Should the Japanese tax system be more progressive? An evaluation using the simulated SMCFs based on the discrete choice model of labor supply

Abstract

This study proposes a method to obtain the social marginal costs of public funds (SMCF) that allows for heterogeneity on a household basis as well as labor supply responses along both the extensive and the intensive margins. To demonstrate our methodology, we take the example of the 1999 national income tax reform in Japan and evaluate it by estimating the SMCFs for changing marginal tax rates in different income brackets. We estimate the discrete choice model (DCM) of labor supply using a 1997 data set of Japanese households, and we use the estimates to generate the SMCFs with a DCM micro-simulation. We evaluate the simulated SMCFs with various distributional weights and find that the value of the SMCF for a 1 % increase in the marginal tax rate in any given income bracket decreases as we move across brackets from the bottom to the top. This finding suggests that the national government should have made the Japanese income tax system more progressive rather than less progressive as carried out in the 1999 reform. Our method is readily transferrable to tax reforms in other countries as well.

Appendices

Appendix 1

Appendix 1 explains the method of obtaining each of the components in (9) and (10). After we estimate parameters in $U(\cdot )$, we calibrate the random component $e_{ij}$ in (2) as follows.

• [1 ] Draw a vector of $J$ random numbers that follow the I-EV distribution. Let $e_i^q\equiv [e_{i1}^q$, $e_{i2}^{q}$,..., $e_{iJ}^{q}$]’ be the $q$-th draw of such a vector. This yields a set of utility levels $U_i^q\equiv \left[ {U_{i1}^q,U_{i2}^q,\ldots ,U_{iJ}^q} \right]$’, where $U_{ij}^q\equiv V(h_{ij} ,\varvec{Z}_i ,\varvec{\tau } )\,+e_{ij}^q$. If observed labor choice $h_{i}$ coincides with the choice among the $J$ alternatives that picks up the maximum value in $U_i^q$, store $e_i^q$ vas a “successful” draw. Repeat this process to obtain $K$ successful draws $\left\{ {e_i^1,\ldots ,e_i^k,\ldots , e_i^K} \right\}$.

• [2 ] For a given successful vector draw ${\varvec{e}}_{i}^{k}$, we can construct pairs of utility level

  $$\begin{aligned} U_i^{0k} =V(h_i ,{\varvec{Z}}_i ,{\varvec{\tau }})+e_{i*}^k \end{aligned}$$

  (21)

  and individual tax liability

  $$\begin{aligned} R_i^0 =R(W_i h_i ,{\varvec{Z}}_i ,{\varvec{\tau }}) \end{aligned}$$

  (22)

  where “$j$ = *” in (21) indicates that the choice * is optimal for a vector of ${\varvec{e}}_{i}^{k}$. Since observed choice $h_{i}$ is the optimal choice by construction, different ${\varvec{e}}_{i}^{k}$ does not affect the optimal choice and the values of $V(h_i ,\varvec{Z}_i ,\varvec{\tau } )\hbox { and}\;R(W_i h_i ,\varvec{Z}_i ,\varvec{\tau } )$.

• [3 ] Change the tax parameters “slightly,” from $\varvec{\tau }$ to $\varvec{\tau }^{1}$. This will change the deterministic part of utility in (2) from $V(h_{ij}, \varvec{Z}_i, \varvec{\tau } )\;\hbox { to}\;V(h_{ij}, \varvec{Z}_i, \varvec{\tau }^1)$. Using ${\varvec{e}}_{i}^{k}$ above, we can predict a new labor choice $h_{i**}^{k}$ by selecting the choice that yields the maximum values among $\left\{ {U_{i1}^{1k},U_{i2}^{1k},\ldots ,U_{iJ}^{1k}} \right\}$, where $U_{ij}^{1k}\equiv V(h_{ij} ,\varvec{Z}_i ,\varvec{\tau }^1)\,+e_{ij}^k$. The new utility and tax liability are

  $$\begin{aligned} U_i^{1k}&= V( {h_{i**}^k,{\varvec{Z}}_i ,{\varvec{\tau }}^1})+e_{i**}^k\; {\hbox {and}} \end{aligned}$$

  (23)
  $$\begin{aligned} R_i^{1k}&= R(W_i h_{i**}^k ,{\varvec{Z}}_i ,{\varvec{\tau }}^1). \end{aligned}$$

  (24)

• [4 ] Simulate the “marginal” utility of income with a “small” lump-sum increase in nonlabor income $\Delta$, which yields $V^\lambda (\cdot )$. Using ${\varvec{e}}_{i}^{k}$ again, select $U_{i*}^{\lambda k}=\hbox { max}\{U_{i1}^{\lambda k},U_{i2}^{\lambda k},\ldots ,U_{iJ}^{\lambda k}\}$, where $U_{ij}^{\lambda k}\equiv V^\lambda (h_{ij}, \varvec{Z}_i ,\varvec{\tau } )\,+e_{ij}^k$, which yields an estimate for $\lambda _i^k$ as

  $$\begin{aligned} \lambda _i^k =\frac{U_{i*}^{\lambda k} -U_i^{0k} }{\Delta }. \end{aligned}$$

  (25)

Appendix 2

Appendix 2 explains our method of calculating tax liabilities and cash transfer benefits. Personal income tax liabilities consist of income tax imposed by the central government, inhabitant tax levied by local governments (prefectures and municipalities), and social insurance premiums for public pensions, health insurance, and unemployment insurance. The tax codes for FY 1997 are shown in Table 8. In calculating the tax liabilities, we make the following compromises due to data limitations. First, while local governments base their taxes on income in the previous year, we use the current income as a surrogate. Second, while local governments can change their tax rates, we use the standard uniform local tax rates as set by the national law, since most local governments adhere to the standard rates. Third, while public insurance premiums differ by place of work, we assume that combined premium rates for the three public insurances—those employed by firms with less than 1,000 employees, firms with 1,000 employees or more, and the public sector—are 13.3, 14.3, and 12.9 %, respectively, with specified ceilings on premium payments. Fourth, for deductions and exemptions, we consider employment income deduction, basic exemption, spousal exemptions, exemptions for dependents, and social insurance premium deductions, with the additional assumption that 20 % of nonlabor income is deductible.

Table 8 Outline of Japan’s income tax system, 1997 (thousands of yen)

For cash benefits, we consider child allowance. If households with children have annual earnings that are less than certain levels of income, they are entitled to child allowance. The eligibility depends on the level of household income as well as the number of children at specific ages. The income threshold varies as family size changes. In FY 1997, families with children aged three years or younger were eligible if their annual income was below 3.25 million yen (US$ 32,500), with the income threshold increasing with the number of dependents within the household. The 1997 monthly benefits per child depended on the number and composition of the children: 5,000 yen each for the first and second children and 10,000 yen each for the third and subsequent children.

The government also offers another type of child allowance called the child rearing allowance. However, since this targets single mothers only, we do not consider it here. Public assistance (PA) benefits may also be worth considering. PA benefits compensate for the difference between the minimum cost of living and the maximum possible earnings of a household. However, it is impossible to obtain individual PA benefits from the observed income data, as income is not the only variable used to determine eligibility for benefits. The means tests consider multiple characteristics of the applicants. The applicants have to exhaust all of their financial assets and prove that they have no support from their family and relatives. If the authority considers that the applicants are able to work, their chances for receiving benefits become quite slim. Fortunately, the proportion of households receiving PA in our samples is negligible; in 1997, the ratio of the number of households receiving PA to the total number of households was 1.41 %. In addition, the heads of 93.3 % of households receiving PA were the elderly ($\ge 65$ years if male and $\ge 60$ years if female), single mothers, the injured, the sick, and the disabled, all of whom are excluded from our samples. Therefore, the share of households receiving PA in our sample should be less than $.095\,{\%} [= .0141 \times (1 - .933)]$, and plausibly, even less. See Hayashi (2010) for a general explanation about Japan’s social protection system.