Intensive margin, extensive margin, and allowance for spouses: A discrete choice analysis Shun-ichiro Bessho∗
Masayoshi Hayashi†
January 25, 2013
Abstract This paper investigate the effects on married women’s labor supply of Japanese tax reform, abolishment of allowances for spouses, based on a micro-simulation method. To do so, we estimate a structural discrete choice household labor supply model. We take into account husband’s response and the extensive margin of labor supply, assuming fixed cost of working and a unitary household model. Our estimates of the average own intensive elasticity are 0.032 for husbands and 0.031 for wives, while that on the extensive margin is 0.009 for husbands and 0.055 for wives. Our simulations show that the complete abolishment of allowances for spouses increase by 1.62% the average working hours of wives, consistent with the previous papers, which ignore the extensive margin and find the positive effects on labor supply of such abolishment. JEL: J20, H24 Key Words: Female labor supply; discrete choice model; tax reform; Japan.
1 Introduction The evaluation of the household response to taxation has profound implications for the assessment of tax policy. A progressive tax on labor income have the possible negative effects on ∗
Faculty of Economics, Keio University, 2-15-45, Mita, Minato-ku Tokyo, 108-8345 Japan.
[email protected] † Faculty of Economics, The University of Tokyo
1
E-mail:
labor supply. Japanese income taxation system, which has a progressive tax rate structure as well as unique and complicated allowances and deductions, is no exception. In particular, the allowances for spouses of Japanese income tax system create a non-convex and piecewise linear budget line and could depress labor supply of married women. This topic has been investigated intensively by Japanese labor economists (e.g., Akabayashi 2006, Abe 2009, Takahashi 2010). They all observe a negative effects of the allowances for spouses on labor supply of married women. As Takahashi (2010) points out, structural estimations employed in Akabayashi (2006) and Takahashi (2010) find more modest impact than those with reduced-form equations. These previous researches maintain two assumptions that would not be plausible. First, they only focus on wife’s labor supply behavior and ignore the possibility that husband could respond to the tax reform as well, presuming the husband’s income is fixed. If a couple maximizes the household’s utility jointly or interacts strategically, it would not be the case. Second, a continuous labor supply function is assumed. In other words, only the intensive margin, i.e., hours-of-work margin is taken into account, and the extensive margin, i.e., participation margin is assumed to be zero. When labor supply of married women is not necessarily large enough, the extensive margin could play a important role. Given this state of Japanese literature, the purpose of this paper is to examine the effect of the allowances for spouses on labor supply, considering husband’s response and the extensive margin of labor supply. To do so, we estimate a structural discrete choice household labor supply model a la Van Soest (1995). This kind of discrete choice household labor supply models is robust to (i) the non-convexities of budget sets (Blundell and MaCurdy 1999), (ii) changes in the number of choices (Flood and Islam 2005), (iii) measurement errors in wages and labor hours (Flood and Islam 2005), and (iv) unobserved heterogeneity (Haan 2006). To evaluate the effect, we employ a micro-simulation method developed by Creedy and Kalb (2006), using the estimated parameters. The contribution of this paper is twofold. First, we take into account husband’s response and the extensive margin of labor supply, assuming fixed cost to enter a labor market and a unitary household model. When there is a fixed cost to participate labor market, she would not choose a small amount of labor supply, rather does not work at all. This generates a positive participation margin. When a couple maximize the household’s utility that is drawn from consumption and leisure of each member, a tax reform on allowances for spouses could affect labor supply of
2
both husband and wife. Thus we do not assume the husband’s labor supply is fixed. Second, the labor supply elasticity with respect to wage rate are computed. Although the labor supply elasticity is one of key parameters to evaluate a public policy, only a few Japanese studies on labor supply behavior adequately include the effect of the tax system on the budget constraints of households (Bessho and Hayashi 2011). We provide elasticities of gross labor supply on both intensive and extensive margin with respect to own wage rate, as well as “cross” elasticity that, for example, describes the husband’s response to a change in wife’s wage rate. Our estimates of the average own intensive elasticity is 0.032 for husbands and 0.031 for wives, while that on the extensive margin is 0.009 for husbands and 0.055 for wives. These elasticities seem to be modest compared to the previous research in Europe and America, but we believe that these values are quite plausible. Our simulations show that the abolishment of allowances for spouses increase the average labor supply of wives. The average labor working hours of wives whose labor incomes are less than 1.03 million yen and whose annual working hours are less than 1,550 hours increase by 0.13% responding to the abolishment. These results are consistent with, but more modest compared to the Japanese literature, which find the positive effects on labor supply of such abolishment. This difference may be due to the fixed costs associated with positive hours worked, which creates the positive extensive margin of labor supply. This paper is organized as follows. The following section (Section 2) provides a brief explanation of the institutional background. Section 3 presents an econometric specification and micro-simulation method of a labor supply model. Section 4 explain our data, and the results are shown in Section 5. Section 6 concludes.
2 Institutional backgrounds Japanese income tax system assumes individuals as tax units. The amount of income tax for each individual is calculated as follows. Income tax on individuals include “income tax”, a national tax, and “inhabitants tax”, a local tax. The principle to compute the amount of tax is almost the same between income tax and inhabitants tax. First, “employment income ” is calculated as the salaries the individual receives minus “employment income deduction” in case of employed workers. Second, “taxable income ” is defined as the “employment income”
3
minus some kinds of deductions and allowances, plus taxable non-labor income. Finally, the tax rates are applied to taxable income and some tax credits are subtracted, if any, to obtain the tax amounts. In addition, employed workers must pay the social insurance premiums, which we take here as tax. The social insurance premium differ as places of work differs. Allowances for spouses and special allowance for spouses are applied when “taxable income” is computed(1) . Assume a husband to be a household’s primary earner, his wife to be a secondary earner, and both to be employed workers. Under the tax code in 1997, when our data was collected, if the wife’s “taxable income” is less than 760 thousand yen (in case of the national income tax) and if the husband’s “taxable income” is less than 10 million yen, allowance for spouses and special allowance for spouses are applied for the husband’s tax calculation. If the wife has no other income, 760 thousand yen of “taxable income” means 1.41 million yen of her gross income, because of employment income deduction. The amount of sum of allowance for spouses and special allowance for spouses is phased out almost linearly as the wife’s gross income increases, as long as the husband’s taxable income is less than 10 million yen (Figure 1). The local inhabitants tax has the same allowances, but the amount is smaller than the national income tax by 50 thousand yen. Japanese social insurance system assumes individuals as units, as well, but has special treatment for spouses. In our case, if the annual personal gross income of the wife is below a certain threshold (1.3 million yen), she is eligible for “Category III” insured, dependent on her husband, and does not have to pay the premium. These tax and social insurance systems are well known to married women in Japan (Akayabashi 2006) and thus may well affect the labor supply behavior of the married couples. As Akabayashi (2006) and Takahashi (2010) put emphasis, these system makes the wife’s budget line non-convex and could depress their labor supply because of a dip in the budget set.
(1)
See Ayabayashi (2006) and Takahashi (2010), among others, for details.
4
3 Model 3.1 Econometric specification We apply a structural discrete choice household labor supply model following van Soest (1995). Houdehold i consumes a numer´aire xi and leisure of husband lhi and wife lwi to obtain utility ui = ui (xi , lhi , lwi ). We assume a unitary model in the sense that household members jointly maximize the utility, ui , given the before-tax wage rate and tax codes. The time endowment is expressed as T so that husband’s hours worked is given as hhi = T − lhi , and wife’s as hwi = T − lwi . xi is equal to the family’s after tax income, including husband’s and wife’s earnings. Denoting the parameters of income tax code as τ , the family’s after income tax income is represented as: xi = [Whi hhi − T (Whi hhi , Wwi hwi ; Zi , τ )] + [Wwi hwi − T (Wwi hwi , Wwi hwi ; Zi , τ )] (1) where T (·) is a income tax function, Zi is a vector of the family’s characteristics and Whi and Wwi are pre-tax wage rate of husband and wife, respectively. Note that husband’s (wife’s) income tax depends on wife’s (husband’s) income in Japan through, for example, allowance for spouses. Because the family’s utility depends on the tax code, we can write the utility as ui = u(xi , lhi , lwi ; Zi , τ ). The choice of labor supply is discretized, so that each household is assumed to choose among the alternatives in the choice set of income leisure combinations {(xij , lhij , lwij ) : j = 1, 2, ..., J} to maximize u(xij , lhij , lwij ; Zi , τ ). We work with the standard translog specification of the direct utility function: u(·; ·) = V (xij , lhij , lwij |Zi , τ ) + εij = βx ln xij + βh ln lhij + βw ln lwij + βxx (ln xij )2 + βhh (ln lhij )2 + βww (ln lwij )2 +βxh (ln xij )(ln lhij ) + βxw (ln xij )(ln lhij ) + βhw (ln lhij )(ln lwij ) +βhf 1(hhij > 0) + βwf 1(hwij > 0) + εij
(2)
where εij is an additive random disturbance and βs are coefficients to be estimated. The last two terms reflect the fixed cost associated with working, thus the expected signs of βhf and βwf are negative.
5
We assume that household characteristics linearly affect five of the coefficients, (βx , βh , βw , βhf , βwf ). These coefficients are assumed to be dependent on the husband’s and wife’s respective ages, the number of children, and the place of residence. While we consider other cases where some of these five coefficients are independent of household characteristics, the likelihood ratio tests emphatically reject the restricted models. The random disturbance, εij , follows the type I extreme value distribution identically and independently. Because the family choose j for which the utility is the largest, we can use the multinomial logit model to estimate βs (van Soest 1995, Creedy and Kalb 2006).
3.2 Tax reform simulations Our labor supply model can be interpreted as a structural one, we can simulate the labor supply behavior when the tax code τ changes. Three tax reforms are simulated. The new allowance schedules are shown in Figure 1. First, the special allowance for spouses is partly abolished, that is, the special allowance applied for those with spouses with “taxable income” less than 380 thousand yen. The allowance become constant until the spouse’s taxable income become 380 thousand yen, and phased out almost linearly as the spouse’s gross income increases. The phase-out part is unchanged. This reform was actually implemented in 2004. Second, the allowance for spouses is abolished, and the special allowance for spouses is partly abolished. In this case, the allowance become phase out from the beginning and those with spouse’s taxable income more than 380 thousand yen become ineligible for the allowance. In other words, the allowance schedule shifts parallel to the origin. Third, both allowance for spouses and special allowance for spouses are abolished. While the manifesto of Democratic Party of Japan included this tax reform in 2009 General Election, this reform has not been implemented. Allowances for spouses and special allowance for spouses are applied for both national “income tax” and local “inhabitants tax”. Since these allowances for these taxes have been changed simulataneously in the past Japanese tax reforms, the above-described reforms are applied for both national and local taxes in the following simulations.
6
3.3 Micro-simulation method We base our calculation on the behavioral DCM micro-simulation (Creedy and Kalb 2006). Recall that utility (2) consists of a deterministic part and a random component εij . First, we estimate the parameters of the utility function (2), as explained in Section 3.1. Second, we calibrate the random components so that they become the observed labor choices. We draw vectors of J random numbers that follow the the type I extreme value distribution for each household i. Let εqi ≡ [εqi1 , εqi2 , · · · , εqiJ ]′ be the q-th draw for such numbers, which yields uqi ≡ [uqi1 , uqi2 , · · · , uqiJ ]′ , where uqij ≡ V (xij , lhij , lhij |Zi , τ ) + εqij . Store εqi as a “successful” q∗ q draw if the labor choice from this draw (hq∗ hi , hwi ) ≡ argmax{ui } coincides with observed
(hhi , hwi ). If not, discard such εqi and try again. This process is repeated until either “successful” draw(2) . We repeat this process to obtain K successful draws εi ≡ vec[ε1i , · · · , εki , · · · , εK i ], where k indexes the draws in place of q. Here we set K = 100. Third, we change the tax parameters τ to τ 1 so that allowance for spouses and special allowance are for spouses are changed as explained above. This changes the deterministic part of the utility from V (xij , lhij , lwij |Zi , τ ) to V (xij , lhij , lwij |Zi , τ 1 ). The k-th draw εki yields 1k 1k ′ 1k 1 k u1k ≡ [u1k i i1 , ui2 , · · · , uiJ ] with uij ≡ V (xij , lhij , lwij |Zi , τ ) + εij and the optimal choice 1∗ 1k becomes (h1∗ hi , hwi ) ≡ argmax{ui }. 1∗ Since there are K successful draws, there are at most K pairs of (h1∗ hi , hwi ) for each house-
hold(3) . These K pairs constitute the empirical distribution of the labor supply after the tax change. We calculate labor supply elasticity with respect to wage rate for reference, using the same micro-simulation method where we raise the before-tax wage rates by 1%. We focus on the aggregate labor supply, and the intensive and extensive margin elasticities, as defined by Kleven and Kreiner (2006), are calculated for K successful draws. (2)
(4)
Since both husband and wife
If “successful” is not obtained after 100 trial, such household i is discarded. Less than 10% of the sample is
discarded for each draw. (3) Successful draws are not obtained for a small portion of the sample for each draw after the trials, thus there are less than K pairs for many households. See footnote (2). (4) Let superscript 0 denote a variable before the increase in wage rate, and superscript 1 denote that after the increase. The aggregate labor supply, H, is the product of the fraction of individuals who participate in the labor ¯ The rate of change of aggregate labor supply, market, P , and the average labor supply of the participants, h. (H 1 − H 0 )/H 0 , is decomposed as the sum of that of participation rate, (P 1 − P 0 )/P 0 , and that of the average
7
can work in this setting, we calculate “cross” elasticity, husband’s labor supply elasticity with respect to wife’s wage rate and wife’s elasticity with respect to husband’s wage rate, as well.
4 Data 4.1 Sample The data used in the sample are from Syugyo Kozo Kihon Chosa [Employment Status Survey] conducted by the Statistical Bureau of the Japanese Government in 1997. This survey is conducted every five years and the most comprehensive labor survey in Japan: it produces a large sample that contains about 11 million individual observations with a variety of household characteristics. We focused on the labor supply of nuclear families whose heads are prime age (25-55) males. We omitted the following observations from the sample: (a)self-employed workers, (b) board’s members of private companies and non-profit organization, (c) family workers for SMEs, (d) the unemployed due to illness, (e) those who had changed residence or job within one year, and (f) those who had children within one year. These omissions reduces the sample size down to 43,011 . The ESS codes hours worked as interval data. Using these intervals, we set up eight choices of hours worked per year as shown in Table 1. Note that the choice set contains 8 × 8 = 64 alternatives for married couples. We set T = 16 × 365 = 5, 840 hours per year. The variables included in Zi are standard in the literature. They include the dummies for the following variables: five age groups (30-34, 35-39, 40-44, 45-49, 50-54), residence in one of the three major urban areas (the Greater Tokyo area, the Chukyo area, and the Kinki area), and three educational backgrounds (junior high school graduates, junior college graduates, and university graduates or higher). They also include the numbers of each of the three groups of children aged 6 and below, 7-14, and 15 and above. The sample statistics are shown in Table 2.
¯1 − h ¯ 0 )/h ¯ 0 . The former one corresponds to the extensive margin, and the latter labor supply of the participants (h one does to the intensive margin.
8
4.2 Before-tax wage rate We use as before-tax wage rate predicted values. Since the data for days worked and labor income are provided as intervals, we first calculate before-tax wage rate as quotient of middle values of days worked and labor income. The predicted before-tax wage rate is defined as a fitted value of a wage rate regression for each gender where the dependent variable is log of the before-tax wage rate and the explanatory variables include dummies for age, residence, education and their cross term. Since nonnegligible portion of wives choose zero hours worked, the wage rate regression for females is estimated by Heckit sample selection model, where excluded instruments are quadratic terms of residuals obtained from a regression for non-labor income (family income minus husband’s labor income).
4.3 Tax code To estimate the direct utility function, we need to know the family budget set computing family’s after-tax income, xij , for each alternative in the choice set. Income tax on individuals include national “income tax”, local “inhabitants tax”(5) and social insurance premium, as described in Section 2. The principle to compute the amount of tax is almost the same between income tax and inhabitants tax. First, we derive “employment income ” as the salaries the individual receives minus “employment income deduction”. Second, “taxable income ” is defined as the “employment income” minus some kinds of deductions and allowances including allowance for spouses and special allowance for spouses, plus taxable non-labor income. Here, we assume that 20% of non-labor income is deductible. Finally, we apply the tax rates to taxable income and subtract some tax credits, if any, to obtain the tax amounts. The amounts and rate are shown in Table 3. The available deductions, allowances or tax credits differ as individual characteristics differ. Thus, we cannot take into account some of them because of data limitation. What we employ are basic allowance, allowance for spouses, special allowance for spouses, allowance for depen(5)
The amount of inhabitants tax is calculated based on the income in the previous year in practice. Since our data
sets is not panel data, however, inhabitant tax is assumed to be computed using the current income. In addition, while prefectures and municipalities can charge their own tax rates, we use the standard tax rates uniformly set by the national law. However, this should not cause us a major problem since most local governments stick to these standard rates.
9
dents, employment income deduction and deduction for social insurance premiums. We assume public pension insurance, public health insurance and public unemployment insurance as social insurance. The premiums of social insurance differ as places of work differs. The data does not contain, however, such information needed to calculate social insurance premium. We assume that the social insurance premium is 13.3% if the firm where the individual works employs less than 1,000 people, 14.3% if more than 1,000 people, 12.9% if the individual is a public servant. We consider the upper limit of the social insurance premium.
5 Results 5.1 Labor supply elasticity The log-quadratic specifications used in this paper do not impose a priori restrictions such as the positive marginal utility of income or the quasi-concavity of preferences. Using the parameters shown in Table 4, we check these two properties. We calculate HC and du/dy, defined in Eq. (3) and Eq. (4) in Van Soest (1995), for all individual households in our sample. If HC is positive, the preferences of the households are quasi-concave. If du/dy is positive, the utility of the households increases with income. While there are households with negative values, the proportion of such households is just 0.34%. Most households thus display quasi-concave utility functions that increases with income. We compute the elasticities of aggregate labor supply using the micro-simulation method described in Section 3.3, which provides us K = 100 sets of the simulated elasticities. Table 5 presents the average and standard deviation of those elasticities. Panel A of Table 5 shows the elasticities with respect to own wage rate, the average of which is 0.041 for husbands, 0.087 for wives. The elasticities of the intensive margin, i.e., the hours-of-work elasticities are shown in the third column of Panel A of Table 5. The average own elasticity is 0.032 for husbands and 0.031 for wives. Both values are small compared to those obtained in the discrete-choice models, which range between 0.0 and 0.3 for husbands and between 0.1 and 0.7 with a few exceptions (see Tables 1, 2 and 4 in Bargain et al. 2011). The elasticities of the extensive margin, i.e., the participation elasticities are shown in the
10
rightmost column of Panel A of Table 5. The average own elasticity is 0.009 for husbands and 0.055 for wives, which are again very modest results. Indeed, the discrete-choice labor supply models show larger elasticities ranging between 0.0 and 0.2 for husbands and between 0.1 and 1.0 for wives (see Tables 1, 2 and 4 in Bargain et al. 2011). Although our elasticities of extensive margin are small in comparison with the discrete-choice literature, they seem comparable to corresponding values on intensive margin (0.032 for husbands and 0.031 for wives). This nonnegligible magnitude of the participation elasticities may be due to the fixed utility cost of working. Because we assume that husband and wife jointly maximize the household’s utility, a change in the husband’s gross wage rate can affect the wife’s labor supply behavior, and vice versa. Thus the cross-elasticities, i.e., those of the wife’s labor supply with respect to the husband’s wage rate and those of the husband’s labor supply with respect to the wife’s wage rate, can be defined. These cross-elasticities, shown in Panel B of Table 5, are −.007 for husband’s and .169 for wives, respectively. These results imply that a husband responds to an increase in his wife’s wage by reducing his working hours, while his wife responds to a similar increase in his wage in the opposite manner. Although Bargain et al. (2011) do not list the values for cross-wage elasticities, Van Soest (1995, Table 4) finds that the cross-elasticities of husband’s are negative for 80 percent confidence interval of the three different models. While the signs are same in the husband’s case, our average (−.007) is smaller than his smallest median (−.015). In contrast, our average estimates of the wife’s cross-elasticities are positive (.169) and larger than the largest median (.051) of Van Soest (1995).
5.2 Tax reform simulations Using the parameters shown in Tables 4, we simulated three tax reforms, all of which cut back the allowances for spouses described in Section 3.2 and Figure 1. The first one (Reform 1) abolish the special allowance applied for those with spouses with “taxable income” less than 380 thousand yen. The second (Reform 2) makes the allowance schedule shift parallel to the origin. The third (Reform 3) abolish both allowance for spouses and special allowance for spouses. The simulated distributions of wife’s labor supply are shown in Panel A of Table 6(6) , while (6)
The observations for which “successful” random components are not obtained after 100 trial are discarded in
11
those of husband’s are in Table 7. Reform 1 and 3 generate similar results. The ratio of nonparticipants, i.e., those with zero labor supply decrease by 0.5 percent points and that of those with short working hours also decrease by a small portion, less than 0.1 percent point. In addition, the ratio of those whose annual hours of work are more than 1,250 hours increase by 0.4 to 0.5 percent points. These changes in distributions make average annual working hours increase by 1.6 percents. Although abolishment of allowances for spouses changes after-tax income of alternatives of all households, this may especially affect wives with relatively small labor income and thus with husbands eligible for these allowances more than others. Panel B of Table 6 shows simulated distribution of labor supply of wife who originally works with their labor income less than 1.03 million yen, with annual working hours less than 1,550 hours and with husbands whose income is less than 10 million yen. 1.03 million yen is perceived as “ceiling” for wife’s working hours (Abe 2009). Despite that the ratios of non-participants slightly increase, the average annual working hours increase by 0.1 to 0.2 percents for Reform 1 and 3. The magnitude of the abolishment effect in this paper is smaller than the previous studies. Akabayashi (2006) and Takahashi (2010) simulate wife’s labor supply behavior when the allowances are abolished and find that the abolishment increase hours worked by 5.53% (Akabayashi 2006) or 0.7% (Takahashi 2010). Since these existing estimates are obtained for the sample of working wives, the figures should be compared with those in Panel B of Table 6. As Takahashi (2010) summarizes, these two estimates seems even modest compared with other estimates. This difference come from the difference in wage elasticity estimates. Our estimated intensivemargin elasticity of wives is 0.031, which is smaller than Akabayashi’s (2006) estimate of 0.16 and Takahashi’s (2010) estimate of 0.19. This is perhaps due to the estimation method. They consider a usual linear labor supply function and utilize an estimation method a la Hausman (1979), assuming out the fixed cost. In addition, they might not take the local tax into account. Our small estimates might also be due to the discrete nature of labor supply choice, though we believe this is not problematic to consider the aggregated labor supply because of enough number of draws. The simulation for Reform 2 provides a little different picture, showing that wife’s labor the draw.
12
supply decrease slightly in both Panel A and B of Table 6. This can be because of the fixed utility cost of working, that is, positive participation elasticity and small intensive-margin elasticity. Since our estimation includes the fixed cost, wives can leave the labor market if their labor incomes are not sufficiently large. To see this, we computed simulated distribution conditional on original (before reform) labor supply choice for Reform 2. The sample is same as the one used in Panel B of Table 6. Due to the discretized nature of labor supply, more than 99% of wives do not change their choice after the reform. Panel B of Table 8 displays the shares of those who changed their choice. This panel shows that 53.2% of those whose original annual working hours less than 900 hours and who change the labor supply choose zero working hours after Reform 2. In case of those with original working hours between 1,250 to 1,550 hours, the corresponding ratio is 62.6%. Since the usual income effect works, some workers do increase the labor supply, as the previous researches describe. In a Round-table talk of labor and management, participants says that the abolishment of these allowances should make some people work longer, others shorter (Hirano et al. 2010, p.63). Panel B of Table 8 is just consistent with this view. Table 9 shows the simulated average household income and tax payment. Although the aggregate labor supply slightly increases, the before-tax incomes decrease because of the changes of labor-supply distribution. In all of three cases, tax payments increase. As a result, the reforms makes after-tax incomes smaller.
6 Conclusion This paper investigate the effects of Japanese tax reform, abolishment of allowances for spouses, based on micro-simulation method. To do so, we estimate a structural discrete choice household labor supply model assuming a unitary household model. Our estimates of the average own intensive elasticity are 0.032 for husbands and 0.031 for wives, while that on the extensive margin is 0.009 for husbands and 0.055 for wives. Our simulations show that the abolishment of allowances for spouses can raise the aggregate labor supply of wives. The complete abolishment of allowances for spouses increase by 1.6% the aggregate labor supply and by 0.1% the average working hours of wives whose labor incomes are less than 1.03 million yen and whose annual working hours are less than 1,550 hours. These results are even more modest than previous
13
papers, which find the positive effects on labor supply of such abolishment. We also point out the possibility that the allowance abolishment could decrease the women’s labor supply. This difference may be due to the fixed cost associated with positive hours worked, which is ignored in the Japanese literature. Our analysis has some limitations. First, our labor supply is static and does not take a dynamic decision making into account (Abe 2009). Second, we assume a unitary household whose members jointly maximize the household’s utility. A strategic interactions within a household may matter as discussed in Bargain et al (2006). Third, we assume a perfect knowledge of tax codes and ignore the possibilities of tax avoidance or evasion of households. All these are topics for our future research.
References [1] Abe, Y. 2009. The effects of the 1.03 million yen ceiling in a dynamic labor supply model. Contemporary Economic Policy 27(2), 147-163. [2] Akabayashi, H. 2006. The labor supply of married women and spousal tax deductions in Japan — a structural estimation. Review of Economics of Household 4, 349-378. [3] Bargain, O., Beblo, M., Beninger, E., Blundell, R., Carrasco, R., Cuiuri, M-C., Laisney, F., Lechene, V., Moreau, N., Myck, M., Ruis-Castillo, J., Vermeulen, F. 2006. Does the representation of household behavior matter for welfare analysis of tax-benefit policies? An introduction. Review of Economics of Household 4, 99-111. [4] Bargain, O., Orsini, K., Peichl, A. 2011. Labor supply elasticities in Europe and the US. IZA Discussion Paper Series No. 5820. [5] Bessho, S., Hayashi, M. 2011. Labor supply response and preferences specification: Estimates for prime-age males in Japan. Journal of Asian Economics 22(5), 398-411. [6] Blundell, R., MaCurdy, T. 1999. Labor supply: A review of alternative approaches. Ashenfelter, O., Card, D., (Eds.) Handbook of Labor Economics 3A, 1559-1695. [7] Creedy, J., Kalb, G. 2006. Labor Supply and Microsimulation: The Evaluation of Tax Policy Reforms. Edward Elgar Publishing. 14
[8] Flood, L., Islam, N. 2005. A Monte Carlo evaluation of discrete choice labor supply models. Applied Economics Letters 12(5), 263-266. [9] Haan, P. 2006. Much ado about nothing: conditional logit vs. random coefficient models for estimating labour supply elasticities. Applied Economics Letters 13, 251-256. [10] Hausman, J.A. 1979. The econometrics of labor supply on convex budget sets. Economics Letters 3, 171-174. [11] Hirano, M, Staffs on Human Resources Department/Union Officials. 2010. The 1.03/1.3 Million Yen Borderline: Employment Management and Worker. Japanese Journal of Labor Studies 52(12), 54-67. [12] Kleven, H. J., Kreiner, C. T. 2006. The marginal cost of public funds: Hours of work versus labor force participation. Journal of Public Economics 90, 1955?1973. [13] Takahashi, S. 2010. A structural estimation of the effects of spousal tax deduction and social security systems on the labor supply of Japanese married women. GSIR Working Papers, Economic Analysis and Policy Studies EAP10-4, International University of Japan. [14] Van Soest, A. 1995. Structural models of family labor supply: A discrete choice approach. Journal of Human Resources 30, 63?88.
15
Figure 1. Allowance schedule
Husband's allowance Current allowance schedule
Reform 1
Special Allowance for spouses
Allowance for spouses
Reform 2
Wife's taxable income
- 16 -
Table 1. Discretized level of annual hours worked Choices
Annual Hours Worked
Lower Limit (hours)
Upper Limit (hours)
1 2 3 4 5 6 7 8
0 492 1,177 1,484 1,741 1,849 2,140 2,679
0 1 901 1,251 1,551 1,751 2,001 2,201
0 900 1,250 1,550 1,750 2,000 2,200 3,000
Total
- 17 -
Ratios in samples Couple: Couple: Husband (%) Wife (%) 2.14 1.37 26.63 25.79 13.62 8.53 10.79 11.13
41.98 16.04 21.79 10.39 3.85 2.85 1.97 1.13
100.00
100.00
Table 2. Sample statistics mean
s.d.
min
median
max
1{25 ≤ age ≤ 29}
.028
0.164
0
0
1
1{30 ≤ age ≤ 34} 1{35 ≤ age ≤ 39} 1{40 ≤ age ≤ 44} 1{45 ≤ age ≤ 49} 1{50 ≤ age ≤ 54}
.044 .075 .182 .349 .322
0.206 0.263 0.386 0.477 0.467
0 0 0 0 0
0 0 0 0 0
1 1 1 1 1
1{junior high school} 1{senior high school} 1{junior college} 1{university}
.144 .451 .055 .350
0.351 0.498 0.227 0.477
0 0 0 0
0 0 0 0
1 1 1 1
1{25 ≤ age ≤ 29}
.053
0.224
0
0
1
1{30 ≤ age ≤ 34} 1{35 ≤ age ≤ 39} 1{40 ≤ age ≤ 44} 1{45 ≤ age ≤ 49} 1{50 ≤ age ≤ 54}
.029 .089 .295 .377 .152
0.168 0.285 0.456 0.485 0.359
0 0 0 0 0
0 0 0 0 0
1 1 1 1 1
Wife: Education
1{junior high school} 1{senior high school} 1{junior college} 1{university}
.103 .505 .244 .148
0.305 0.500 0.430 0.355
0 0 0 0
0 1 0 0
1 1 1 1
Number of Children
age ≤ 6 7 ≤ age ≤ 14 15 ≤ age
.153 .474 .777
.448 .776 .854
0 0 0
0 0 1
4 5 4
Residence in one of the three major urban areas
.629
.483
0
1
1
Husband: Gross wage rate (fitted) (10,000 yen/hour)
.427
.123
.182
.424
.753
Wife: Gross wage rate (fitted) (10,000 yen/hour)
.060
.042
.003
.049
.276
Total household income (10,000 yen/year)
951
605
25
850
3,000
Husband: Age
Husband: Education
Wife: Age
The sample size is 43,011.
- 18 -
Table 3. Outline of income taxation system, 1997 (Thousands of yen) Income tax
Inhabitants tax
Basic Deduction Exemption for Spouses Special Exemption for Spouses Exemption for Dependents (for specific dependents)
380 380 380 380 530
330 330 330 330 410
Employment Income Deduction
Not over 1,800, 40% Not over 3,600, 30% Not over 6,600, 20% Not over 10,000, 10% Over 10,000, 5%
Not over 1,800, 40% Not over 3,600, 30% Not over 6,600, 20% Not over 10,000, 10% Over 10,000, 5%
650
650
Not over 3,300, 10% 5% Over 3,300, 20%
Not over 2,000,
Lower limit Tax rates
- 19 -
Over 2,000,
const. # children age ≤ 6 # children 7 ≤ age ≤ 14 # children 15 ≤ age Husband: 30 ≤ age ≤ 34 Husband: 35 ≤ age ≤ 39 Husband: 40 ≤ age ≤ 44 Husband: 45 ≤ age ≤ 49 Husband: 50 ≤ age ≤ 54 Wife: 30 ≤ age ≤ 34 Wife: 35 ≤ age ≤ 39 Wife: 40 ≤ age ≤ 44 Wife: 45 ≤ age ≤ 49 Wife: 50 ≤ age ≤ 54 Husband: junior high school Husband: junior college Husband: university Wife: junior high school Wife: junior college Wife: university Residence in urban areas
Table 4. Estimated parameters βx βh βw 10.667 61.423 156.546 (1.866) (3.201) (3.908) 0.042 -0.458 -0.738 (0.018) (0.113) (0.171) 0.011 -0.154 0.615 (0.009) (0.065) (0.055) 0.017 -0.192 -0.097 (0.009) (0.058) (0.048) -0.021 -3.467 (0.045) (0.388) -0.013 -3.094 (0.042) (0.308) -0.062 -2.963 (0.031) (0.245) -0.032 -1.294 (0.020) (0.164) -0.007 -0.529 (0.016) (0.112) -0.077 -0.559 (0.042) (0.223) -0.106 -1.460 (0.047) (0.286) 0.003 -1.282 (0.037) (0.206) -0.047 -0.652 (0.022) (0.128) -0.040 -0.320 (0.018) (0.107) -0.059 -1.201 (0.016) (0.180) -0.013 0.017 (0.026) (0.198) -0.007 -0.134 (0.017) (0.139) -0.063 -0.818 (0.016) (0.131) 0.014 0.698 (0.017) (0.103) 0.055 0.625 (0.028) (0.166) 0.015 0.386 -2.380 (0.012) (0.105) (0.080)
- 20 -
βhf 0.153 (0.046) -0.403 (0.142)
βwf -0.961 (0.026) -0.903 (0.046)
Table 4. (contd) Estimated parameters
const.
const.
βxx 0.060 (0.004) βxh -0.573 (0.381)
βhh -9.357 (0.289) βxw -1.820 (0.122)
βww -20.775 (0.475) βhw 3.487 (0.341)
Note: Standard errors are in parentheses. Log-likelihood is -151399.2. Sample size is 43,011.
- 21 -
Table 5. Wage elasticities Panel A. Change in own wage rate Total
Intensive
Extensive
Husband
.041 (.010)
.032 (.009)
.009 (.005)
Wife
.087 (.031)
.031 (.016)
.055 (.021)
Panel B. Change in spouse’s wage rate Total
Intensive
Extensive
Husband
-.007 (.007)
-.003 (.006)
-.004 (.004)
Wife
.169 (.035)
.063 (.019)
.105 (.024)
Note: Standard deviations of 100 successful draws are in parentheses.
- 22 -
Table 6. Simulated distribution of wife’s labor supply Panel A. Total Choices 1 2 3 4 5 6 7 8
Annual Hours Worked 0 492 1177 1484 1741 1849 2139 2679
Original
Reform 1
Reform 2
Reform 3
46.75 16.66 21.92 9.39 2.80 1.81 0.64 0.03
46.20 16.63 22.03 9.57 2.93 1.91 0.69 0.03
46.79 16.65 21.85 9.38 2.82 1.82 0.65 0.03
46.21 16.62 22.01 9.56 2.94 1.93 0.70 0.04
576.03
585.05
575.90
585.36
(std dev)
---
1.39
1.33
1.43
Rate of change (%)
---
1.57
-0.02
1.62
Average hours worked
Note: Unit is percent, except for average hours worked (annual hours). This table shows average densities of each choice for 100 draws. Panel B. “Eligible” wives Choices 1 2 3 4 5 6 7 8
Annual Hours Worked 0 492 1177 1484 1741 1849 2139 2679
Original
Reform 1
Reform 2
Reform 3
0.00 36.89 45.34 17.77 0.00 0.00 0.00 0.00
0.14 36.61 45.21 17.84 0.09 0.08 0.04 0.00
0.23 36.85 45.16 17.69 0.02 0.02 0.02 0.00
0.23 36.58 45.15 17.80 0.10 0.09 0.05 0.01
Average hours worked
978.68
980.58
976.43
979.98
Rate of change (%)
---
0.19
-0.23
0.13
Note: Unit is percent, except for average hours worked (annual hours). This table shows average densities of each choice for 100 draws. Sample is limited to those who under current system works with their labor income less than 1.03 million yen and with annual working hours less than 1,550 hours. - 23 -
Table 7. Simulated distribution of husband’s labor supply Panel A. Total Choices 1 2 3 4 5 6 7 8
Annual Hours Worked 0 492 1177 1484 1741 1849 2139 2679
Original
Reform 1
Reform 2
Reform 3
1.07 1.25 28.00 27.20 14.22 8.67 10.59 9.00
1.11 1.40 27.91 27.10 14.23 8.73 10.60 8.92
1.09 1.32 27.99 27.17 14.21 8.67 10.58 8.96
1.13 1.79 28.08 26.99 14.08 8.64 10.48 8.81
Average hours worked
1614.77
1612.19
1613.12
1604.62
Rate of change (%)
---
-0.16
-0.10
-0.63
Note: Unit is percent, except for average hours worked (annual hours). This table shows average densities of each choice for 100 draws. Panel B. with “Eligible” wives Choices 1 2 3 4 5 6 7 8
Annual Hours Worked 0 492 1177 1484 1741 1849 2139 2679
Original
Reform 1
Reform 2
Reform 3
0.75 1.33 32.42 31.03 13.20 9.39 9.02 2.85
0.79 1.45 32.35 30.95 13.20 9.40 9.01 2.85
0.79 1.43 32.39 30.97 13.20 9.38 9.00 2.84
0.82 1.86 32.46 30.79 13.06 9.30 8.91 2.81
Average hours worked
1521.25
1519.59
1519.59
1512.94
Rate of change (%)
---
-0.11
-0.11
-0.55
Note: Unit is percent, except for average hours worked (annual hours). Sample is limited to those with such wives who under current system works with their labor income less than 1.03 million yen and with annual working hours less than 1,550 hours. This table shows average densities of each choice for 100 draws.
- 24 -
Table 8. Simulated conditional distribution of wife’s labor supply Panel A. Distribution (%) Choices 1 2 3 4 5 6 7 8
Annual Hours Worked
Original Choice 2
0 492 1177 1484 1741 1849 2139 2679
3 0.15 99.73 0.03 0.03 0.02 0.02 0.02 0.00
4 0.27 0.09 99.57 0.02 0.01 0.02 0.02 0.00
0.32 0.10 0.04 99.47 0.01 0.02 0.02 0.00
Panel B. Share of movers (%) Choices
Annual Hours Worked
1 2 3 4 5 6 7 8
0 492 1177 1484 1741 1849 2139 2679
Original Choice 2
3 53.62 --11.90 12.10 7.66 8.42 5.68 0.62
Note: Unit is percent. The case of Reform 2 is presented.
- 25 -
4 62.61 21.08 --3.80 2.89 4.49 4.63 0.49
60.36 19.74 8.45 --2.72 3.97 4.11 0.66
Table 9. Simulated average household income and tax Original 8,648.3 -1,666.3 -8,648.3 --
Reform 1
Reform 2
Before-tax income 8,646.0 8,642.6 -0.03% -0.07% Tax payment 1,821.2 1,738.3 9.30% 4.33% After-tax income 8,646.0 8,642.6 -2.25% -1.11%
Note: Unit is thousand yen.
- 26 -
Reform 3 8,642.6 -0.07% 1,903.0 14.20% 8,642.6 -3.47%
