|
Article Excerpt
INTRODUCTION
The individual income tax system in the United States operates on a self-assessment basis, in which individuals annually determine their tax liability and pay what they deem due. Over time, a taxpayer learns more about the tax system, so that reports today may depend in part on reports in the past. An individual can also communicate with or be influenced by other fliers in choosing how much to report, so that her/his reporting decision may depend in part on the decisions of others. Put differently, insofar as a taxpayer may recall his or her own past filing experience and may communicate in some ways with other taxpayers, these issues help shape the taxpayer's current report. Thus, two phenomena emerge from past reporting experience and from the exchange of information and experiences with other filers, one relating to dynamic effects (or persistence) in the taxpayer's reporting decision and the other relating to spatial dependence (or interdependence) across all taxpayers.
Dynamic effects suggest that this year's evasion decision is affected by past evasion experience. There is, therefore, an element of persistence in individual income tax evasion decisions. As suggested by Dubin (2007), this may also be due to the delayed audit completion cycle. He argues that "... taxpayers may adjust their reported taxes based on a mixture of taxes reported in the previous year and the optimal level
There are several possible explanations for spatial dependence in tax evasion. Individuals typically exchange their experiences with others, so they influence and are influenced by the tax evasion behavior of other taxpayers. (1) Further, if one taxpayer successfully evades because he or she is not audited, then (given the audit resources available to the tax administration) this may increase the probability that another taxpayer will be audited. Still another explanation may be due to a "social norm" of compliance. A social norm can be distinguished by the feature that it is process-oriented, unlike the outcome-orientation of individual rationality (Elster, 1989). A social norm represents a pattern of behavior that is judged in a similar way by others and that, therefore, is sustained in part by social approval or disapproval: if others behave according to some socially accepted mode of behavior, then the individual will also behave appropriately, but if others do not so behave, then the individual will respond in kind. Consequently, an individual will comply as long as he or she believes that compliance is the social norm; if noncompliance becomes pervasive, then the social norm of compliance disappears. (2) Manski (1991) and McFadden (2006) suggest still another, related explanation for interdependence. Both argue that individuals faced with dynamic stochastic decision problems that pose immense computational challenges may look to others to infer satisfactory policies, so that interdependence may come simply from imitating others, perhaps through social networks.
These twin issues of persistence and spatial dependence are seldom raised in the theoretical or empirical analyses of tax evasion, at least not in tandem. This is our purpose in this paper. We first extend the original Allingham and Sandmo (1972) model of income tax evasion by incorporating both issues. We then test their empirical validity in the context of the U.S. federal individual income tax evasion. We use state-level, time-series, cross-section data for the years 1979 to 1997, collected from a variety of sources, to estimate the factors that affect annual per return evasion in the individual income tax. The estimation methods use several econometric models that incorporate both spatial dependence and dynamic considerations and that also consider the potential endogeneity of the audit rate. When estimation methods appropriately consider these issues, the empirical results indicate strong and robust support for both spatiality and persistence in tax evasion. The results also show a large deterrent effect from higher audit rates. (3)
In the next section we briefly review past empirical studies conducted on individual income tax compliance. We then extend the Allingham and Sandmo (1972) model by incorporating persistence and spatiality. The following section deals with data and related methodological issues, including the methods employed in the empirical work, analytical issues related to the construction of variables, and descriptive statistics. We then present our estimation results, first examining the spatiality and persistence issues separately and then suggesting and implementing a way to combine them. In the final section we summarize our findings and discuss their implications.
SOME PREVIOUS EMPIRICAL STUDIES
We focus here on those empirical studies that directly link to our twin points of departure, especially those that use Internal Revenue Service (IRS) time-series data, and we confine ourselves to individual income tax evasion only. (4)
The first time-series studies are by Dubin, Graetz, and Wilde (1987; 1990), who estimate the determinants of income tax reporting as a function of audit rates and various socio-economic variables, using state-level, time-series, cross-section data from the Annual Report of the IRS for the years 1977-1985. Both studies recognize the potential endogeneity of the audit rate, and accordingly use the two stage least squares (2SLS) method to control for endogeneity. Both find that the audit rate is endogenous; after controlling for this endogeneity, both find that higher audit rates often have a positive impact on income tax reporting but one that varies by audit class and one that is not always statistically significant. (5) These studies also find that there is a spillover effect from tax audits; that is, taxpayers who are not themselves audited pay more in taxes when audit rates increase.
Plumley (1996) extends the analysis of Dubin, Graetz, and Wilde (1987; 1990), using state-level data from 1982 to 1991. He finds that criminal investigation enforcement activities are significant and positively related to compliance. Dubin (2007) also builds on these analyses by using state-level, cross-section, time-series data from 1988 and 2001 to examine the effects of criminal investigation enforcement activities on taxpayers' compliance behavior. Using a dynamic panel estimation method, he concludes that criminal investigation activities often have a measurable and significant effect on voluntary tax compliance; he finds that incarceration and probation have the most significant effect on reported taxes, while sentenced cases and media attention do not have a significant influence. (6)
In related work, Witte and Woodbury (1985) and Dubin and Wilde (1988) use Taxpayer Compliance Measurement Program (TCMP) data for the single year 1969, aggregated to the three-digit zip code level and broken down by audit class, to estimate the impact of audit rates on a predicted measure of noncompliance. Witte and Woodbury (1985) do not control for potential endogeneity of the audit rate in the three-digit zip code location, while Dubin and Wilde (1988) use 2SLS methods to control for endogeneity. Their results are quite different: Witte and Woodbury (1985) find a significant deterrent effect of audits on noncompliance, while Dubin and Wilde (1988) find that the audit rate has a significant impact in only one of the seven audit classes. See also Jou (1992). (7)
In sum, these past studies, even without always confronting the issues of dynamics and interdependence, present conflicting evidence as to the relative importance of sanctions and audit rates on tax compliance. Enforcement is negatively related to evasion in most all theoretical models of tax compliance, but audit rates are often statistically insignificant in empirical studies. Importantly, there are no empirical studies that address the twin issues of dynamics and interdependence. In the next sections, we develop a simple theoretical model that incorporates both issues, and we then test empirically their importance.
INDIVIDUAL INCOME TAX COMPLIANCE WITH DYNAMICS AND INTERDEPENDENCE
Let us start with the Allingham and Sandmo (1972) model of income tax evasion. Consider an individual i whose true income at time t, [[GAMMA].sub.it], is known only to him or her but not to the tax authority. Income taxes are levied at a constant rate [[tau].sub.it] on declared income. With some probability [p.sub.it],, taxpayer i will be subject to investigation by the tax authority in time t, and, if found to have underreported income, the taxpayer will be subject to a penalty rate [[theta].sub.it] ([[theta].sub.it] > [[tau].sub.it]) on the evaded income [E.sub.it]. To incorporate dynamics and interdependence, taxpayer i in evading the amount [E.sub.it] from the tax authority in period t takes into consideration the amount successfully evaded in the previous period [E.sub.it-1] by himself or herself, as well as the amount [E.sub.jt] contemporaneously evaded by another, "average" taxpayer j ([not equal to]i). The IRS audit rule is assumed to be one in which the decision to audit individual taxpayer i this year depends on how much that individual i evaded last year as well as on how much another individual j evades this year. (8)
Individual i's after-tax income [W.sub.i] is state dependent, and is given as
[1] [W.sub.i1] = [y.sub.it] - [[tau].sub.it]([y.sub.it] - [E.sub.it])
if not caught cheating, and
[2] [W.sub.i2] = [y.sub.it] - [[tau].sub.it] [y.sub.it] - [[theta].sub.it][[tau].sub.it][E.sub.it]
if caught cheating.
The individual is assumed to maximize von Neumann-Morgenstern expected utility. Taxpayer i then chooses [E.sub.it] to maximize
[3] E([U.sub.it]) = (1 - [P.sub.it]([E.sub.it-1][E.sub.jt]))[U.sub.it]([y.sub.it] - [[tau].sub.it] ([y.sub.it] + [P.sub.it] ([E.sub.it-1][E.sub.jt]) [U.sub.it] ([y.sub.it] - [[theta].sub.it][[tau].sub.it]([- [E.sub.it]))
where E is the expectation operator, and [p.sub.it] ([E.sub.it-1] [E.sub.jt]) implies that evasion of income by individual i at time t is conditional on his or her own evasion in the previous period together with the contemporaneous evasion by the average taxpayer j.
If conditioning of [p.sub.it] on ([E.sub.it-1] [E.sub.jt]) is ignored, then the familiar Allingham and Sandmo (1972) first-order condition can be derived as
[4] (1-[p.sub.it] (*))[U'.sub.it]([W.sub.i1]) = [p.sub.it](*)[[theta].sub.it][U'.sub.it]([W.sub.i2]),
where a single prime on [U.sub.it]() shows the first partial derivative of [U.sub.it] with respect to income and [W.sub.i1] and [W.sub.i2] are after-tax incomes in the two states of nature as defined by equations [1] and [2]. In this case, all of the basic results of the original model follow. However, if the probability of being audited is conditional on ([E.sub.it-1] [E.sub.jt]), then this greatly complicates the analysis of the optimal choice of [E.sub.it], since both [E.sub.it-1] and [E.sub.jt] are now arguments on both sides of the first-order condition [4]. The influence of [E.sub.it-1] or [E.sub.jt] on [E.sub.it] can be obtained by totally differentiating [4], treating all other parameters constant, to give where the double prime on [U.sub.it] shows the second partial derivative with respect to income. From the second-order condition characterizing the optimal choice of [E.sub.it], the first term in the denominator of the right hand sides of [5] and [6] must be negative. Also, the first term in the numerator of the right hand sides of [5] and [6] must be positive. Hence, the impact of a change in [E.sub.jt] or [E.sub.it-1] on [E.sub.it] depends on the sign of the second term in each of the cases. Insofar as the sign of [dp.sub.it](*)/ d[E.sub.jt] or d[p.sub.it(*)/d[E.sub.it-1] cannot be determined a priori, the sign of the impact of either [E.sub.jt] or [E.sub.it-1] (or both) must be determined empirically.
[5] [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII],
[6] [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII],
The solution of the taxpayer's utility maximization results in the following general functional form:
[7] [E.sub.it] = f([E.sub.jt],[E.sub.it-1], [X.sub.it]),
where [X.sub.it] is a set of characteristics that influences the evasion behavior of the individual and [E.sub.it], [E.sub.it-1], and [E.sub.jt] are as defined above. Importantly, [E.sub.it] depends upon the dynamic (or persistent) behavior of individual i through [E.sub.it-1], and also upon the interdependent (or spatial) behavior of other taxpayers through [E.sub.jt]. In the next section, we discuss our data, and we also describe our approach to incorporating these dynamic and spatial effects.
DATA AND METHODOLOGICAL ISSUES
General Specification
We posit a general panel model in which individual i's income tax evasion [E.sub.it] depends on individual j's income tax evasion behavior, on individual i's...
|
