Determinants of aggregate income-tax-evasion behavior: The case of the US

Banca Nazionale del Lavoro Quarterly Review, Sep 1998 by Cebula, Richard J

In order to measure variables AUDIT and PEN, data indicating the percentage of filed federal income tax returns in any given year that were actually audited by the IRS and the total penalty (penalties plus interest) assessed by the IRS per dollar of reported AGI were obtained from the IRS (1971-1996).

Next, we consider the data for measuring aggregate income-taxevasion behavior, i.e., the relative size of the underground economy. A number of studies have estimated the size of the underground economy over the years. Among the well-known past major contributions in this area in terms of the US are those by Tanzi (1982 and 1983), Feige (1989 and 1994), Bawley (1982), Carson (1984), Pozo (1996) and Pyle (1989). Based on such studies, there appear to be three primary approaches to estimating the size of the underground economy for the US:

1) the taxpayer Compliance Measurement Program;

2) the AGI gap approach; and

3) Currency Ratio Models.

The third of these approaches includes the General Currency Ratio model (GCR).

In this study, to measure the relative size of the underground economy, we adopt the series generated by Edgar Feige. Feige has generated revised and updated estimates of aggregate unreported income as a percent of reported aggregate adjusted gross income (AGI) based on the GCR model, employing an IRS estimate of unreported income for 1973 as the base year. Because these data are available for the years 1973-94, and because they appear to be the most current complete data set presently available on the relative size of the underground economy in the US, they are used as the dependent variable (UGE/ RAGI) in the empirical estimates.

4. Initial empirical findings

The AEPIT, AESST, AUDIT and PEN data were obtained from the IRS (1970-1996); the ACIT data were obtained from the Council of Economic Advisors (1997); the DIS series was obtained from the University of Michigan's ISR; and the estimated data for the UGE/RAGI data were provided by Edgar Feige. The AEPIT variable is lagged two periods due solely to multicollinearity problems. The time series examined in this study are annual and cover the 1973-94 period. The Phillips-Perron (PP) and Augmented Dickey-Fuller (ADF) tests indicate that the following four variables in equation (9) are non-stationary in levels but stationary in first differences: AESST, ACIT, PEN and AUDIT. The remaining variables are stationary in levels. These PP and ADF test statistics are summarized in Table 1. Thus, in estimation equation (10), variables AESST, ACIT, PEN and AUDIT are expressed in first differences form.

Since the (dependent) variable (UGE/RAG)t is contemporaneous with variable DISt, the possibility of simultaneity bias exists. As a result, equation (9) is estimated using an Instrumental Variables (IV) technique to correct for possible simultaneous equation bias, with the instrument being Ut 3, the average unemployment rate of the civilian labor force in year t-3, as a percent. Table 1 shows that the instrument is stationary in levels. The choice of instrument is based on the finding that DISt is highly correlated with Ut 3, whereas the error terms in the system are not contemporaneous with the lagged (by three periods) instrument. Data for Ut-3 were obtained from the Council of Economic Advisors (1997, Table B-40). To correct for heteroscedasticity, the White (1980) procedure is used.