Chapter 2 Tax Evasion and Avoidance among U.S Households
2.4 Estimating evasion and avoidance
2.4.8 Robustness checks and discussions
Sample selection
The main sample used in this paper is families turning from employees to self-employed.
What if we include firms that were already in existence initially? In row 9 and row 10 in
table 2.3 we use overall sample and obtain estimates of income under-reporting fraction close
to our main estimate, which provides evidence that considering a broader sample won’t affect
the main results.
Use the same tax formula for a household
The tax liability for our main regressions are calculated using the current year tax
formula. One concern is that the tax liability change could be due to either behavioral
responses or simply due to mechanical tax formula change. To address such concern, we use
the same tax formula for a household along its employee-self-employed cycle. In particular, we
use the tax formula in the last year before a family turns from an employee to self-employed.
Table 2.17 shows that using the same tax formula does not affect our main results. Although
the concrete numbers change, the main result that the corporate self-employed rely more
on avoidance while the non-corporate self-employed rely more on evasion remains. Despite
the similar results, we use tax liability calculated using current year tax formula because
it would allow us to compare the tax liability of the households turning from employees to
self-employed with other households. For example, if we use the tax formula in the last year
before a family turns from an employee to self-employed to calculate the tax liability, the
calculate.
Accounting for both federal and state tax liability
Since the TAXSIM model includes state tax code from 1977, to best utilize all years
available PSID data, we focus on federal tax liability throughout the paper. Using overall tax
liability including both federal and state tax for the post-1977 sample renders similar result.
Controlling for net wealth
Previous papers, e.g. Hurst et al. (2014), do not include wealth in their specification.
One reason is that wealth information is not available in most years. Another reason is that
wealth may be a function of other socioeconomic characteristics such as age and education.
Thus by controlling Xit we have included a good portion of wealth. In this session, we try to
directly control for net wealth for years when such information is available. We follow Hurst
and Lusardi (2004) to use the fifth-order polynomial of net wealth (in 100000 dollar unit),
as an approximation of some nonlinear wealth model. We also try to include a third-order
polynomial because it may suffice to approximate a nonlinear wealth function.
Table 2.18 shows the results. In all regressions we focus on the observations with
wealth information. Comparing columns 4-6 to columns 1-3, we see that controlling for family
fixed-effect would have our estimates of evasion close to that obtained using the full sample
but without controlling for net wealth. Based on the F.E. regressions in columns 5 and 6,
we see that around 8-9% income is under-reported, whether we control for the third-degree
or fifth-degree polynomial of net wealth. These estimates are close to that obtained in
column 4, which does not control for net wealth at all. This suggests that by controlling
the socioeconomic characteristics and family fixed-effect, we may have controlled for most
information that net wealth would offer. Thus, we could be confident with results obtained
without controlling for net wealth. Controlling the log form of net wealth or negative net
Different measures of consumption
The life-cycle consumption model implies a log-linear Engel curve relation between
total income and total consumption, which motivates our using of total consumption as
our main consumption measure. Previous research (Pissarides and Weber (1989), Hurst
et al. (2014)) instead focuses on food consumption because food consumption is consistently
available over most years. Replacing total consumption by food consumption, we get an
income under-reported fraction of 4.9%, lower than the 10% estimate using total consumption.
Spousal wage/salary
When the head becomes self-employed, the wife might have behavioral responses. A
possible response is that the wife may increase labor supply as precautionary behavior, in case
that the head might not be successful in his self-employed business. In fact, many businesses
experience losses in the beginning. Consider when the head becomes self-employed, his income
decreases while his wife makes compensated income increase, which makes the total reported
income largely unchanged and consumption increase due to potential expenditure increase
for the self-employed, as we observe in our data. If that is true, then the evasion explanation
may not hold. We thus focus on the households with employee wives, and examine whether
spousal wage/salary increases when the head becomes self-employed. We find that on average,
the spousal earnings do not change much. Even in the first year when the head becomes
self-employed, the change in spousal income is not significantly positive (table 2.19).
Occupation change within an employee status
The main framework we use in this paper is to compare the consumption-income
relation around a family’s head turns from an employee to an self-employed. By doing this,
we ignore the potential heterogeneity within employees. Consider if a person faces job change
or occupation change as an employee leading to comparable changes in consumption as
income and tax liability? If the income-consumption relation changes when a person changes
occupation/job as an employee, then our framework should be revised to a comparison with
the person who experience occupation/job change as an employee.
To explore such possibility, we examine the income-consumption relation when a head
experiences occupation change as an employee. From 1968 to 2001 waves, PSID provides
consistent 3-digit 1970 census code, for 2003-2011 the 3-digit 2000 census code is adopted.
Since the match between the two codes are complicated and imperfect, we focus on the
3-digit occupation using 1968-2001 waves data for such analysis. We find that when the head
experiences an occupation change as an employee, his income and consumption do not change.
Furthermore, there is no change in his reported total income conditional on consumption,
and no change in his tax liability conditional on total income. These results suggest that our
framework is appropriately used and there is no need to compare with the the person who
experience occupation/job change as an employee (table 2.20).
Accounting for transitory income shock
In our main specification, we implicitly assume that any observed change in income
is permanent. But the more transitory the change in income, the larger change in income
we would observe conditional on consumption. This raises the necessity to explicitly deal
with the problem of transitory income shock. Both Pissarides and Weber (1989) and Hurst
et al. (2014) have done this, but with some difference. Pissarides and Weber (1989) take a
more parametric approach towards the income process that adjusts for the greater volatility
of transitory income fluctuations for the self-employed. Ignoring the differences in variance
between groups could overestimate the extent of under-reporting. Hurst et al. (2014) in their
online appendix show that adjusting for the differences in variance using the Pissarides-Weber
model decreases the estimated under-reporting only slightly. But Pissarides and Weber (1989)
only adjust for differences in transient income volatility. Hurst et al. (2014) show that if
the bias. After also correcting for differences in permanent income volatility, the estimates
is close to original estimates. Using the same data (PSID) and fundamentally the same
assumption as in Hurst et al. (2014), we tend to rely on their result to justify our approach
in this paper.21
2.5
Conclusion
In this paper we have made some first attempts to estimate the scale of tax evasion
and avoidance using panel survey data. We estimate the scale of income under-reporting, tax
evasion and avoidance for families when they become self-employed. Compared with previous
literature (Hurst et al. (2014)) using survey data to estimate income under-reporting scale of
the self-employed relative to employees, we obtain an estimate much closer to that implied by
the IRS tax audit data. Another finding is that for tax saving purpose, evasion may be more
important for the non-corporate self-employed, while avoidance may be more important for
the corporate self-employed. The rich information contained in PSID allows us to examine
several aspects of family tax saving behavior. But there are also limitations. One limitation
is that since the corporate income is not directly observed from PSID data, we infer it based
on capital gains and dividends information. Another limitation is that since S or C status of
a corporation is not reported in PSID, we take an approach assuming the corporation would
take the organizational form that minimizes tax burden. To concur these limitations, future
research may consider using administrative data linking individual/household to firms, which
has become a popular trend recently. This paper does not examine heterogeneity in evasion
and avoidance across occupations and industries, which may be an interesting direction for
future research.
Chapter 2 is currently being prepared for submission for publication of the material.
The dissertation author, Xiaxin Wang, was the sole author of this paper.