Experimental Estimates of Taxpayer Ethics

Abstract

This paper extends the existing literature on taxpayer ethics in three ways. First, we construct a two-stage model of decision making, which allows us to disentangle risk preferences from ethical motivations for income tax compliance. Second, we develop a new experimental data set, which permits us to estimate the magnitudes of the relevant personality traits, risk aversion and morality, at the individual level. Third, we combine the experimental data with participant surveys so that ethical preferences are not only measured but also linked to demographic characteristics. We find that ethical preferences are correlated with risk aversion, age, gender, and marital status, among other characteristics.

Keywords:

• tax evasion
• ethical preferences
• shadow price of morality

Notes

The gross tax gap in the US, before Internal Revenue Service enforcement efforts, was estimated to be $345 billion in 2001, and after audits and recovery, the net tax gap was $290 billion, or 13.7 per cent of the tax revenues owed (Slemrod 2007). There does not appear to be any clear demographic pattern to evasion (Slemrod 2007).

Of course, many taxpayers do not avail themselves of the information. Still, the accessibility of tax information is quite different from the uncertainty regarding capture, conviction, and punishment for street crimes.

Expected utility theory has also been criticized for maintaining axioms which appear to have been violated in experiments, but Harrison (1994) points out that the reported anomalies have largely been generated by invalid experimental procedures.

This assumption is not essential to the model; it was adopted for convenience in the experimental application, as it greatly simplifies the mental computation required to calculate an expected rate of return.

As in the traditional model of Allingham and Sandmo (1972), we assume an individual taxpayer whose preferences are governed by a utility function. The model is not intended for application to corporations or other large organizations. As noted by Slemrod (2007), most of the tax gap results from individuals who under-report their income; corporate income tax evasion accounts for only about 10 per cent of the tax gap.

Researchers seeking to estimate intertemporal discount rates have similarly found it important to disentangle them from risk aversion using a two-stage procedure. Subjects in those studies first reveal their degree of risk aversion by selecting among lottery options or indicating a reservation price for a risky asset, and then reveal their rates of time preference by answering questions about present and future values. As in the current study, the estimated level of risk aversion imposes structure on the utility function so that the parameter of interest can be isolated. See, for example, Ventura (2003) or Andersen et al. (2008).

That particular subject was also unique in that (s)he invested exclusively in bonds, thereby consistently exhibiting zero tolerance for risk. As we would expect from such an individual, (s)he avoided any possibility of a penalty by complying with the tax in all three of the rounds in which (s)he participated.

This includes, for example, studies by Friedland et al. (1978), Spicer and Becker (1980), Friedland (1982), Spicer and Thomas (1982), Benjamini and Maital (1985), Spicer and Hero (1985), Baldry (1986, 1987), Becker et al. (1987), Robben et al. (1990), Webley et al. (1991), Alm et al. (1992), Bosco and Mittone (1997) and Mittone (2006). For surveys, see Alm and McKee (1998) and Torgler (2002).

The use of actual monetary incentives tends to increase the extent to which experimental behavior conforms to the predictions of rational-actor theory (Smith and Walker 1993).

The authors developed a computer program that automatically calculated income based on investment decisions, deducted tax payments based on declared income, subtracted any penalties imposed on audited returns, and added the net income from each round to the subject's accumulated wealth for reinvestment.

The funds were, in fact, used for this purpose, insofar as the tax receipts financed additional rounds of the experiment.

In a classroom experiment, Li et al. (2008) demonstrate that individuals are willing to make voluntary tax payments, comparable to charitable donations, for specific government programs including education. In practice, tax check-offs have become an important source of revenue among states in recent years.

In practice, audit rates vary by the type and amount of income. Through the early 1990s, audit rates exceeded 4 per cent for high-income individuals and those with business income filing schedule C; more recently, audit rates have declined. The relatively high audit probabilities used in the experiment tend to encourage compliance but should not bias the calculation of the ethical parameters.

With the rare exception of highly publicized cases involving elected officials or extraordinarily wealthy individuals such as famed hotel owner Leona Helmsley, tax evasion is generally prosecuted privately. In most instances, the audit and fine are imposed without any media exposure. Cho et al. (1996) note that revenue agencies often renegotiate penalties with offenders in private, which allows the taxpayers to avoid the embarrassment of bankruptcy and allows the government to avoid the transaction costs involved in seizing and liquidating assets. Thus, Grasmick and Bursik (1990) find empirically that while social reprobation is a significant deterrent to other crimes, it is not so for tax evasion. For a discussion of the difference between social embarrassment and true morality, see Gordon (1989).

There were nine sessions, each lasting between nine and eleven rounds. Each subject participated in at least two practice rounds to ensure an understanding of the procedure, before the actual experiment began.

The principal mechanism by which wage and salary workers evade income tax is overstating deductions.

Relative risk aversion is simply the product of absolute risk aversion (α) multiplied by wealth (w), and risk tolerance is the inverse of risk aversion. Thus, absolute risk tolerance and relative risk tolerance are the inverses of absolute and relative risk aversion, respectively. Equations (A1) through (A5) in the Appendix show the derivation of absolute risk aversion.

Naturally, some individuals exhibited greater behavioral variation than others, though in general behavior was reasonably consistent for each individual across rounds. On average, the standard error of absolute risk tolerance for an individual was about 1/6 as large as the mean, which also implies that the mean values were significantly different from zero.

A regression of the compliance rate (YD/Y) on t, p, α, and λ confirmed that behavior responds as expected to changes in the tax rate, the audit probability, risk aversion, and ethical preferences. Because the fine for evasion was always twice the tax rate, it could not be used as an independent variable.

The shadow price of morality was even more stable across rounds than risk tolerance. On average, the standard error across rounds was only about 4 per cent of the individual's mean λ, which also implies that the mean values were significantly different from zero.

However, it is precisely because of limited research funding—i.e., low stakes—that student volunteers are typically recruited. Whereas the sums used as incentives for this and similar experiments may seem negligible to non-students, college students with little income are more likely to perceive them as meaningful. Since each session in the present study was completed in 1.25 hours or less, each subject earned at least $12.94 per hour by participating in the experiment, and some earned more than $39.50 per hour. Such rates compare favorably to most students' opportunity costs. At the time of the study, the federal (and state) minimum wage was $5.85 per hour, and the median wage rate among the subjects was $8.59 per hour.

The exponential utility function, U(w) = −(1/α)exp(−wα), is often employed in the finance literature, and is consistent with empirical evidence of constant absolute and increasing relative risk aversion. Recent work by Botti et al. (2008) indicates that the constant absolute risk aversion function is more empirically relevant than the constant relative risk aversion function.