The effect of death tax on the labour supply of donors: evidence from TRA97

ABSTRACT

Exploiting estate tax cuts from the Taxpayer Relief Act of 1997 (TRA97), this paper estimates the effect of death tax on the labour supply of living potential donors. To this end, difference-in-difference with multiple imputation approach is applied to micro-level panel data. This paper finds that the estate tax cuts makes no difference in labour force participation or working hours of potential donors in a statistically meaningful way, although the TRA97 reduces marginal estate tax rates by 37.51% on average. This finding suggests that the death tax causes no meaningful distortion of living potential-donors’ labour supplies at either extensive or intensive margin.

KEYWORDS:

• Estate taxation
• labour force participation
• working hours

JEL CLASSIFICATION:

• H24
• J22
• H30

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

¹ We estimate the expected time of death of each individual based on his (her) gender, race, and age as of the year 1996, according to the CDC life expectancy table. The HRS does not contain data on the expected life expectancy of individual respondents; at most, it provides the information on the probability for respondent individuals to be alive at the age of 75 or 85.

² This quintile regression method that Bernheim, Lemke, and Scholz (2004) adopted is similar to that of Holtz-Eakin and Marples (2001). As wealth is not distributed symmetrically, 10 quantile regressions are better than one OLS regression. This paper improves upon their approaches in the following three aspects. First, we obtain statistical validity of our estimation for missing data on expected terminal wealth by relying on the multiple imputation method, which is theoretically proven by Little and Rubin (1987). Second, to enhance the accuracy of our imputation using quintile regressions, we increase the number of sub-groups to 10 from 5 (i.e., run 10 quantile regressions instead of 5). Third, we add a new step for improving the accuracy of imputed terminal wealth by comparing imputed past wealth and actual past wealth, which will be detailed below.

³ In detail, the final estimate for each coefficient is the average of the 30 estimated coefficients from the 30 runs of a regression equation with the 30 imputes. Moreover, the standard errors of each final estimate for inference are obtained by summing the following two terms: the average of the 30 standard errors from the regressions with the 30 imputes (within-imputation variability) and product of the variance of the 30 standard errors and $1+\frac{1}{30}$(between-imputation variability).

⁴ In setting the tolerance limit, if we set the limit too low, then we end up with a too small number of observations for treatment and control groups for regression analysis.

⁵ As an alternative, synthetic control method, which is also used for causal inference studies like Abadie and Gardeazabal (2003) and Peri and Yasenov (2015), could be considered for identifying the effect of the TRA 97 estate tax cuts. While the difference-in-difference approach uses one pre-treatment period, synthetic control approach uses multiple pre-treatment periods to construct a synthetic counterfactual that estimates what a treated individual would do if he were not treated, based on matching estimation. Like the difference-in-difference approach, this matching estimator also can take account of the unobservable unique characteristic of an individual. However, synthetic control approach is less suitable for identifying the effect of the TRA 97 estate tax cuts. To understand why, notice that for unbiased estimation, conditional mean independence assumption and common support assumption should be met. In particular, conditional mean independence assumption means that clearly given treatment assignment is independent of outcome (labour supply) after a proper matching, entailing that the mean outcomes of both what individuals of control group do and what treated individuals would do with no treatment (which is not observable). Firstly, treatment assignment is not clearly given in the TRA97 estate tax cuts treatment. Secondly, while it is possible to test the parallel assumption for unbiased difference-in-difference estimation utilizing pre-treatment periods, as we do in this paper (Section 4), it is not possible to test the conditional mean independence assumption for unbiased estimation of synthetic control approach. In addition, notice that whether an individual is treated is not clearly given in the TRA estate tax cuts experiment, which leads us to rely on the multiple imputation method which provides an articulate way for dealing with the missing variable of treatment assignment for regression estimation, but not for matching estimation. Especially, the conditioning variable for conditional mean independence assumption is given, not constructed by imputation. In other words, applying the multiple imputation method is more suitable for difference-in-difference regression than for matching estimation of synthetic control approach.

⁶ Although it is not written in bold font, ${\beta }_4$ is a vector of parameters which is taken for an inner product with ${\mathbf{X}}_{it}$.

⁷ The HRS data are available free of charge to registered users. The HRS (Health and Retirement Study) is sponsored by the National Institute on Aging (grant number NIA U01AG009740) and is conducted by the University of Michigan. The data were downloaded at http://hrsonline.isr.umich.edu.

⁸ As we utilize data of individual respondents who represent their own household so that a household does not appear in our sample more than once.

⁹ To control the possible influence of marital deduction of estate taxation that was effective independent of the TRA97, the variables of ‘Married’ and ‘Widowed’ are included. In addition, to control potential possibility that the parents of respondents might also be affected immediately by the TRA97, the binary indicator for whether the parent of each respondent is alive and older than 85 (‘Parent Older Than 85 is Alive’) is included.