the effects of the 1986 and 1993 tax reforms on self-employment

National Tax Association

THE EFFECTS OF THE 1986 AND 1993 TAX REFORMS ON SELF-EMPLOYMENTAuthor(s): Kevin MooreSource: Proceedings. Annual Conference on Taxation and Minutes of the Annual Meeting ofthe National Tax Association, Vol. 95 (2002), pp. 323-332Published by: National Tax AssociationStable URL: http://www.jstor.org/stable/41954301 .

Accessed: 10/06/2014 18:01

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp

.JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range ofcontent in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact [email protected].

.

National Tax Association is collaborating with JSTOR to digitize, preserve and extend access to Proceedings.Annual Conference on Taxation and Minutes of the Annual Meeting of the National Tax Association.

http://www.jstor.org

This content downloaded from 195.34.79.195 on Tue, 10 Jun 2014 18:01:17 PMAll use subject to JSTOR Terms and Conditions

http://www.jstor.org/action/showPublisher?publisherCode=nta

http://www.jstor.org/stable/41954301?origin=JSTOR-pdf

http://www.jstor.org/page/info/about/policies/terms.jsp


THE EFFECTS OF THE 1986 AND 1993 TAX REFORMS ON SELF-EMPLOYMENT

Kevin Moore , Johns Hopkins University and the Federal Reserve Board of Governors*

The ure seeking

entrepreneur in capitalist

profit, the

has economic entrepreneur

always been theory.

creates

a key While new

figure in capitalist economic theory. While seeking profit, the entrepreneur creates new

jobs for other individuals, introduces new ideas and technology, and can foster competition in the economy. These predicted positive effects have sparked policy interest in promoting entrepreneurship, with the key issues being how do we encour- age individuals to become entrepreneurs and what policy tools work best to achieve this end.1 One policy tool that many have championed is tax policy.

There are many links between entrepreneurship (self-employment) and taxation. In general, the self-employed have more flexibility in reporting taxes and may be more sensitive to tax policy. There are efficiency concerns if the tax system is encour- aging people to become self-employed for the wrong reasons, say to undertake tax evasion, or if high tax rates discourage individuals who would be highly productive entrepreneurs. Furthermore, if the expected returns to success accrue more directly to the self-employed than employees, the progressive structure of the tax system could be more of an issue for the self-employed.2 There are many more interactions between taxation and self- employment, but this paper will focus on how changes in marginal or average tax rates affect the propensity to become self-employed.

PREVIOUS LITERATURE The quantity of research on self-employment has

increased over the past 25 years. Much of this research focused on the choice of self-employment in the absence of taxation, and this area of literature has provided important insights (theoretically and empirically) into how human capital, ability

*The author would like to thank Robert Avery, Ana Azicorbe, Chris Carroll, Arthur Kennickell, Robert Moffitt, Nicole Nestoriak, Matt Shum, the participants at the JHU Summer Seminar, and various economists from the Federal Reserve Board for comments on the ideas in this paper. Special thanks go to Daniel Feenberg and NBER for allowing the author to use the TAXSIM program. All opinions expressed in this paper are those of the author alone and do not necessarily reflect those of the Board of Governors of the Federal Reserve System. All errors are the responsibility of the author.

measures, household characteristics, and macro- economic variables influence the choice to become self-employed. A good review of the empirical literature is found in Le (1999).

Focusing on the literature on self-employment and taxation, many of these studies use aggregate time-series or cross-sectional data. Long (1982a, 1982b), Moore (1983), Blau (1987), Parker (1996), and Schuetze (1998) find that higher marginal tax rates lead to increases in the self-employment rate. Robson and Wren (1998) find a negative relationship between the self-employment rate and marginal tax rates, but a positive relationship between average tax rates and the self-employment rate. Fairlie and Meyer (1998) and Bruce and Moshin (2002), however, find that marginal tax rates have no effect on the self-employment rate.

Some recent studies have emphasized the importance of using individual level data to estimate the tax response. Examples in this literature ex- clude Carroll et al. (2000a, 2000b), Bruce (2000, 2001), Bruce and Holtz-Eakin (2001), Gentry and Hubbard (2000), and Cullen and Gordon (2002). Of these studies, this paper is closest in spirit to Bruce (2000), who finds a positive relationship between marginal tax rates and entry into self- employment, and Gentry and Hubbard (2000), who find that a more progressive tax rate structure leads to a decrease in entry into self-employment. This paper extends this literature by using repeated cross-section data from the Survey of Consumer Finances with the 1986 and 1993 tax reforms as natural experiments to gauge the effects of tax rate changes on the choice to become self-employed.

ECONOMETRIC METHODS

Identifying Tax Effects It is well known that identifying the effects of a

change in the federal tax law is hindered by the fact that all individuals with a given set of characteristics face the same tax rate at a given time. This lack of variation makes estimating the effect of taxes difficult, and in some cases identification arises only from the non-linearity of the tax func-

323



NATIONAL TAX ASSOCIATION PROCEEDINGS

tion. Plus, since earnings are a component of the tax function, and earnings and the employment decision occur simultaneously, the tax rate is endogenous.

The issue of identifying tax effects figures promi- nently in the recent literature on taxes and labor supply and the literature on taxes and taxable income. One method that has gained widespread use is "differences in differences" (DID). This method is similar to the treatment effects idea, where one group is exposed to some sort of catalyst (treatment group), while the other group is not (control group). Usually, DID is utilized to examine the effects of a natural experiment on two different groups. The two main assumptions behind the DID methodology are that individuals are grouped by an exogenous variable and that the growth rate of the dependent variable for the two groups is equal in absence of the treatment.3 DID estimates can be computed using a simple tabular method or by regression analysis.4

Figure 1 provides a simple tabular example of how the DID methodology works. In Figure 1, yjt is the mean value for group j at time t, with the subscript c representing the control group and the subscript T representing the treatment group.

The simplicity of the DID methodology masks some important underlying econometric issues, which are detailed in Moffitt and Wilhelm (2000). They show that when identifying tax effects using DID, the grouping variable acts as an instrumental variable for the tax function, since the tax function itself is endogenous. Another important result is that the grouping variable should be exogenous and time-invariant. This latter result is important when using RCS data because we cannot follow the same individuals across time and must assume that no individuals switch groups.5

A final issue with DID is the assumption that the growth rate of the dependent variable for the two groups is equal in absence of the experiment.

This assumption is important because it ensures that the changes in the dependent variable are due to the experiment and not to other factors. Since we have data from a period without any marginal tax rate changes, this assumption is testable.

Repeated Cross-Section Analysis The application of repeated cross-section (RCS)

analysis in microeconomics most commonly en- tails constructing birth cohorts in the data so that the population of individuals in different years re- mains the same. Basically, by stratifying the data by birth cohorts, members of the population age into or age out of the sample.6 While early RCS studies, such as Browning, Deaton, and Irish (1985), use group means in estimation, Moffitt (1993) shows that estimation of RCS models is possible using the individual level data. Moffitt also provides insight on identification issues, stresses the importance of stratifying the data by time- invariant characteristics, and shows that RCS estimation is a version of instrumental variable estimation.

In using the DID and RCS methodology to estimate the self-employment choice model used in this paper, we face two main issues. The first issue is identifying the tax effects, while the second issue is the endogeneity of the tax rate in the self- employment choice equation. To solve the endogeneity problem, we utilize instrumental variables, which also solves the identification problem. However, since the instruments are also the grouping variables for the DID methodology, it is important to chose a variable that is relatively time- invariant. The solution to these problems is to use a combination of time-invariant and time- varying variables as the instruments, such as an education and year interaction.

Equation (1) shows the model estimated in the analysis, where the dependent variable is a binary response for whether an individual is self-

Figure 1: DID Example

Pre- intervention Post-intervention _________ Difference

a*m' Äi Ä! &ř,-ýq-ý*

Ijvi

in

t»y,-yTi-yn Difference in Differences ^ =

324



95™ ANNUAL CONFERENCE ON TAXATION

employed at time t , CD. is a vector of cohort dummies, S. is contains all the time-invariant independent variables (except the grouping variable), YRt is a dummy variable equal to 1 if the observa- tion is from the post-intervention year, G. is a dummy variable equal to 1 if the individual be- longs to the treatment group, Tt(Z.) is the tax rate function, W.{ is a vector of the time- varying independent variables, and r'.t is the error term.

(1) C* = CD/5, + S'S2 + S3YRi + Sfi, + 85T(Z.) + W/& it 6 + 7J, 'it it 6 'it

C =lifC*>0 = 0ifC*<0 It It It

Since the dependent variable is binary, we estimate equation (1) using a probit model. Due to endogeneity of the tax function, we use the procedure outlined in Newey (1987) for asymptotically efficient estimation of a binary choice model with endogenous independent variables.7

DATA AND SAMPLE CHARACTERISTICS

Data The data sets we utilize in this paper are the

Survey of Consumer Finances (SCF) for 1983, 1989, 1992, 1995, and 1998. The Board of Gover- nors of the Federal Reserve System conducts this survey of household assets and liabilities trienni- ally. Sample weights are provided for the data, missing values are multiply imputed (the 1983 data is singly imputed), and sample sizes for the SCF range from 3,143 to 4,305 households.8

Due to richness of the SCF data, we are able to control for a variety of factors that may influence the decision to become self-employed, such as education, current job tenure, labor force experience, industry, occupation, geographic location, marital status, number of children, health, attitudes toward risk, expectations of inheritance, wealth, credit worthiness, and previous self-employment experience. Comprehensive income data (for the year prior to the survey) are also collected by the SCF, and this, in conjunction with the other data in the survey, allows for the computation of federal and state tax rates (marginal and average) and liabilities via the NBER TAXSIM model.9

While marginal tax rates represent the tax rate on additional earnings, average tax rates provide a measure of the tax rate for total taxable income.

Both tax rates are computed on a household level basis, with marginal rates computed by adding $1 to earnings. The average tax rate is the ratio of the total tax liability (federal and state) to total income (minus non-taxable income) of the household. Married couples are assumed to file jointly, as are couples that live together.10 The computation of the tax rates includes using the SCF data to calculate the value of the itemized deductions; this includes the mortgage interest deduction, property tax deduction, charitable contributions deduction, invest- ment interest expense deduction, personal interest deduction (in applicable years) and state tax liability deduction.

In addition to federal and state tax rates, we also include the payroll tax rate in the total tax rate measure. We assume that employees pay both the employee and employer part of the payroll tax. The marginal payroll tax rate is computed by adding $1 to the earnings of the household head, while the average payroll tax rate is the ratio of the household head's payroll tax liability to earnings. Since the self-employed faced a lower rate than employees prior to 1990, and the payroll tax is a significant percentage of earnings (up to a maximum amount), it is important to include the payroll tax in the total tax rate measure.11

The sample period of 1983-1998 covers the major federal tax rate changes of 1986 and 1993, and the lesser change in 1990. Table 1 shows how rates have changed for married filing jointly households. We focus on the Tax Reform Act of 1986 and the Omnibus Reconciliation Act of 1993. TRA86 reduced the number of statutory personal tax rates from 12 to two (15 percent and 28 percent), with a "bubble" rate (33 percent) for a certain range of taxable income. OBRA93 created two new high- income rates (36 percent and 39.6 percent), and made permanent the limits on itemized deductions and the phaseout of personal exemptions.

Sample Characteristics Although many definitions of self-employment

are possible, this paper uses a self-reported measure whereby individuals are classified by their response to a question about whether they work for someone else or they are self-employed.12 This definition does not require that the household reports owning an actively managed business.13 The fact that a self-employed individual bears all the earnings risk is the key element of our definition, which is derived from the ideas put forth by Knight ( 192 1 ).

325




Table 1 Tax Brackets for Married Filing Jointly Taxpayers

(in 2001 dollars adjusted by the CPI-RS) 1982 Min 1988 Min 1991 Min Rates Taxable Ine Rates Taxable Ine Rates Taxable Ine 12 $5,984 15 $0 15 $0 14 $9,680 28 $42,840 28 $43,180 16 $13,376 33 $103,536 31 $104,331 19 $20,944

f 1 994 Min 1997 Min „ . Rates Taxable Ine Rates Taxable lne ¿y 3>43,Z9o 33 $52,624 15 $0 15 $0 39 $61,952 28 $44,840 28 $45,320 44 $80,608 31 $108,383 31 $109,560 49 $105,600 36 $165,200 36 $166,925 5 0 $150,656 3916 $295,000 39J> $298,155

As was mentioned in the econometric section of the paper, the choice of grouping variable is a key element in correctly implementing the DID methodology. The main grouping variable we use in this paper is the education level of the household head, where the control group is individuals with a some college or less, and the treatment group is individuals with at least a college degree. By restricting the sample to male household heads aged 25 to 64 who are currently working, we attempt to remove individuals who have not com- pleted their education or have retired. No restric- tions on the minimum number of hours worked per week or weeks worked per year are used, since flexibility of work schedule is a potential benefit of self-employment. Since we are focusing on

personal tax rates, self-employed individuals whose primary business is organized as a privately held C-corporation are also dropped from the sample.

The top panel of Table 2 shows self-employment rates for the full sample and by education group. In the TRA86 sample, the overall self-employment rate is rising, with a sizable jump between 1983 and 1989, while for the OBRA93 sample there is small decrease in the rate between 1992 and 1995. Similar trends are evident in the education specific self-employment rates, except for the some college or less group, where there is a small drop in the self-employment rate between 1989 and 1992 in the both samples, and a rise in the rate between 1992 and 1995 in the OBRA93 sample. Overall,

Table 2 Self-Employment Rates and Tax Rates, by Year and by Education,

1983, 1989, 1992, 1995, and 1998 SCF (for male household heads ages 25-64 who are currently working)

TRA86 Sample OBRA93 Sample Year 1983 1989 1992 1989 1992 1995 1998 Self-reported SE rates All households 11.4 16.6 18.6 15.2 16.5 15.8 17.7

Some college or less 11.6 16.2 15.9 14.7 14.6 15.7 16.9 College or higher 11.0 17.5 22.2 16.1 19.3 15.8 18.9

Tax rates All Households MTR 42.1 36.0 37.3 36.0 37.0 37.4 38.5

ATR 29.3 28.2 28.9 27.5 28.4 27.9 29 Some college or less MTR 40.7 35.1 36.6 34.9 36.2 36.3 37.2

ATR 28.0 27.2 27.5 26.4 27.1 26.8 27.5 College or higher MTR 45.1 37.5 38.3 37.9 38.2 39.1 40.4

ATR 32.0 29.8 30.9 29.5 30.3 29.6 31.0 Note: All estimates are weighted by the sample weights, the MTR and ATR include the payroll tax.

326




Table 2 shows some evidence of a negative relationship between self-employment rates and changes in tax rates.

The information on self-employment rates by education group in Table 2 can also help determine if the growth of the self-employment rate is similar for the two groups in periods without a major tax rate change. For the TRA86 sample, while the self-employment rate rises for the highly educated group over the 1989 to 1992 time period, it is es- sentially flat for the some college or less group. These results also hold true in the OBRA93 sample for 1989 to 1992. Of course, since there was a small tax change in 1990, the 1989 to 1992 time period may not be untainted. Examining the 1995 to 1998 time frame for the OBRA93 sample, we see a rise in the self-employment rate for both groups, with a higher growth rate for the college degree or higher group. Although there was no change in marginal tax rates in the Taxpayer Relief Act of 1997, the tax rate on capital gains decreased for most assets. Any interactions between capital gains rates and marginal rates could call into question labeling 1995 to 1998 as a period without tax rate changes. The first-stage regressions will provide an additional test on the validity of the grouping variable.

Another issue related to the choice of grouping variable is whether there is significant variation in tax rates between education groups. By definition, the treatment group should be affected more by the change in tax rates than the control group. The bottom panel of Table 2 shows the total marginal and average tax rates by education group. Over the 1983 to 1989 time period, the average marginal tax rate for the highly educated group fell 7.6 per-

centage points, while the some college or less group saw a decline of 5.5 percentage points. For 1992 to 1995, the highly educated group saw close to a 1 percentage point increase in the average marginal tax rate, versus a 0. 1 percentage point increase for the some college or less group. Average tax rates vary much less than marginal tax rates across all time periods. Overall, the changes in marginal or average tax rates in both samples are larger for the highly educated group.

RESULTS As a check on our sample, we estimate the self-

employment choice model without the tax rates.14 The results from these regressions are similar to those found in the rest of the literature and they are consistent across both samples and most time periods.

Turning to the models with the tax variables, Table 3a and 3b show the results of the first stage regressions of the tax rate variables on the various instruments and the other exogenous variables.15 To ensure we have strong instruments we include the F-statistic for the test of the null hypothesis that the coefficient on the instrument is weak. As shown in Stock, Wright, and Yogo (2002), a significant coefficient on the instrument and a low p- value for the F-statistic is not enough to guarantee a strong instrument. Stock and Yogo (2002) provide critical values of for the F-statistic that iden- tifies strong instruments (Table 2 in their paper). For a single endogenous regressor and a single instrument, the critical value for a 5 percent Wald test that has an actual size potentially exceeding 10 percent is 16.38, and for 15 percent is 8.96.

Table 3a First Stage Regressions for the Marginal or Average Tax Rate,

TRA86 Sample, 1983, 1989, and 1992 SCF (for male household heads ages 25-64 who are currently working)

1983-1989 1989-1992 , College*Yr89 College*Yr92 Instrument , Tax Rate MTR ATR MTR ATR College * 1989 -6.718 -3.653

(0.731)*** (0.503)*** College * 1992 0.841 1.456

(0.781) (0.493)*** R-Squared 0.32 0.30 0.14 0.20 F-stat for instrument 87.92 55.36 1.31 9.20 p- value of F-stat 0.00 0.00 0.28 0.00 # of Observations 3008 3008 2973 2973

327




Table 3 b First Stage Regressions for the Marginal or Average Tax Rate, OBRA93 Sample,

1989, 1992, 1995, and 1998 SCF (for male household heads ages 25-64 who are currently working)

1989-1992 1992-1995 1995-1998 Instrument College*Yr92 College*Yr95 College*Yr98 Tax Rate MTR ATR MTR ATR MTR ATR College * 1992 0.893 0.952

(0.793) (0.487)* College * 1995 1.631 0.612

(0.784)** (0.489) CoUege * 1998 0.784 0.682

(0.764) (0.520) R-Squared 0.14 0.23 0.11 0.22 0.10 0.24 F-stat for instrument 1.48 4.24 5.03 1.67 1.20 2.01 p-value of F-stat 0.25 0.05 0.03 0.21 0.30 0.18 # of Observations 3029 3029 3486 3486 3634 3634 Regressions corrected for multiple imputation, the other independent variables included in the regressions are omitted to save space. Notes: Standard errors in parentheses. ♦significant at the 10 percent level, **significant at the 5 percent level, ***significant at the 1 percent level

In Table 3a, for the 1983 to 1989 time period and either tax rate, the instrument has an F-statistics that far exceeds 16.38. Over the 1989 to 1992 time frame, the results are not as consistent, as only the instrument for average tax rates has a large enough F-statistic. In Table 3b, for all time periods in the OBRA93 sample, the instruments are weak for both tax rates.

We test other instruments for the 1992 to 1995 time period, including using a high-school or less (control group) versus a some college or higher (treatment group) education grouping, and a blue-collar (control group) versus white-collar (treatment group) occupation grouping. Both instruments are weak over 1992 to 1995. 16 Next, we attempt to use the education and occupation variables together as instruments, but these instruments are also weak over 1992 to 1995. The cause of the weak instrument problem may be that OBRA93 affected only high-income taxpayers (taxable income greater than $165,000 if married filing jointly), and only a small fraction of our treatment group is above that threshold. While using income as the grouping variable (instrument) may seem like a possible solution, since income is endogenous and time varying, it is problematic. Without strong instruments, the results for the OBRA93 sample are suspect; therefore, we focus primarily on the results for the TRA86 sample.17

Beginning with the TRA86 sample for 1983 to 1989, Table 4a shows a negative and significant

effect of the total marginal tax rate on the propensity to choose self-employment. This relationship exists regardless of the tax rate measure, but the coefficient is almost twice as large for the average tax rate. For 1989 to 1992, we only have a strong instrument for average tax rates, but the coefficient is insignificant. In Table 4b, for the OBRA93 sample, since all the instruments are weak, the results for the tax variables are suspect. Nonethe- less, over the 1992 to 1995 period, the sign on either tax rate is negative, which is consistent with the results from the TRA86 sample.

The negative relationship between tax rates and self-employment is at odds with the idea that individuals chose self-employment for tax avoid- ance reasons, but is consistent with the idea that higher tax rates erode the expected returns to self- employment. This result is consistent with Hubbard and Gentry (2000), who focus explicitly on the progessivity of the tax schedule. While our study does not focus explicitly on the progressivity issue, since TRA86 drastically reduced the top tax rates and OBRA93 added back in higher rates, there certainly was a change in the progressivity of the tax rate schedule.

A few examples using the marginal effects in Table 4a give a better idea of magnitude of the tax effects. For the 1983 to 1989 time period, a 6 percentage point decrease in the marginal tax rate leads to a 6.6 percentage point increase in the probability of an individual choosing self-employment.

328



95th ANNUAL CONFERENCE ON TAXATION

Table 4a IV Probit for Self-Employment Choice, TRA86 Sample

(for male household heads ages 25-64 who are currently working) Self-reported self-employment measure

1983-1989 1989-1992 , College*Yr89 College*Yr92 Instrument , Tax Rate MTR ATR MTR ATR Total MTR -0.011 0.034

(0.004)** (0.069) Total ATR -0.019 0.020

(0.008)** (0.027) 1 989 Year dummy 0.034 0. 1 02

(0.050) (0.034)*** 1992 Year dummy -0.046 0.019

(0.191) (0.045) College or higher 0.059 0.092 0.161 0.098

(0.020)*** (0.029)*** (0.066)** (0.049)** Age 45-54 in 1983 -0.074 -0.091 0.020 -0.022

(0.041)* (0.043)** (0.228) (0.096) Age 35-44 in 1983 -0.113 -0.141 0.077 0.008

(0.052)** (0.055)** (0.344) (0.138) Age 25-34 in 1983 -0.149 -0.193 0.043 -0.037

(0.058)** (0.063)*** (0.435) (0.177) Ag / Mining / Const 0.161 0.149 0.352 0.295

(0.034)*** (0.036)*** (0.158)** (0.053)*** Manufacturing -0.118 -0.118 -0.176 -0.155

(0.017)*** (0.017)*** (0.058)*** (0.026)*** Wholesale /Retail 0.054 0.045 0.142 0.137

(0.026)** (0.027)* (0.070)** (0.048)*** FIRE / Bus / Repair 0.132 0.146 0.269 0.205

(0.029)*** (0.028)*** (0.165) (0.041)*** Head is white 0.080 0.089 0.126 0.090

(0.017)*** (0.017)*** (0.052)** (0.033)*** High financial risks 0.051 0.058 0.129 0.086

(0.019)*** (0.020)*** (0.068)* (0.032)*** Log of house value 0.006 0.007 0.010 0.010

(0.002)*** (0.002)*** (0.005)* (0.004)** Longest prior job SE 0.060 0.071 0.245 0.194

(0.034)* (0.034)** (0.092)*** (0.041)*** # of Observations 3008 3008 2973 2973 Regressions corrected for multiple imputation, and also include controls for tenure at the current job, labor force experience, poor health, marital status, number of children under 18, region, state unemployment rate, spouse's earnings, credit worthiness, and expectation of inheritance. Notes: Coefficient values are marginal effects, standard errors in parentheses. * significant at the 10 percent level, ** significant at the 5 percent level, *** significant at the 1 percent level

While for average tax rates, a 1 percentage point decrease leads to a 1 .9 percentage point increase in the probability of choosing self-employment. 18

When comparing the effects of the average and marginal tax rates, average tax rates have a larger coefficient, and therefore a larger effect than marginal tax rates. However, as shown in Table 2, average tax rates do not vary over time as much as marginal tax rates. Even after the sweeping rate reductions of TRA86, the most the mean average tax

rate decreased was about 2 percentage points for the college or higher group (from 32 percent to 29.8 percent). This is in contrast to a 7.6 percentage point reduction in marginal rates for this same group. So, although average tax rates have more bite, much larger broad marginal rate reductions must occur to induce large changes in average tax rates.

An important question is how the magnitude of the tax effects compares to the effects of other variables in the self-employment choice equation. Note

329




Table 4b IV Probit for Self-Employment Choice, OBRA93 Sample

(for male household heads ages 25-64 who are currently working) Self- reported self-employment measure

1989-1992 1992-1995 1995-1998 rear College*Yr92 College*Yr95 College*Yr98 Instrument 2 Tax Rate MTR ATR MTR ATR MTR ATR Total MTR 0.040 -0.041 0.079

(0.069) (0.028) (0.117) Total ATR 0.037 -0.113 0.089

(0.043) (0.116) (0.095) 1992 Year dummy -0.065 -0.015

(0.163) (0.057) 1 995 Year dummy 0. 130 0.1 36

(0.066)** (0.104) 1998 Year dummy -0.123 -0.112

(0.253) (0.175) College or higher 0.119 0.031 0.113 0.335 -0.022 -0.147

(0.036)*** (0.095) (0.032)*** (0.235) (0.173) (0.254) Age 45-54 in 1989 -0.024 -0.028 -0.144 -0.221 0.147 0.273

(0.093) (0.070) (0.123) (0.230) (0.387) (0.390) Age 35-44 in 1989 0.033 -0.008 -0.225 -0.361 0.093 0.303

(0.150) (0.089) (0.154) (0.311) (0.465) (0.510) Age 25-34 in 1989 0.005 -0.032 -0.306 -0.472 0.080 0.311

(0.214) (0.126) (0.141)** (0.277)* (0.556) (0.589) Ag/ Mining /Const 0.391 0.324 0.200 0.093 0.447 0.419

(0.210)* (0.088)*** (0.088)** (0.214) (0.210)** (0.141)*** Manufacturing -0.172 -0.152 -0.077 -0.096 -0.159 -0.120

(0.052)*** (0.025)*** (0.031)** (0.044)** (0.062)** (0.037)*** Wholesale / Retail 0.121 0.111 0.095 0.015 0.191 0.191

(0.090) (0.059)* (0.043)** (0.117) (0.114)* (0.075)** FIRE /Bus /Repair 0.244 0.190 0.162 0.194 0.114 0.064

(0.141)* (0.049)*** (0.045)*** (0.058)*** (0.082) (0.102) Head is white 0.119 0.077 0.086 0.147 0.115 0.007

(0.041)*** (0.039)* (0.029)*** (0.069)** (0.057)** (0.109) High financial risks 0.126 0.065 0.066 0.200 0.033 -0.033

(0.066)* (0.046) (0.029)** (0.122) (0.051) (0.093) Log of house value 0.010 0.008 0.021 0.033 0.003 -0.004

(0.005)* (0.005) (0.004)*** (0.019)* (0.020) (0.021) Longest prior job SE 0.262 0.181 0.228 0.364 0.330 0.134

(0.134)* (0.047)*** (0.058)*** (0.115)*** (0.146)** (0.123) # of Observations 3029 3029 3486 3486 3634 3634 Regressions corrected for multiple imputation, and also include controls for tenure at the current job, labor force experience, poor health, marital status, number of children under 18, region, state unemployment rate, spouse's earnings, credit worthiness, and expectation of inheritance Notes: Coefficient values are marginal effects, standard errors in parentheses. * significant at the 10 percent level, ** significant at the 5 percent level, *** significant at the 1 percent level

that the results for the non-tax variables change very little when properly instrumented tax rates are added to the model. From Table 4a, for the 1983 to 1989 time frame, we see that both tax measures have less of an influence on the probability of self- employment than any of the other significant variables shown (except house value). Variables like education, wealth (house value), industry, race, attitude toward risk, and previous self-employment

experience consistently play a more influential role in determining who becomes self-employed.

CONCLUSIONS Understanding why individuals choose to be-

come self-employed and how public policies affect this choice is an important area of research. Altering tax policies is often the mechanism by

330




which the government attempts to influence the behavior of individuals and households.

This paper shows that the sweeping rate reductions in TRA86 did have an affect on the propensity of individuals to choose self-employment; a 6 percentage point decrease in marginal tax rates leads to a 6.6 percentage point increase in the probability of choosing self-employment. The rate increases in OBRA93, however, had no significant effect on the choice to become self-employed. Thus, it seems that relatively small or narrowly focused marginal tax rate changes will not significantly affect the self- employment rate. Other covariates, such as education, industry, wealth, attitudes toward risk, and previous self-employment, are a more constant influence on the self-employment decision. Of course, other elements of the tax code may matter more than marginal tax rates, such as the capital gains tax rate, but those are topics for future research.

Notes 1 This policy question is even more interesting given the recent finding by Blanchflower, Oswald and Stutzer (2001) that 70 percent of workers would prefer to be self-employed, while only about 8-14 percent are self- employed. 2 Of course, the self-employed also accrue the expected returns to failure more directly than employees. 3 The first assumption is analogous to a condition nec- essary to apply the Wald estimator.

4 Meyer (1995) provides an excellent overview of the use natural experiments and DID.

5 See Heckman (1996) for an example of the bias these switchers may cause.

6 This assumes the population has no in or out flows of immigrants and no births or deaths.

7 The Stata command that performs this procedure was written by Joe Harkness and is available to download at http://ideas.uqam.ca/ideas/data/Softwares/ bocbocodes415801 .html.

8 Data for the 2001 SCF should be available to the public in early 2003. For more details on the SCF, visit www.federalreserve.gov/pubs/oss/oss2/scfindex.html. 9 For a description of the TAXSIM model, see Feenberg and Coutts (1993) or visit www.nber.org/-taxsim/. 10 The SCF does ask about the household's filing status, and some married couples do report filing separately. Unfortunately, due to the difficulty in dividing up the income flows and tax deductions, we assume that married filing separately and cohabitating couples file jointly. 1 1 Aitnougn worKers ao derive social security ana Medi- care benefits from payroll tax contributions, Feldstein and Samwick (1992) show that the net marginal social security tax rate varies widely among workers.

12 The text of this question has changed somewhat over the course of the survey, but not in a way that should greatly influence responses. The specific variable used in the self-reported measure is B4540 in the 1983 SCF, and X4106 in the 1989, 1992, 1995 and 1998 SCFs.

13 Note that in any given survey year, about 70 percent of the self-reported self-employed own an actively managed business.

14 All regressions are adjusted for the effects of multiple imputation on the coefficients and standard errors. Because the 1983 data is only singly imputed, we created five replicates of the 1983 data to facilitate the computation of the adjustment for multiple imputation over the 1983 to 1989 time period. To assess the influence the replicating had on the results, we repli- cate one implicate of the 1989, 1992, 1995 and 1998 data five times and estimate our models. While we would expect the replicated data to produce smaller standard errors in the regressions, this was not always the case. In fact, often the standard errors were higher in the replicated data. Overall, the replicating the 1983 data produces standard errors that are about 2 percent larger, which will not affect the highly significant coefficients in the regressions. 15 While all other exogenous variables are included in the regressions, only the instruments are listed in Table 3a and 3b to save space. 16 Both these instruments are strong instruments over the 1983 to 1989 period, and choice models estimated with these instruments yield similar results to those obtained using the original instrument.

17 Work on finding strong instruments for the 1993 tax act continues.

18 1 he percentage point decrease or increase m marginal and average tax rate is chosen to approximate the average change in tax rates for all households over the given time period (see Table 2, first row of bottom panel).

References Blanchflower, David, Andrew Oswald and Alois Stutzer.

Latent Entrepreneurship across Nations. European Economic Review 45 (2001): 680-691.

Blau, David. A Time-series Analysis of Self-employment in the United States. Journal of Political Economy 95 (1987): 445-467.

Browning, Martin, Angus Deaton and Margaret Irish. A Profitable Approach to Labor Supply and Commod- ity Demands over the Life Cycle. Econometrica 53 (1985): 503-544.

Bruce, Donald. Effects of the U. S. Tax System on Transitions into

Self-employment. Labour Economics 7 (2000): 547-575.

Taxes and Entrepreneurial Endurance: Evidence from the Self-Employed. National Tax Journal 55 (March 2001): 5-24.

331




Bruce, Donald, and Douglas Holtz-Eakin. (2001) Taxes and the Allocation of Entrepreneurial Effort. Knox- ville: Department of Economics, University of Ten- nessee, 2001. Unpublished Working Paper.

Bruce, Donald, and Mohammed Moshin. Tax Policy and Entrepreneurship: New Time Series Evidence. Paper presented at the National Tax Association 95th An- nual Conference on Taxation, Orlando, November 2002.

Carroll, Robert, Douglas Holtz-Eakin, Mark Rider and Harvey Rosen. Income Taxes and Entrepreneurs' Use of Labor.

Journal of Labor Economics 18 (2000): 324-351. (2000a)

Personal Income Taxes and the Growth of Small Firms. Cambridge: National Bureau of Economic Research, 2000. NBER Working Paper No. 7980. (2000b)

Cullen, Julie, and Roger Gordon. Taxes and Entrepre- neurial Activity: Theory and Evidence for the U.S. Cambridge: National Bureau of Economic Research, 2002. NBER Working Paper No. 9015.

Fairlie, Robert, and Bruce Meyer. Trends in Self-employment among White andBlackMen: 1910-1990. Cam- bridge: National Bureau of Economic Research, 1998. NBER Working Paper No. 7182.

Feenberg, Daniel, and Elisabeth Coutts. An Introduction to the TAXSIM Model. Journal of Policy Analysis and Management 12 (1993): 189-194.

Feldstein, Martin, and Andrew Samwick. Social Secu- rity Rules and Marginal Tax Rates. National Tax Jour- nal 45 (March 1992): 1-22.

Gentry, William, and R. Glenn Hubbard. Tax Policy and Entrepreneurial Entry. New York: Columbia Univer- sity, 2000. Working Paper.

Heckman, James. Labor Supply and the Economic Re- covery Act of 1 98 1 : A Comment, Empirical Founda- tions of Household Taxation. NBER Research Project Report. Chicago: University of Chicago Press, 1996, pp. 32-38.

Knight, Frank. Risk, Uncertainty, and Profit. Chicago: University of Chicago Press, 1921.

Le, Anh. Empirical Studies of Self-Employment. Jour- nal of Economic Surveys, 13 (1999): 381-416.

Long, James. Income Taxation and the Allocation of Market

Labor. Journal of Labor Research 3 (1982): 259- 276. (1982a)

The Income Tax and Self-Employment. National Tax Journal 35 (March 1982): 31-42. (1982b)

Meyer, Bruce. Natural and Quasi-Expenments in Eco- nomics. Journal of Business and Economic Statistics 13(1995): 151-161.

Moffitt, Robert. Identification and Estimation of Dynamic Models with a Time Series of Repeated Cross-Sec- tions. Journal of Econometrics 59 (1993): 99-123.

Moffitt, Robert, and Mark Wilhelm. Taxation and Labor Supply Decisions of the Affluent. In Joel Slemrod, ed. Does Atlas Shrug? The Economic Consequences of Taxing the Rich. Cambridge: Harvard University Press for the Russell Sage Foundation, 2000.

Moore, Robert. Self-employment and the Incidence of the Payroll Tax. National Tax Journal 36 (September 1983): 491-501.

Newey, Whitney. Efficient Estimation of Limited Depen- dent Variable Models with Endogenous Explanatory Variables. Journal of Econometrics 36 (1987): 231-250.

Parker, Simon. A Time Senes Model of Self-employment Under Uncertainty. Economica 63 (1996): 459-475.

Robson, Martin, and Colin Wren. Marginal and Average Tax Rates and the Incentive for Self-Employment. Southern Economic Journal 65 (1999): 757-773.

Schuetze, Herbert. Taxes, Economic Conditions, and Recent Trends in Male Self-employment: A Canada- US Comparison. McMaster University, 1998. Work- ing Paper.

Stock, James, and Motohiro Yogo. Testing for Weak In- struments in Linear IV Regression. Cambridge: Na- tional Bureau of Economic Research, 2002. NBER Technical Working Paper No. 284.

Stock, James, Jonathan Wright and Motohiro Yogo. A Survey of Weak Instruments and Weak Identification in Generalized Method of Moments. Journal of Busi- ness and Economic Statistics 20 (2002): 518-529.

332



the effects of the 1986 and 1993 tax reforms on self-employment

Documents