of taxation and allocation of the tax burden across different levels of income as given--as having been decided (in some sense correctly, or socially optimally) by Congress. (30)
IV. STRICT TAX LIABILITY AND THE OPTIMAL MERITS PROBABILITY THRESHOLD
To begin to answer these questions, consider the example of Joe Taxpayer in greater detail. Imagine Joe is considering a single transaction or investment that is guaranteed to produce a pre-tax profit of $75. Now assume further that there are only two possible tax treatments of that transaction: either it will produce a tax liability of $100 or it will produce no tax liability at all. (31) Thus, the overall profitability of the transaction depends on the ultimate tax consequences; it either produces an after-tax gain of $75 or an after-tax loss of $25. (32) The problem, of course, is that Joe does not know with certainty what the tax consequences will be. This is one of those transactions for which neither Congress nor the Treasury Department has clearly specified the tax treatment. Hence, the best that Joe can do is estimate, or have his lawyer or accountant estimate, the probability that the zero-tax interpretation of the transaction will be upheld if detected and then make his investment decision accordingly.
Let us also assume that the tax law uncertainty that Joe faces will be resolved only after he has made the investment and only after several years--however long it takes for the Service to select Joe's return for audit and either reject or accept the tax position in question or for the statute of limitations on that return year to run. This assumption also implies that Joe cannot, at a reasonable cost or within a reasonable time, get a private letter ruling to resolve the uncertainty before the investment is made. Assume that, once the transaction is undertaken, it cannot reasonably be unwound or reversed should the tax treatment on which the taxpayer relies happen to be struck down. Rather, in the event the tax position turns out to be wrong, Joe simply has to absorb the extra taxes and penalties, which in this case would again mean that the transaction, from an ex post perspective, would be a net after-tax loser. On a related point, I assume initially that it is impossible to purchase private insurance against the possibility of an adverse tax decision and, in any event, that Joe is risk neutral in the sense of being indifferent between two prospects with differing levels of variance but equal expected values. I relax these assumptions below. (33)
Finally, begging the reader's indulgence, a few additional assumptions common to the economic analysis of law are necessary for the analysis to proceed. The particular implications of these assumptions will be explored later in this article. First, recall the assumptions from the introduction that the taxpayer is a rational actor in the traditional sense and cares only about maximizing after-tax returns. These are obviously essential to the deterrence analysis that follows and are customary in the relevant deterrence literatures. Second, it is assumed throughout this article that the federal tax laws, as ultimately interpreted by a court, represent the will of Congress and are therefore presumptively social-welfare-maximizing. This assumption has two components. First, whatever Congress intends for the tax laws to mean is what maximizes social welfare; this includes both the allocative and distributive consequences of the tax laws. Thus, if Congress decides that a taxpayer who earns $30,000 should pay 10% in taxes and a taxpayer who makes $200,000 should pay 30% in taxes (perhaps because, for example, our society has decided that high-income people place a lower marginal value on money than do low-income people) then that result is, in some sense, social-welfare-maximizing. (34) Second, whenever it is unclear what tax treatment Congress intended for a particular transaction, that question is answered definitively and accurately when a court renders a tax decision. Both of these assumptions are obviously unrealistic. We all know how Congress (with the President's help) can and does mismanage the tax system. Courts are likewise notorious for getting tax decisions wrong. Nevertheless, to render the deterrence analysis tractable, these assumptions--that Congressional intent is welfare maximizing and that courts are always right--are necessary. Moreover, if one is especially troubled by the quality of Congress's tax lawmaking record or by the courts' performance in tax cases, those issues should be addressed directly. (35)
With that lengthy but necessary setup, we can now begin to isolate the factors that determine what the optimal ex ante tax compliance incentives would be in Joe Taxpayer's situation. This part of the analysis relies on the traditional lens of deterrence theory as it has been developed in the economic analysis of legal rules. (36) In that literature, the two most important normative questions are (1) what is the optimal liability rule--strict liability, some variant of negligence, or fault; and (2) what is the optimal level of damages in the private enforcement context, or the optimal fine in the public enforcement context. (37) Applying this framework, it should be clear as an initial matter--although it is never expressed this way in the tax compliance literature--that if Joe Taxpayer's position is certain to be scrutinized by the Service because the probability of detection is one, then the optimal tax liability rule is strict liability and the optimal fine is simply the amount of additional taxes owed plus an appropriate interest charge to account for the time value of money. This combination will produce the right ex ante compliance incentives and will induce Joe to make the investment described above only if it is efficient for him to do so.
Interestingly, this conclusion suggests that such a strict tax liability rule will induce the taxpayer to behave optimally with respect to the question of legal uncertainty; in particular, it will give him the incentive to choose the optimal threshold probability of success on the merits. In the above example, the social-welfare-maximizing choice would be for Joe to make the investment if the probability that the position will be upheld on the merits is greater than 0.25, but not otherwise. For any probability of success on the merits greater than 0.25, the transaction is a positive net-present-value investment after taxes, but not otherwise. (38) A few simple examples illustrate this point. If the probability of success on the merits for this transaction was 30%, the expected tax cost associated with the investment would be $70, still less than the pre-tax profit of $75; hence the deal, for Joe and for society, is worth pursuing. (39) If the probability of success on the merits was 20%, the expected tax cost would be $80, making the deal a $5 loser in after-tax terms. (40) Given these numbers, the optimal merits probability threshold is 25%. This number is entirely an artifact of the arbitrary parameters of the example but the more general point still holds. Assuming detection certainty--the idea that every tax issue gets evaluated by a court and thus that all ex ante uncertainty gets resolved ex post--a rule of strict tax liability that requires the taxpayer to pay the additional taxes ex post in the event the Service or a court finds his position to have been wrong will induce him to internalize his ex ante expected tax liability. This will work with whatever numbers are used in the example.
To put the point even more generally, when taxpayers face conditions of legal uncertainty, social welfare is maximized if they make investment decisions on the basis of their best estimate of the ultimate resolution of that uncertainty. To achieve this result, society wants to make individuals and firms internalize the expected value of the harm that their decisions might cause; harm, in this instance, would be the amount of under-paid taxes. This conclusion is consistent with the conventional wisdom in the economic analysis of tort law, where cost internalization through strict liability is well understood to achieve efficiency in certain settings and under assumptions similar to those made in this article. (41) For example, in the products liability context, if a product manufacturer is trying to decide whether to manufacture and sell a particular product (or whether to make a particular safety innovation in an existing product), the existence of a strict liability tort rule induces that company ex ante to take into account the expected harm that its product might cause (or the expected reduction in harm that the safety innovation might yield). In essence, this is what a strict liability rule in the tax context does as well.
Thus, a regime that makes Joe Taxpayer pay the $100 in back-taxes in the event the Service and courts reject his tax position will induce him to make the social-welfare-maximizing ex ante choice regarding when to take advantage of a given legally uncertain tax benefit. (42) It is also worth making two other aspects of this conclusion explicit. First, it highlights the fact that there is no a priori correct merits probability threshold apart from what is optimal under the circumstances as described above. (43) Second, the optimal merits probability threshold will depend on a number of factors including the amount of the potential tax cost (or tax savings) associated with the transaction if the taxpayer loses (or wins), the amount (and the certainty) of the pre-tax profit expected from the transaction, and, if we lower the assumption of risk neutrality, the taxpayer's taste for risk, though I have assumed risk neutrality to this point. Thus, all else equal, the higher the potential tax cost (or savings) associated with the transaction, the higher the minimal threshold probability of success on the merits will be. (44) This makes intuitive sense. If the tax aspect of a particular transaction is very large relative to the expected pre-tax profit from that transaction, we want the taxpayer--and the taxpayer herself should want--to be certain about the substantive law in question before going forward with the deal. Along the same lines, the relationship between the expected pre-tax profit and the optimal merits probability threshold also makes intuitive sense. If the nontax aspect of a particular deal is relatively large, compared to the tax savings, the taxpayer can afford (and society would want her) to be more aggressive in her interpretation of uncertain tax laws.
A number of interesting observations follow from this analysis. First, even if the probability of success on the merits for a given tax position is extremely low, it can be socially optimal for the taxpayer to engage in the transaction and take the questionably legal position, so long as she believes that the expected pre-tax profit from the transaction exceeds the expected tax liability. This is a point that is sufficiently counterintuitive (and interesting) to bear restating in a slightly different way. Socially optimal behavior, in a world with substantive legal uncertainty, can and often will include actions that turn out, after the substantive legal uncertainty is resolved, to have been illegal. That is just another way of saying that whenever there is substantive legal uncertainty, it is not the case that inaction--or declining to interpret the uncertain law in one's favor--is always the best, most social-welfare-maximizing approach. Where the substantive law is uncertain, a conservative interpretation of the law is not always optimal; to the contrary, sometimes optimality calls for aggressiveness in the face of substantive legal uncertainty.
This conclusion depends critically on the probability of detection being one--that is, the complete absence of detection uncertainty. If the probability of detection is less than one, this conclusion obviously does not follow, unless there is a penalty large enough to approximate the effect of detection certainty, as will be discussed at some length below. So let me be very clear that I am not advocating a regime that encourages or allows taxpayers to take tax positions that have a very low probability of success on the merits--which is one way of understanding what is normally meant by an "aggressive" tax position--unless we have in place a deterrence regime that either makes detection a certainty or imposes an ex post punitive penalty that has roughly the same ex ante effect as certain detection.
A second interesting observation that flows from the analysis above is that the strict tax liability rule works even for transactions that promise no pre-tax profit, that is, in circumstances in which the tax position in question--for example, the special deduction or credit--is the factor making the deal potentially profitable. Strict tax liability produces optimal compliance incentives not only in situations in which the transaction depends on the tax outcome for its overall profitability, but also in situations in which a transaction is expected to produce a pre-tax loss. These sorts of transactions are of course the source of much debate in the tax literature and are sometimes given pejorative labels such as "abusive tax shelters" or transactions that "lack economic substance." (45) Even for such transactions, if the probability of detection is 100%, and everyone knows this, a simple strict tax liability rule optimizes taxpayer compliance incentives.
The Joe Taxpayer example can be tweaked slightly to illustrate this last point. Assume now that the transaction is expected to lose $5 before taxes but promises the possibility of either producing a tax liability of $50 (for an after-tax loss of $55) or, perhaps by producing a credit or a loss deduction that can be used to offset taxes from some other source, a net tax savings of $100 (for an after-tax gain of $95). Given those possibilities, the break-even or optimal merits probability threshold for Joe's situation is roughly 0.3. Thus, Joe will be willing to engage in the transaction--and, from an efficiency perspective, should be willing to engage in the transaction--if the probability that this position will win on the merits is anything greater than 30%, but not otherwise. (46) Of course, the precise probability threshold is entirely an artifact of the numbers that I have chosen for the example, but the point is fairly general. Because the strict tax liability rule forces Joe to internalize the ex ante expected value of his ex post tax liability, he will invest in the transaction only if it is optimal to do so, so he will choose the optimal merits probability threshold, whatever that may be under the circumstances.
V. THE PROBLEM OF DETECTION UNCERTAINTY AND THE BENTHAM-BECKER SOLUTION
The preceding analysis assumed, among many other things, that the probability of detection--the combined probability that the taxpayer's return would be selected for audit and the particular issue in question would be scrutinized--was equal to one. This is an especially fanciful assumption in the tax context, at least with respect to the sort of sophisticated transactions that this article is focusing on--transactions that involve navigating the intended and unintended loopholes in the Code. For such transactions, the probability of detection is notoriously low. This is true both because the audit rate itself is far less than 100% and because, even when a return is audited, there is a good chance that such tax positions will go unnoticed by the Service. (47) Although no one knows what the precise probability of detection is for any given type of tax position (even the Service, which does have data on audit rates for different categories of taxpayers and has confidential information about its own enforcement priorities, cannot know what the ultimate detection probability is for a particular tax position), such probability is almost certainly very low for many of the transactions that characterize the zone of substantive legal uncertainty. The result, of course, is that a tax penalty regime of strict liability in which the only penalty is the "harm"--the additional taxes plus interest--woefully under-deters. Thus, with probability of detection significantly below one, unless we assume massive levels of risk aversion, taxpayers have an incentive to engage in transactions (and take tax positions) that are far too aggressive from an overall social welfare perspective. Indeed, the current problem of the tax gap, the difference between the taxes owed and the taxes paid, is almost certainly a result of the fact that the vast majority of taxpayers who underpay their taxes never get caught, and everyone knows that.
The effect of detection uncertainty on taxpayer aggressiveness can easily be illustrated by making a small but important change to the Joe Taxpayer example. Imagine the situation as described in the original example above (a $75 expected pre-tax profit with a potential tax liability of $100 or $0), except that the probability of detection is not 100% but is, for example, 1%. If the fine is then set equal to the amount of back-taxes owed with no additional penalty, the taxpayer would have an incentive to make the investment, in reliance on the tax position in question, even if the probability of winning the position on the merits were zero and that particular tax position were clearly illegal. (48) Indeed, in the current Joe Taxpayer example, Joe would have an incentive to make the investment, despite the position being unquestionably illegal, for any probability of detection less than 75%. (49) Again, the particular numbers are arbitrary, but the principle is well known among tax practitioners and analysts: the lower the probability of detection--all else being equal--the larger the incentive that taxpayers have to take aggressive tax positions. (50)
The problem of low probabilities of detection is one of the oldest and most thoroughly discussed issues in the entire deterrence literature. At least since Jeremy Bentham, theorists have understood the relationship between the probability of detection, the magnitude of the penalty, and the appropriate level of deterrence. (51) It was Gary Becker who first formalized this relationship in the criminal law context and who further identified the optimal solution to the deterrence problem, assuming that the cost of increasing the probability of detection--that is, money spent on detecting law violators--is a deadweight social loss. Becker's solution is to spend relatively little on detection, but to increase the ex post penalty until the potential perpetrator is induced ex ante to act as if the probability of detection were one. (52) Such a Bentham-Becker penalty is calculated by dividing the harm caused by the probability of detection. To illustrate, if a crime is expected to cause a harm of 100, such that the optimal cost-internalizing sanction would be $100, but the probability of detection is 0.01, the optimal fine would be $10,000--that is, the amount of the harm (100) divided by the probability of detection (0.01). Or, if you prefer multiplying, the amount of harm multiplied by the reciprocal of the detection probability. The theory is that such a penalty makes the expected value of the fine equal the harm. Thus, in general, so long as (1) the ex post adjudicator can accurately determine both the ex ante probability of detection and the amount of the harm (for purposes of this article, the actual taxes owed as determined by the adjudicator in resolving the legal uncertainty), (2) taxpayers or their advisors are aware of this fact, and (3) taxpayers have sufficient assets at risk to care about large ex post penalty, then the use of such an ex post penalty regime can create the proper ex ante tax compliance incentives. (53) Indeed, under such a regime, people should behave as if the probability of detection were equal to one and the fine equaled the harm. This same analysis has been applied to justify punitive damages in tort law and environmental law, among many other areas. (54) A similar analysis could be applied to tax law as well. According to such an analysis, under certain restrictive assumptions, the optimal fine for tax underpayments ought to be the amount of tax underpayment divided by the probability of detection.
To see the operation of this idea in the tax context, take the original Joe Taxpayer example but assume a probability of detection of 1%. In such a case, Joe's ex ante compliance incentive would be optimized by applying a strict liability standard that assessed an overall fine--including both the back-taxes and penalty--equal to the harm (the additional taxes owed of $100) multiplied by 100 (i.e., 1 / 0.01). Thus, in the event the tax position was detected and rejected, Joe would be required to pay not only the underpaid tax, but also a punitive fine (or a kicker) of $9900. (55) By adopting a rule that would impose such a fine in the event of an adverse determination, the strict-liability-plus-punitive-kicker rule forces Joe, ex ante, to internalize the expected tax liability associated with the transaction, which, as shown above, leads to the optimal tax compliance incentives even under conditions of legal uncertainty. (Again, this conclusion assumes Joe has at least $10,000 worth of assets that might be subject to the ex post penalty.) Recall that under the original example, Joe's optimal merits probability threshold was 25%. He should only make the investment if he assessed the likelihood of winning on the merits to be greater than 25%. Now, if the probability of detection were 1% rather than 100%, but the potential total tax liability were $10,000 rather than just $100, Joe would still have an incentive to make the investment (and claim the $100 tax benefit) only if doing so were socially efficient--that is, only if the 25% merits probability were exceeded. In sum, a strict liability rule with a Bentham-Becker punitive penalty can induce optimal ex ante tax compliance incentives in a world with substantively uncertain tax law and in which taxpayer aggressiveness regularly goes undetected.
Applying the strict-liability-with-punitive-kicker regime, however, raises a number of conceptual and practical problems. These are the ex post unfairness concern, the judgment-proof taxpayer problem, the over-deterrence problem, and the administrability problem. I discuss the first three in this part and the fourth in Part VI. Some but not all of these concerns will apply to both a strict liability and a fault-based regime. All of them are connected with, and even attributable to, the punitive-kicker aspect of the tax penalty regime proposed in the previous section.
First, even if we assume that taxpayers are perfectly rational and informed--and thus would be optimally deterred by a Bentham-Becker penalty--there is the view that the Bentham-Becker punitive penalty would create a kind of ex post unfairness because of the disparity between the size of the penalty and the magnitude of the offense. In the criminal law context, this complaint against Bentham-Becker penalties is often stated in terms of the punishment being out of proportion to the crime. Thus, for example, if an individual were to break the law and cause a social harm of $100, a sanction of $10,000 on that single individual would, on this view, be considered excessive. The same criticism could be made in the tax context. If a taxpayer underpays her income by $100, and the particular mistake has only a 1% chance of being detected, it seems intuitively unfair that the one person out of 100 who gets caught will have to pay $10,000 while the other ninety-nine go free. Such an outcome seems especially problematic when the activity in question, from an ex ante perspective, is not clearly illegal but is only of uncertain legality.
This injustice may be seen as further compounded by the fact that the size of the tax penalties imposed on taxpayers may in many cases reward dishonesty in the following sense: if the Service were to concentrate its enforcement efforts on identifying those taxpayers who are most likely to cheat on their taxes (a reasonable strategy), the probability of detection for those taxpayers would increase relative to the probability of detection of the taxpayers who are not so much the focus of the Service's scrutiny. The irony of this approach, however, is that the punitive-kicker would correspondingly be lower for those relatively dishonest taxpayers (the ones on whom the Service is focusing) and higher for the more honest taxpayers (to whom the Service gives less frequent or less intense scrutiny). Thus, if we loosely equate audit rates with detection rates, taxpayers subject to only a 1% audit rate because of their historically high levels of compliance would face a potential noncompliance penalty equal to 100 times their tax understatements; whereas the taxpayers subject to a 50% audit rate because of their historically low levels of compliance would face a potential penalty of only two times their understatements. Of course, audit rates are not necessarily inversely correlated with relative honesty. For example, it may be administratively or politically more feasible to audit certain taxpayers at higher rates than others. But the perception of unfairness could still be a problem. (56)
Besides the ex post unfairness problems, there is a potential deterrence problem as well: a rational taxpayer of the sort I have been assuming throughout this analysis (including an assumption of risk neutrality) would ignore the threat of any ex post fine that exceeds the amount of her assets that are available to satisfy a tax judgment. This fact--sometimes called the judgment-proof problem--limits the ability of large ex post fines to produce optimal ex ante compliance incentives. The problem is well known in the deterrence literature, and it suggests that even with a Bentham-Becker punitive penalty in place, if taxpayers are judgment-proof, they will be under-deterred. (57)
Interestingly, there is also an over-deterrence concern with the punitive-kicker penalty, although it requires that we assume risk aversion on the part of the taxpayers. Putting aside the judgment-proof problem for the moment, the prospect of large tax penalties in conjunction with a strict liability rule can actually over-deter, that is, induce taxpayers to under-invest in legally uncertain tax positions. The point is simple enough to understand: risk-averse taxpayers facing a potentially large ex post tax penalty in the event their uncertain tax positions end up losing in court might be deterred from taking such positions in the first place--even when taking such positions, although not certain to be upheld on the merits, is well above the efficient merits probability threshold. That is, they would be deterred from making investments that they should make. Thus, for example, when the likelihood of succeeding on the merits for a particular transaction is, say, 40%, we know that it would be efficient for the taxpayer to make the investment if the optimal merits probability threshold is less than 40%. (In the Joe Taxpayer example above, recall, that threshold was 25%.) If the taxpayer is risk-averse, the possibility of a $10,000 ex post tax liability might dissuade the taxpayer from going through with even a tax-efficient transaction, that is, even though the expected tax liability would be significantly less than the expected pre-tax profit. In that sense, then, if taxpayers are risk-averse, a full punitive-kicker penalty can over-deter and discourage efficient transactions.
Some readers may scoff at this over-deterrence concern. They might say, for example, that while over-deterrence may be a theoretical possibility, it is a miniscule concern in the real world of tax enforcement. The much bigger problem, the argument goes, is under-deterrence. The whole problem of corporate tax shelters, for example, which has occupied so many pages in Tax Notes, The Wall Street Journal, and The New York Times in recent years, is in essence a problem of under-deterrence. Moreover, the Treasury Department continues to report a substantial federal tax gap of roughly 16%--that is, federal tax revenues are estimated to be approximately 16% less than they would be if all taxpayers paid what they actually owe. (58) Thus, one might reasonably ask, how serious can the over-deterrence problem really be? My own view, based on very little evidence, is that at present the over-deterrence problem is probably small, though not nonexistent. Even now, I have heard anecdotal accounts of taxpayers who are not willing to engage in certain types of transactions that are tax sensitive because the tax law is too uncertain, and because the Service will not issue them a private letter ruling on the issue. I have no idea how large this problem is. The much more important point is that, even if it were the case that there is presently relatively little over-deterrence under the current tax penalty regime, we obviously do not currently have anything resembling a Bentham-Becker penalty regime. As discussed above, maximum penalties are usually limited to 20%, rarely higher than 75%, of the underpaid tax. These amounts are obviously far, far less than the penalties that would be the norm under a Bentham-Becker model, which again would be five, ten, or even 100 times the tax underpayment. If such large penalties were authorized, it is reasonable to suppose that it would not be very long before complaints of over-deterrence would dominate the tax news coverage.
VI. THE POTENTIAL ROLE OF (AND THE PROBLEMS WITH) TAX LIABILITY INSURANCE
Before we get to the administrability problems with the strict tax liability punitive-kicker regime, consider a possible, at least partial, solution to the ex post unfairness and over-deterrence concerns. These concerns could be reduced if we allowed (and the judgment-proof problem could be reduced if we required) taxpayers to purchase insurance against the possibility that a particular tax position on their return might be rejected by the Service and back-taxes and fines imposed. The insurance would cover the back-taxes, interest, penalties, and perhaps the legal fees as well. (59) What would the effect of such insurance be? In theory, assuming actuarially priced insurance, such an innovation would convert the large ex post fines that would threaten taxpayers under the Bentham-Becker kicker regime into something approximating the taxpayer's ex ante expected tax liability. This theoretical result would be as close to the ideal tax treatment of legally uncertain tax transactions as we can get, both from a deterrence perspective and a distributive justice perspective. To see this point, go back to the original example where the efficient merits probability threshold was 25%, owing to the $75 expected pre-tax profit and $100/$0 potential tax liability. Additionally, assume again that the detection probability for the transaction is 1%, such that the optimal punitive penalty is $9900, plus the $100 tax liability. Now imagine that Joe Taxpayer finds a particular transaction that fits this profile and that has a merits probability of 40%. Joe's expected tax liability is thus $60. (60) Because his expected after-tax profit is $15, he should clearly make the investment. For the critical point, assume further that there are 99,999 other taxpayers in the economy taking a tax position with the same payoff structure as Joe's, an expected tax liability of $60. Assume also that all of these tax positions are uncorrelated with one another, in the sense that when one uncertain position is resolved by the Service or a court ex post, that decision has no effect on how the other cases will be resolved. According to these assumptions, only 1000 of these taxpayers will be audited, 600 of whom will lose their cases and be required to pay the $100 in taxes plus the $9900 Bentham-Becker punitive fine. Again, such a rule produces the right ex ante incentives. The problem, according to the ex post unfairness complaint, is that 600 unlucky souls are required to pay $10,000 each, while the other 99,400 taxpayers, who are in exactly the same position and who took exactly the same risk, pay nothing. That result could be seen as distributively unjust.
Tax liability insurance can alleviate this injustice. Since each of the 100,000 taxpayers in this example has an expected tax liability of $60--each is taking a 0.006 chance of paying a $10,000 tax-plus-penalty liability--tax liability insurance, in its idealized form, would allow these taxpayers to shift the risk of a $10,000 payment to the insurer in exchange for paying the insurer a premium of roughly $60. If such insurance were provided, when the 1000 out of the 100,000 taxpayers were selected for audit, and 600 of those ended up losing their cases, the insurer would pay the $10,000 total tax liability for each of the 600 unlucky ones. (61) In so doing, the insurer would be acting in effect as a private ex ante tax collector, collecting premiums of $60 from each of the taxpayers in the pool in advance and then eventually paying those premiums over to the government (less an administrative fee).
Ideally, this type of tax insurance could reduce the over-deterrence problem as well as the ex post unfairness (or perceived unfairness) associated with the Bentham-Becker punitive sanction. The over-deterrence problem goes away because the threat of a large ex post fine goes away. The ex post unfairness concern is addressed through the operation of insurance risk pooling. So, unlike the Bentham-Becker penalty without insurance, where $6,000,000 was spread across 600 taxpayers, here $6,000,000 is spread equally among 100,000 taxpayers. Moreover, if we assume that the insurer could set and adjust premiums perfectly to correlate with the expected tax liabilities of its insured (this is what I meant above by the assumption of actuarially fair insurance), the strict liability rule with a Bentham-Becker penalty would still produce optimal ex ante compliance. Because taxpayers would still ex ante face the expected tax liability associated with the investment, they would be induced to make the efficient decision regarding when to invest in reliance on uncertain tax law.
This rosy picture ignores many complications that might make such tax liability insurance infeasible. One problem is more fundamental than the others: to the extent the tax insurer is not able to classify risks perfectly, there would be a degree of both ex post unfairness and inefficiency. Imagine that the insurer in the example above could not distinguish between a taxpayer who is engaging in a transaction with an expected tax liability of $60 and a taxpayer who is making an investment with an expected tax liability of $70 or $50. To the insurer, these all look the same. If that were true, there would be inefficiency, because taxpayers would not face their own expected tax liability. Some taxpayers who are insured would thus be induced to make investments that are excessively aggressive. For example, a taxpayer might represent to an insurer that it was going to engage in a transaction with an expected tax liability of $60 and then turn around and, in fact, engage in a transaction with an expected tax liability of $70. This is a version of the well-known problem of moral hazard that plagues all insurance arrangements to some extent. Likewise, those taxpayers who expected to be engaging in relatively tax-risky transactions (the $70 expected value transactions) would find the $60 premium to be a bargain and would be disproportionately likely to purchase insurance. This would force premiums up over time, causing the lower-risk taxpayers on the margin to opt out of buying insurance, thus leaving them uninsured and, again, potentially over-deterred. This is of course the tax-insurance version of adverse selection. When there is variance within insurance risk pools, there is concomitantly cross-subsidization from the low risk to the high risk insured, which arguably presents an ex ante unfairness concern.
Moral hazard, adverse selection, and cross-subsidization occur to some extent in all insurance markets and are not considered per se fatal to the enterprise. Insurers generally try to combat these problems in a myriad of ways, not the least of which is by engaging in ex ante efforts at risk classification and contractual protection. For example, a tax-risk insurer who is approached about covering a particular tax transaction might hire its own tax experts to evaluate the legal merits of the proposed tax treatment. This would help the insurer to decide whether to offer the coverage, at what price, and under what terms. To mitigate the problem of moral hazard--that is, the taxpayer who says she is planning a transaction with a $60 tax risk but then proceeds to do a transaction with a risk of $70--the insurer could, in advance of the transaction, require the taxpayer to provide detailed representations regarding the proposed transaction (details suggested by the insurer's tax experts), and the breach of these representations by the taxpayer would be grounds for the insurer to void the contract. Plus, insurers could use a combination of large deductibles and contractual policy exclusions to try to reduce moral hazard and adverse selection concerns. (62) Insurers have used these tools for decades in liability and other insurance markets and, interestingly, they are beginning to use them now in the small but growing tax-risk insurance market. (63)
Another serious problem with tax-risk insurance, which potentially confronts all premium-financed insurance arrangements, is the possibility of correlated risks. In the example above, I assumed that all 100,000 taxpayers were engaging in transactions that posed the same tax risk but that were uncorrelated with each other. Those are the perfect conditions for the insurance risk-spreading mechanism to work. I have already discussed what happens when the risks being insured have different expected values--the problem of insufficiently precise risk classification. But what happens if the tax risks are correlated? If the insurer has not reinsured the risk, and if the correlated risks represent a large fraction of the insurer's overall book of business, the result can be bankruptcy for the insurer. To see this, change the example so that all of the 100,000 taxpayers are doing precisely the same transaction, or they are doing transactions that turn on exactly the same uncertain legal question, such that if the legal uncertainty is resolved in one case, it gets resolved in all of them, virtually simultaneously. In that case, if those taxpayers' risks were the only risks that the insurer covered, and the insurer had no reinsurance and no large surplus of assets, it would not be able to price the policies. The insurer could not charge just $60 for the policies, because if a court were to decide against one of the taxpayers, there would not be sufficient funds to cover all of the claims. The insurer would have $6 million in premiums, but would need $100 million, since all 1000 taxpayers selected for audit would lose their cases. In addition, the insurer would not be able to charge much more than $60 because the taxpayer-insureds would not be willing to pay more. Certainly taxpayers would not be willing to pay a premium of $100 (equal to the taxes they hoped to avoid paying on the transaction), which would be necessary to give the insurer certainty that it would be able to pay the claims when they come in. (64)
This correlated-risk problem is potentially troubling, given that some uncertain tax issues are correlated with each other. This is especially true insofar as taxpayers tend to copy each other's tax transactions. However, the correlated-risk problem seems ultimately unlikely to prevent a tax-risk insurance market from arising if a Bentham-Becker punitive-kicker tax penalty regime were adopted. Why? First, there are many different sources of transactional tax law uncertainty, almost as many as there are sections in the Code. Second, there would rarely be perfect correlation of risks, even when taxpayers attempt to engage in very similar transactions. Often the tax treatment will turn on highly fact-specific issues that will be peculiar to the particular transaction and particular taxpayer. For that reason, it would be surprising if an insurer could not put together a portfolio of tax law risks that were relatively diversified or uncorrelated with each other. Moreover, insurers can, and usually do, reinsure their risks with other, larger insurers that have larger and more diversified portfolios. For the very large insurers, since their tax insurance business would be only a small part of their own portfolio, this diversification function would be performed largely in house.
One complaint that gets lodged against allowing tax-risk insurance in the current regulatory climate is that the availability of tax insurance will exacerbate an already existing under-deterrence problem. (65) The argument is that, given the trivial penalties that currently exist for tax noncompliance, the existence of tax law uncertainty and taxpayer risk aversion are necessary to prevent the tax gap from increasing dramatically. Or, if risk-averse taxpayers are allowed to insure their penalties, competitive market forces will eventually motivate insurers to offer coverage for transactions that have zero merits probability, in which the only risk they would be insuring would be the risk of detection. Put differently, the worst case scenario with respect to tax-risk insurance would be insurance for clearly illegal transactions. I agree that this is a concern under the existing penalty structure, and in another article I have suggested some possible solutions, such as requiring taxpayers who purchase tax-risk insurance to report this fact to the Service, thus raising the detection risk, and hence the expected tax liability, for such transactions significantly. (66) This problem of tax shelter insurance, however, would not likely arise--or would be much diminished--under a strict liability penalty regime with a Bentham-Becker punitive penalty. If the penalty is set properly, the insurance premium for the transaction will equal the expected tax liability. In other words, for a transaction with a merits probability of zero, the insurance premium would equal the potential tax liability, making the deal not worth doing for the taxpayer. (67) Thus, if in our example the merits probability were zero, the insurer would charge the taxpayer a premium of $100, efficiently deterring the transaction.
An alternative to allowing risk-averse taxpayers to purchase tax liability insurance from commercial insurance companies would be to allow them to purchase such insurance directly from the government. That is, each of the taxpayers in the example above could pay the U.S. Treasury, instead of an insurance company, $60 and in exchange receive a commitment that the issue in question would not be challenged. This approach seems similar to a private letter ruling, but one where taxpayers are required to pay the expected tax liability in advance. Perhaps the better analogy would be to a settlement agreement between a taxpayer and the Service on some issue that the Service has challenged but with respect to which the legal outcome is uncertain. Interestingly, under either approach to dealing with tax law uncertainty--the private insurance approach and the government insurance (or settlement) approach--the key facts are that the taxpayers would face approximately their expected tax liability when making tax-sensitive investment decisions, the government would get the right amount of revenue under the circumstance ($6 million in our example involving 100,000 taxpayers), and the tax burden for these types of transactions would be allocated fairly across all taxpayers doing the deal ($60 each). Whether government insurance or private insurance would be the better mechanism for this is beyond the scope of this article. The essential point to recognize here would be that, under present assumptions, including the assumption of a strict liability rule with an ex post Bentham-Becker fine, either would optimize ex ante compliance and tax decision-making under conditions of uncertainty.
The most serious difficulty with any attempt to adopt a full-fledged punitive-kicker tax penalty regime on the Bentham-Becker model is not the ex post injustice or the over-deterrence problems or the problems of imperfect insurance markets. Rather, the most troubling concern is the administrability of the penalty provision itself. In the analysis thus far, it has been assumed that the ex post adjudicator, the Service or the court, would be able to determine with perfect accuracy not only the correct answer to the tax question at issue but also, for the purpose of calculating the penalty, the particular ex ante probability of detection for the particular taxpayer and tax issue under scrutiny. Such an assessment is obviously necessary to calculate the precisely correct cost-internalizing penalty. But making such an assessment, at least with any degree of precision, would be impossible. That impossibility may well be why the current tax penalty regime falls so far short of the Bentham-Becker ideal, with most penalties capped at 20% of the underpaid tax. Indeed, some commentators have cited this problem as a reason not even to attempt an ex post punitive penalty approach to dealing with the low-probability-of-protection problem. (68) Moreover, this problem is connected specifically to the punitive-kicker penalty and therefore would apply whether the tax liability rule in place were strict liability (as I have been discussing thus far) or fault-based (to be taken up below).
I agree that neither the Service nor any court evaluating the merits of a particular tax position would be able precisely to calculate the ex ante probability that such a position would be detected. Still, some rough approximation might be possible. For example, the ex post adjudicator, whoever it might be, could begin by relying on the publicly available audit rates for various classes of taxpayers. (69) Such audit rates would almost certainly overstate the probability of detection for many sorts of tax issues that involve substantial legal uncertainty, given that many such issues avoid detection even on audit. However, a penalty based on audit rates would come much closer to the Bentham-Becker ideal than the current regime. Further, as a means of increasing the accuracy of the penalty for the taxpayer in question, the adjudicator might be allowed access to the Service's confidential information regarding audit strategies and audit probabilities. For still further refinement, the adjudicator could then consider evidence specific to the individual or corporation before it. Moreover, if we are worried about giving this much enforcement discretion to the Service or to the courts--that is, the discretion to determine the size of the ex post tax penalty based on all of these factors--Congress could enact a schedule of punitive fines that would be based on audit rates and that could be somewhat tailored to the class of transaction at issue.
By offering these responses to the administrability problem, I do not mean to suggest that this concern should be overlooked or taken lightly. In my view, it is very likely that this concern may be sufficient reason not to adopt such a regime. However, the question is at least worth further study and should get more attention in the literature than it has received.
To summarize the analysis thus far, under the assumptions laid out in Part III above, when the law is substantively uncertain--in the sense that there is a positive, but less than one, probability of success on the merits of the tax position at issue--the optimal tax penalty regime would involve a strict liability rule. Indeed, under present assumptions, a strict liability rule works when the substantive law is certain as well. The optimal level of penalty, however, depends on the probability of detection. If detection is certain, there is no need for a penalty in excess of the taxes owed plus interest. If detection is uncertain, as it always will be in the cases of interest to the present analysis (ones that involve substantive legal uncertainty), the penalty should approximate the back-taxes owed plus interest divided by the probability of detection. That is the famous Bentham-Becker kicker. Such a regime would induce the taxpayer to make optimal decisions regarding whether and under what circumstances to rely on uncertain tax law. One conclusion made clear by my analysis is that there is no a priori appropriate or efficient merits probability threshold or minimum. Rather, how certain the taxpayer ought to be about the substantive law before relying on it will depend on a number of factors, including the size of the potential tax consequences and the potential pre-tax profit expected from the investment. If we are concerned about the potential unfairness and over-deterrence effects of very large potential tax penalties--which could be in the neighborhood of five, ten, or even 100 times the underpaid tax--then tax penalty insurance could be offered by private insurers or the government. Indeed, if a Bentham-Becker penalty regime were enacted, such an insurance market would likely arise on its own unless prohibited by law and whether and how to regulate that market would be the resulting questions.
The preceding analysis looked exclusively at a strict liability approach to tax penalties. Under such a regime, when a taxpayer is singled out for enforcement and has a tax position rejected by the adjudicator, not only is she held strictly liable for any additional tax she is found to owe, but she is also automatically liable for the underpayment penalties as well, penalties that ideally (ignoring all of the caveats discussed above) would be set according to the Bentham-Becker formula. The other option of course involves some form of fault-based or negligence standard for assessing penalties. Indeed, the fault-based alternative is deserving of special attention, given that the current tax penalty regime employs a fault-based approach. The obvious questions are whether the fault-based approach could, like the strict liability approach, create optimal ex ante tax compliance incentives, and, if so, which regime could do so at lower cost. Relatedly, does the fault-based approach face drawbacks similar to those discussed above. In the next part, I explain two fundamental problems with using a fault-based standard for tax penalties, at least if the standard of fault is tied to the question of whether the taxpayer satisfied the optimal merits probability threshold. Then I explain why, notwithstanding these problems, there is a second-best argument for adopting, or (more accurately) continuing to use, some form of fault-based approach to tax penalties.
VII. ASSESSING A FAULT-BASED TAX PENALTY REGIME
A. Defining the Fault-Based Standard: Applying the Optimal Merits Probability Threshold
If a taxpayer takes a position on her tax return that relies on a substantively uncertain interpretation of the tax laws, and the Service and the courts decide that the taxpayer happens to be wrong, then, as to the assessment of the underpaid taxes plus interest, the rule almost by definition has to be strict liability. That is, if the ex post adjudicator determines that a taxpayer owes more taxes than she paid (recall that we are assuming that the Service and courts resolve this uncertainty definitively ex post), then that assessment is final. The taxes must be paid. Joe Taxpayer cannot get out of paying his taxes simply because the law, ex ante, was uncertain, even if his interpretation of the law was reasonable. (70) At bottom, the tax law is distributive in nature, and the adjudicator is resolving the uncertainty as to what the distributive burden of the tax laws ought to be. (71) The interesting question involves the penalty, the amount imposed on the taxpayer over and above the underpaid tax liability. The analysis to this point has been focusing exclusively on the possibility of imposing such penalties on the basis of strict liability, which is a term and concept borrowed from tort law. The obvious alternative to a strict liability approach to tax penalties is a fault-based approach.
What would an idealized fault-based tax penalty regime under conditions of substantive legal uncertainty look like? At the most general level, a fault-based tax penalty regime can be described as follows: if the Service and the court determine that a taxpayer has, in effect, caused "harm" by taking a tax position that turns out to have been wrong, then whether or not the taxpayer will be required to pay a penalty (in addition to the back-taxes plus interest that is by assumption owed) will depend on whether the taxpayer's position was in some sense "reasonable." (72) What constitutes a "reasonable" tax position under conditions of substantive tax law uncertainty is the interesting question. As it turns out, it is also a surprisingly difficult question to answer, even as a conceptual matter. That is to say, although it is simple enough to apply the torts concept of strict liability in the tax context (since strict liability is pretty much the same across all contexts--that is, the offending (or injuring) party is forced to pay the harm she causes plus some additional penalty in cases of detection uncertainty), the same cannot be said of the negligence standard.
There is in fact a negligence penalty in the income tax. (73) And the definition of negligence in tax law owes an obvious debt to tort law. The negligence penalty in tax is imposed for any tax underpayment resulting from the taxpayer's lack of "due care" or failure to do what "an ordinarily prudent person would do under the circumstances." (74) The precise meaning of due care in this context is not clear. Of course, the same can be said of the due care standard in tort law, where the issue of what constitutes a failure of ordinary prudence will typically be determined after the fact by a judge or jury applying an ad hoc analysis that takes into account all of the relevant facts and circumstances. Some conceptual clarity has been brought to the due care standard in torts, with the famous Learned Hand cost-benefit test, but even that seemingly straightforward analysis presents conceptual difficulties. The Learned Hand rule, announced in United States v. Carroll Towing, says that tort defendants will be found to have been negligent--that is, found to have violated the "due care" standard--if it can be shown that they failed to take precautions that would have reduced expected accident costs by more than the cost of the additional precaution. (75) Under Hand's formulation, a party is negligent if the burden of avoiding the accident (B) was less than the product of the probability of the accident (P) and the potential liability resulting from the harm (L) and the party failed to undertake B. (76) Although the Learned Hand standard has been influential among courts and commentators on tort law, it has also generated considerable controversy. For example, some argue that the information that the Learned Hand standard requires is not available to the factfinders who are asked to do the analysis. I will make a similar argument about applying a Learned Hand type standard in the tax penalty context.
Before we get to that criticism, however, what would such a test even look like in the tax context? One possible approach to a Learned Hand type test for penalties--though, as discussed more fully below, not the approach used in the Code--would be to apply a version of the optimal merits probability threshold developed in previous parts of this article. To see how this would work, imagine that the taxpayer has taken a particular tax position on her return, the position has been scrutinized by the Service, and the position has been found wanting; the Service and the court have determined that the taxpayer is not in fact entitled to the deduction or credit she claimed, or must include some item of income that she excluded, and thus that back-taxes and interest are owed. The Service or the court would then ask whether the particular tax position in question, at the time the return was filed, had a probability of success on the merits that exceeded the optimal merits probability threshold for that particular position, taking into account the expected pre-tax profit and the potential tax consequences for that particular transaction. If so, the taxpayer's position would be considered reasonable, or non-negligent, and no penalties would be assessed. If, however, the ex post adjudicator determined that the tax position in question was ex ante below the optimal merits probability threshold for that transaction, the taxpayer would be required to pay the penalty. In other words, the taxpayer would in that case have taken an unreasonable tax position. (77)
To get a sense of how this probabilistic fault-based standard might be applied, return to the Joe Taxpayer example, with $75 expected pre-tax profit, a potential tax liability of either $0 or $100, and a detection probability of 0.01. For this transaction, the optimal merits probability threshold is 25%. That is, it is socially cost-justified for Joe to make the investment if the likelihood of success on the merits is equal to or greater than 25%, but not otherwise. Now imagine that Joe, after doing the requisite legal research (or having his lawyer do it), learns that the particular transaction has a 40% chance of prevailing if detected and he makes the investment. He happens to get selected for audit, however, and ultimately loses the case; the court says Joe owes another $100 of tax, which he does (as we are assuming that the courts are never wrong). Does he owe a penalty as well? Not according to the fault standard. Joe made a reasonable--by which I mean, socially cost-justified--estimation of the law under the circumstances and would have made the same choice if detection had been certain and the penalty equal to the harm. Therefore, under this theoretical tax penalty standard, Joe would owe no penalty. If, however, the probability of success on the merits for the transaction had been, say, 22%, when Joe made his investment, got caught, and lost his case, he would have been required to pay the punitive penalty as well--the additional $9900. Deterrence theory tells us that the punitive penalty should be calculated according to the Bentham-Becker formula, even for a fault-based regime. With such a probabilistic fault-based penalty in place, taxpayers would, under current assumptions, be deterred from taking unreasonably aggressive tax positions under circumstances in which the law is uncertain. This result is analogous to the conclusion reached in the economic analysis of the negligence rule in torts. (78)
B. The Problems with a Probabilistic Fault-Based Tax Penalty Regime
Some of the problems with a probabilistic fault-based tax penalty regime are similar to the problems with a fault-based tort regime while some are unique to the tax context. This section discusses the latter type. Part VII.C below, however, resurrects the case for a modified version of the fault-based standard.
So what's wrong with the above described idealized form of a fault-based tax penalty regime? The most obvious concern, and ultimately perhaps the most troubling, is the problem of administrability. How is the Service or a court, looking at a tax position that it has decided lacks merit, supposed to determine the ex ante optimal merits probability threshold for that particular transaction? (79) We have thus far assumed that courts get everything exactly right when they resolve the tax law uncertainty ex post, but to expect a court to do this analysis seems excessively optimistic. Further, the problem seems more challenging than the problem (discussed earlier) of determining the ex ante probability of detection for the purpose of calculating the appropriate punitive penalty, which again would be required under both the strict liability and fault-based approaches. In assessing ex ante probabilities of detection, at least the court can rely initially on concrete information that is in the hands of the government, such as audit rates, as a starting point for the analysis. With the probabilistic fault analysis, however, the ex post adjudicator must determine the optimal merits probability threshold for this tax transaction, or, more precisely, that probability at the time the transaction was entered into. This means figuring out the expected pre-tax profit from the deal and the potential tax consequences, as well as making an ex post assessment of the ex ante likelihood that the taxpayer would ultimately succeed on the merits. As difficult as this last element of the fault-standard sounds (determining the ex ante probability of success on the merits), the Service and courts today are in fact asked to do this analysis when applying the various understatement penalties under existing law. In cases in which a taxpayer's reporting position is challenged and ultimately overturned by a court, the penalty phase of the analysis requires the court to make an ex post guess of the ex ante strength or weakness of the taxpayer's legal position (literally cast in terms of probability of success on the merits). (80)
Even if a fault-based approach to tax penalties could be made administrable (a topic I will discuss below), this approach still may not be superior to the strict liability tax penalty regime, at least if we continue with the present assumptions. Both types of penalty regimes, by assumption, would entail the ex post analysis of the probability of detection, in order to set the proper Bentham-Becker penalty. However, only the fault-based approach would also require a second costly application of the appropriate merits probability threshold. One possible response to this observation is that, if we return to the standard deterrence assumptions (perfect rationality, no judicial error, etc.), a fault-based approach might be superior because, for game-theoretic reasons, the penalty would never actually have to be imposed. Given the potential liability that the taxpayers face, and the all-or-nothing nature of the fault standard (that is, if you satisfy it, you avoid all of the penalty--all $10,000 in our example), taxpayers would have an incentive to be at least efficiently conservative, in the sense of only making investments with merits probabilities greater than the efficient threshold that the Service would apply to their conduct. As a result, again for game-theoretic reasons, all taxpayers would have an incentive to satisfy the fault standard and hence in theory there would be no need for the penalty to be imposed. To put this point differently, under a fault-based tax penalty regime, taxpayers who are taking advantage of uncertain tax rules can, in effect, insure themselves against the risk of a large tax penalty (should the uncertainty be resolved against them) by taking only reasonable tax positions--only positions that fall within the relevant merits probability threshold. Assuming that courts never make mistakes in their ex post penalty assessments, and taxpayers never make mistakes in calculating the relevant ex ante merits probability threshold, the decision to take only reasonable tax positions would be (almost) equivalent to the purchase of tax-risk insurance. The result is an increase in social welfare, as risk-averse taxpayers would bear less risk. (81) Interestingly, in the general economic deterrence literature, this observation--the fault-based system's ability to reduce risk-bearing because the sanction is never used--is cited as an independent reason, though not necessarily an overwhelming reason, to prefer the fault-based approach over the strict liability approach. (82)
Beyond the administrative cost arguments already discussed, there are two fundamental problems with the use of a fault-based tax penalty standard. The first--the activity-level problem--is well known in the deterrence literature and applies to the use of a fault-based standard in any context. (83) The second--which for lack of a better term I will call the distributional problem--has not previously been discussed and applies only to the use of a fault-based standard in the context of tax penalties.
The activity-level problem is straightforward: to the extent that taxpayers comply with the fault standard, and again assuming courts always reach the right result, taxpayers can be sure that they will not face a tax penalty. This fact, as mentioned above, reduces the risk that they bear. It also means, however, that they are not forced to internalize the cost of those tax positions that happen to turn out to be wrong--wrong but reasonable. As a result, they will have an incentive to engage in the activity beyond the point at which it is socially cost-justified for them to do so. We can see this point in the example from above. The taxpayer is considering a transaction with an expected pre-tax profit of $75 and a 60% chance of causing a $100 tax liability. Under a strict liability approach, she would face an expected tax liability of $60, either because of the expected ex post fine or because of the ex ante insurance premium. Therefore, the taxpayer would make the investment so long as the $15 profit from the deal would be superior to whatever after-tax profit she could get from some alternative use of the investment, but not otherwise. With the fault-based approach however, assuming the merits probability is at least greater than 25%, the taxpayer's expected after-tax profit from making the investment would be $75. Thus, she would engage in the transaction so long as there were no other similar investment that produced more than a $75 after-tax profit. In sum, the taxpayer would be induced to engage in the questionable transaction even when it is not socially cost-justified. (For example, she would invest in this transaction even if there were another transaction with an expected after-tax profit of, say, $70.) More generally, with a fault-based tax penalty regime, there would be an incentive to over-invest or invest too often in legally uncertain tax transactions. This problem does not exist with strict liability, where the taxpayers are forced to internalize the expected tax liability associated with their investments. The relative superiority of strict liability over fault-based standards in dealing with activity-level issues on the part of "potential injurers" is well established in the deterrence literature, and the same arguments would seem to apply in virtually any deterrence context. (84)
The distributional problem with a fault-based tax penalty seems to apply uniquely, or at least especially, to the tax context. (As far as I know, this problem has never been noticed before.) The problem arises insofar as the fault-based penalty eliminates or reduces the actual imposition of the tax penalties. Recall the observation above that, under ideal circumstances, taxpayers are induced to satisfy whatever threshold of reasonableness is set for them by the courts such that no penalties are assessed. Again, in non-tax areas, this fact is cited as one of the advantages of a fault-based standard because it means there is no risk-bearing. The difficulty is that, in the tax area, there is a distributional reason that we want the penalties to be imposed. The failure to impose the penalties actually reduces social welfare, for distributional reasons, not deterrence reasons.
The best way to illustrate this point is by example. Consider our recurring hypothetical, the one with the 100,000 taxpayers all engaging in a tax transaction with a 60% chance of producing a $100 tax liability, and hence the expected tax liability of $60 per taxpayer. Hence, the transaction has 40% chance of winning on the merits. Under the fault-based standard described above, all of the 100,000 taxpayers would have engaged in this hypothetical transaction because the 40% chance of prevailing on the merits would be deemed a "reasonable" position by a court. This is because the 40% probability surpasses the optimal merits probability threshold for this transaction (which, recall, was 25%). According to the assumptions of the example, then, 1000 of those taxpayers would be audited and 600 of those would lose their cases on the merits. For those 600 taxpayers, the fault-based penalty analysis would be applied and the court would inquire as to the reasonableness of the position taken. As noted, the answer would be that the position was reasonable. Thus, none of taxpayers would be assessed a penalty, and the total amount of tax revenue collected from these taxpayers for this transaction would be $60,000-$100 of taxes multiplied by 600, the number of taxpayers who were audited and lost their cases. So what's wrong with this scenario?
The problem is that, in terms of probabilistic distributive justice, $60,000 is approximately $5.94 million too little to be taxing this group of 100,000 taxpayers who are engaging in a series of transactions that almost certainly are collectively producing an income tax liability of $6 million. That is to say, if it were feasible and cost-justified to audit all 100,000 taxpayers who engaged in this transaction, approximately 60,000 of the taxpayers would be found to owe $100, and the other 40,000 taxpayers, nothing. That is what we meant above when we assumed that the 100,000 transactions in question had a 40% chance of prevailing on the merits. Of course, auditing all 100,000 taxpayers is by assumption not feasible. The question therefore is: what can the tax system do, in terms of allocating the tax burden consistently with society's distributional values or preferences? The answer may be somewhat surprising, given that every one of the 100,000 taxpayers engaged in a tax transaction that created an expected tax liability of $60; the best the income tax system can do might be to collect something close to $6 million from the group, getting $60 from each of the 100,000, since they are all equal in the eyes of the law with regard to these transactions. That is, as among the 100,000, the principle of horizontal equity would suggest that each should pay $60 in income taxes for engaging in the transaction in question. (85) Because a fault-based standard can, for game-theoretic reasons, result in the imposition of no penalties (again, assuming all taxpayers rationally decide to comply with the standard), only the strict liability tax penalty system, which imposes the penalty on the few audited taxpayers found to owe taxes, would achieve this distributively desirable result. That is, only the strict liability tax penalty regime forces taxpayers who make investments in reliance on uncertain (sometimes very uncertain) tax positions to bear the expected income tax liabilities associated with those positions. (86)
Note how this result differs from the result in a nontax area of law, such as torts, where we do not usually think of the law as being explicitly distributive in nature. If we imagine tort law being primarily about deterring accidents, rather than about trying to achieve a particular distribution of income, we do not care whether the penalties are ever imposed. So long as, say, drivers are encouraged to drive carefully or manufacturers are induced to make safe products, there is no independent reason to require the payment of fines or damages. The victims of accidents caused by reasonably safe driving or reasonably safe products can be compensated much more efficiently through their own first-party insurance companies than through penalties assessed by the government. But conventional wisdom suggests that tax law is different. (87) Tax law, at least the income tax, is primarily about (1) raising revenue to spend on public goods and (2) allocating the tax burdens in a manner consistent with our society's vision of distributive fairness, whatever that vision happens to be. Hence, when Congress decides that the tax system should collect a certain amount of money from taxpayers who meet a particular description, and thus satisfy certain criteria set out in the law, then failing to collect the tax from those individuals means Congress has to raise the money some other way, such as increasing the national debt or raising rates for everyone. The problem is that either of these latter options produces a distributive result that is different from what Congress intended and from the social optimum. These divergences from the optimal distributive outcome represent losses of social welfare to the same extent as do distortions of labor choices caused by various tax rules. (88)
In sum, the main problems with a probabilistic fault-based tax penalty regime are that such a regime (1) is relatively difficult to administer, (2) owing to the activity-level effect, may result in far too many uncertain (albeit "reasonably uncertain") tax positions being taken, and (3) fails to allocate the tax liabilities associated with those legally uncertain transactions to the group of taxpayers who engag