Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Journal of Business Convergence, Vol.7, No.1(February, 2022), 27~32, ISSN 2765-401X http://dx.doi.org/10.31152/JB.2022.02.7.1.

27

Estimation of Latent Preference for Desirable Personal Income Tax Rate*


1)

Kim, ByungWoo**

This study is interested in the degree of satisfaction of taxpayers for income tax rate, burden and hope for improvement. In special, it focuses on personal labor
income tax. This study develops the model of ordered choice model that finds the determinants of household members for choosing the degree of desirable tax rate:
10%, 20%, and 30%, etc. This model estimates the marginal effects of explanatory variables on unobserved sentiment for optimal tax rate and tax burden of taxpayers.
This study performed the estimation of several types of dependent variable models. Estimation of ordered probit results shows, the choice for desirable tax is not
significant in the case of current annual income as explanatory variable. Ordered model is poreferable to ordinary LS(least squares), since dependent variable measure
is not correct degree of sentiment. This study may help government that have been considering specific implementation of personal income tax such as single or
couple income tax.

Key words: limited dependent variable LS, discrete regression, maximum likelihood (ML), ordered probit

Received: 2021. 11. 22. Revised: 2022. 2. 2. Accepted: 2022. 2. 5.


* ByungWoo Kim gratefully acknowledges the financial support from National Research Foundation of Korea (NRF), grant no. 2018 S1A5A2A01028864. I am responsible for any error.
** Professor, School of Liberal Arts, Korea National University of Transportation, E-mail: bwkim2@daum.net

- 27 -
Journal of Business Convergence, Vol.7, No.1(February, 2022), 27~32

I. Introduction <Table 1> Questionnaire for Desirable Personal Income Tax Rate, and Codding (Source:
NaSTab 2021)

In Korea, the tax rate structure in personal (laborer) income taxation is in <Figure
1>. This study is interested in the desirable rate structure from the viewpoint of
taxpayers. In general bracket system is used in (personal) income tax. This study
develops the model that finds the determinants of household members for choosing
the degree of desirable tax rate: 10%, 20%, and 30%, etc. This model estimates the
marginal effects on unobserved sentiment for optimal tax rate and tax burden of
taxpayers.. In addition, it shows the structure of American household income tax in
2020. In the US, the Tax Reform Act of 1986 was a big change in tax structure.
Samuelson (1986) argued that the income tax law was so complicated in the US.
This issue also have been related to Korean income taxation policy. Makoto, Ida,
Question b1:
Murakami, and Friedman (2013) estimated mixed logit regression model(conjoint What do you think desirable income tax bracket?
1 0%
analysis) for willingness to pay(WTP) of consumers. Due to shortcomings of 2 1-10%.
.......
survey(stated) data, researchers use revealed preference data. In this study, I correct 8 60%-
the bias from analyzing survey data such as NaSTab.
In Hausman and Ruud (1986), the authors analyze a richer specification of the logit
<Table 2> Summary Statistics for Total Income and Resposes from Major Three Categories
model when respondents provide their rankings of the alternatives.1) Alm and of Income (Source: NaSTaB 2021)
Whittongton (2003) estimate the impact of federal individual income tax, on a couple’s AINCOME ARENTINCOME EVA3000 EVA5000 EVA7000
Mean 3836.349 1563.235 1.701068 2.056940 2.459075
decision to marry instead of cohabit. Using household data from the Panel Study on Median 2991.000 720.0000 2
Maximum 20000.00 55000.00 5.000000 6.000000 5.000000
Income Dynamics (PSID), they estimate various models of cohabitation as a function of Minimum -9.000000 -9.000000 1
Std. Dev. 3292.486 4494.943 .634963 .689433 .823276
the change in tax burden. For Sample1, they estimate the factors that determine the Observations 281 281 281 281 281
likelihood that a couple in their first year of living together chooses to do so as
cohabiting or married and derived a reduced-form model of the probability that a couple Note:
AINCOME: Annual Total Labor Income
chooses to marry rather than cohabit.2) By this study, government can get hint for ARENTALINCOME: Annual Rental Income
EVA3000: Response with annual income of 30MN(Won)
implementation of income taxation with regard to single or married income tax payers. EVA5000: Response with annual income of 50MN(Won)
EVA7000: Response with annual income of 70MN(Won)
<Figure 1> shows the income tax structure and depicts the distribution of tax
rates among many income brackets in Korea. In Korea, many kinds of (income)
deduction is implemented. These deductions are profitable to taxpayer with high
marginal tax rate in general. In addition, tax credit is mainly helpful to those with
low marginal rates. This study examines the determinants of desirable tax choice of
Korean taxpayers surveyed in NaSTab data. That is, this study estimates the marginal
effects of household (taxpayer) characteristics on the unobserved sentiment for tax
system such as burden of household. I use limited dependent variable model, since
ordinary LS cannot capture the exact marginal effect on psychology. <Tab 1> shows
partial example in questionnaire. It questions many issues with regard to feeling or
<Figure 1> Distribution of Personal Income Tax Rates (Korea and the US, Mil. %, $)
sentiment for tax rate. I focus mainly on labor income tax rate. And, This study
search for significant determinant for sentiment of workers.
This study examines the determinants of sentiments for desirable tax rate of Korean
<Table 2> shows the summary statistics for characteristics: labor and rent
taxpayers (consumers). Similar research exists. Johnston (2000) study the role of taxpayer
income, etc. This study uses 14,757 cross section household-member data of NaSTab
understanding in marketing decision.3) In <Table 2>, taxpayers with annual income of
2021. The survey data has characteristics of nonexperimental or observational data,
30MN (Won) responded 1.70, on average(1: 0%, 2: 1-10%). The analysis showed that
so, the (response) value may be random. This means that sample selection bias
average annual total and rental income of household may influence the sentiment that is
could cause problem for estimates such as inconsistency. This study corrects this bias
unobservable. Average annual income of household may have a negative influence on
by using ordered model that also estimates each threshold.
desirable tax rate, and rental income may also have a positive relationship.

1) Marcus and Greene (1985) estimated an ordered probit model for the job assignments. The Navy attempts to direct recruits into job classifications that is most productive. The broad
classifications were technical jobs with 3 ranked skill ratings: “medium skilled,”“highly skilled,” and “highly skilled.”
2) The dependent variable (Mi) in their estimation is binary, and equals 1 if couple i chooses to live together as a legally married union and 0 if the couple chooses to be unmarried.
3) They estimated mixture model parameters that show the effects of attributes on consumer choices.

- 28 -
Estimation of Latent Preference for Desirable Personal Income Tax Rate

This study first use 14757 cross section data of KIS 2021 of Korea. Later, I could 1. Desirable Tax Rate
use panel data of 2018 and 2021, but using the sophisticated estimation technique
Basic econometric models for choice of number in questionaire have a base on
is rather difficult due to undevelopment of statistical methods for simulation-based
the ordered probit model. This issue based on the idea that “sentiment” has led to
estimation. <Figure 2> shows the graph for taxpayers’ responses for income.4)
specific choice among alternatives. In general, applying OLS to binary choice data is
not a good idea since though both OLS and Feasible GLS yield consistent estimators,
160,000
the latter produces more efficient estimator.
120,000 For estimating the choice behavior for alternatives, I need to understand the fact
80,000 that the probability of choosing a specific brand of commodity can be affected by
consumer’s characteristics. I express consumers' choices for specific tax bracket by
40,000
the ordered dependent variable:
0

-40,000
2500 5000 7500 10000 12500  =0: taxpayer(household) chooses 0% of rate (2.1)
p13bb002 p13bb008 =1: chooses 1-10% of rate
<Figure 2> Two Types of Annual Income =2:: chooses 11-20% of rate
Note: x-axis denotes the ID of responsents.
=3: chooses 21-30% of rate
=4: chooses 31-40% of rate
AINCOME: Annual Total Labor Income (p13bb002)
ARENTALINCOME: Annual Rental Income (p13bb0008) =5: chooses 41-50% of rate
=6: chooses 51-60% of rate
This paper is organized as follows: Section 2 presents standard econometric =7: chooses 60%- of rate
model. Then, I provide the basis for estimating the parameters and a short
interpretation of the response behavior in Korean taxpayers. Section 4 concludes. This study can set the probability of choosing category as ordered probit
estimation

II. Econometric Models with Limited (Ordered) Prob(Y=1)=Φ [  - (β'x(i)]

Dependnet Variable

<Table 3> Ordered Probit for Desirable Income Tax Rates (Source: NaSTab 2021)5)
Dependent Variable: EVA3000 Dependent Variable: EVA5000 Dependent Variable: EVA7000
Method: ML - Ordered Probit (Newton-Raphson / Marquardt steps) Coefficient covariance computed using observed Hessian Method: ML - Ordered Probit (Newton-Raphson / Marquardt steps)
Sample (adjusted): 1 14757
Var. Coeff. Prob. Var. Coeff. Prob. Var. Coeff. Prob.
AINCOME 1.62E-06 .6361 AINCOME 2.26E-06 .4994 AINCOME 3.42E-06 .2847
Limit Points Limit Points Limit Points
LIMIT_2:C(2) - .237 .000** LIMIT_2:C(2) - .973 .000** LIMIT_2:C(2) -1.471 .000**
LIMIT_3:C(3) 1.563 .000** LIMIT_3:C(3) .837 .000** LIMIT_3:C(3) .140 .000**
LIMIT_4:C(4) 2.483 .000** LIMIT_4:C(4) 2.103 .000** LIMIT_4:C(4) 1.338 .000**
LIMIT_5:C(5) 3.202 .000** LIMIT_5:C(5) 2.855 .000** LIMIT_5:C(5) 2.371 .000**
LIMIT_7:C(6) 3.687 .000** LIMIT_6:C(6) 3.508 .000** LIMIT_6:C(6) 3.124 .000**
LIMIT_7:C(7) 3.689 .000** LIMIT_7:C(7) 3.693 .000**
LR statistic .2239 - .8916 LR statistic .455 - .9785 LR statistic 1.143 -1.1543
Prob: LR statisti) .6360 Avg. log likelihood .4997 .284

There are one kind of approach that treat binary choice model, latent regression The marginal effect in probit model is:
(index function model). If we use logistic or normal distribution, then these are logit
or probit models. There is close relation between two estimates(Amemiya 1961). dp/d   =φ( -β'  )β, m=AINCOME
<Table 3> shows cross- section data estimation results. I used Ordered Probit
estimation under the normal distribution assumption. The size of annual income In this case, marginal effect is not the coefficient, but the multiplication with pdf
increases the probability of choosing higher rates, but is not significant. of standard normal. Estimate for coefficient is –1.62E-06. This means that 10,000

4) I have to notice that the survey data used in this study have characteristics of non-experimental.
5) If estimated coefficient is significant, I denote *, ** at the 5%, 10%, respectively.

- 29 -
Journal of Business Convergence, Vol.7, No.1(February, 2022), 27~32

Won increase in the (household) income cause the probability of choice to increase <Table 5> Heckit Model for Desirable Tax (Income 70MN)
by –1.62E-06 φ (- .2377 + 1.62E-06 
  ). When calculating marginal effect, I used Dependent Variable: EVA7000, Method: ML Heckman Selection (Newton-Raphson/Marquardt steps)
Variable Coefficient Std. Error t-Statistic Prob.
level of income rather than log value instead. Response Equation - EVA7000
C 2.818686 1.336930 2.108328 .0350**
ARENTINCOME -2.06E-05 1.09E-05 -1.893011 .0584**
<Table 4> Ordered Probit for Desirable Tax Rates with Rental Income Selection Equation - ARENT01
Dependent Variable: EVA7000 Dependent Variable: ETCINCOME C -1.898321 .031770 .0000**
AINCOME 1.13E-05 4.99E-06 .0234**
Method: ML - Ordered Probit Method: Least Squares Sum squared resid 217.4482 Schwarz criterion -59.75286 .362495
(Newton-Raphson / Marquardt steps) Log likelihood -1582.202 Hannan-Quinn criter 2.266489 .359335
Variable Coefficient Prob. Variable Coefficient Prob.
ARENTINCOME -2.27E-05 .0819** C 401.210 .195
Limit Points EVA3000 299.9773 .094* 3. Tax Burden for Income Tax
LIMIT_2:C(2) -1.557 .0000**
LIMIT_3:C(3) .166 .0011**
LIMIT_4:C(4) 1.318 .0000**
This study estimated correlation between sentiment of tax burden and desirable
LIMIT_5:C(5) 2.359 .0000**
LIMIT_6:C(6) 2.959 .0000** tax bracket. There is weak negative correlation. Then, I estimated ordered probit
LR statistic 3.114 -1.128 F-statistic 2.833 3.488
Prob(LR statistic) .077596**   Prob(F-statistic) .094670* using financial income such as interest rate and dividend income(INTESTDID). The
coefficent is not significant but, thresholds are significantly positive. I express
taxpayers' choices for specific sentiment by the ordered dependent variable:
10,000
Series: EVA5000
Sample 1 14759
8,000 Observations 14204
 =5: taxpayer(household) chooses very high (2.4)
6,000 Mean 2.052802
Median 2.000000
=4: chooses a little high
Maximum 9.000000
4,000
Minimum 1.000000 =3: common
Std. Dev. 0.658589
2,000 Skewness 0.737995 =2: a little low
Kurtosis 6.478159
=1: very low
0
1 2 3 4 5 6 7 8 9 Jarque-Bera 8449.094
Probability 0.000000 �

<Figure 3> Empirical Histogram for Desirable Tax Rate with 50MN Income (Household Members) <Table 6> Sample Correlation and Effects of Financial Income (Income 30MN)
Note: I used category variable for response of desirable tax rate. BURDEN2 EVA3000
BURDEN2 1 - .0040
EVA3000 1
McFadden(1974) R-squared means likelihood ratio index:
Dependent Variable: EVA3000, Method: ML - Ordered Probit (Newton-Raphson/ Marquardt steps)
Variable Coefficient Prob.
LRI=1–(ln L/ln L0) (2-2) LOG(INTERESTDID) .043662 .5780
Limit Points
LIMIT_2:C(2) - .551323 .1815
LIMIT_3:C(3) 1.611615 .0001**
ln L0: log-likelihood with only a constant term. LIMIT_4:C(4) 2.657685 .0000**
LR statistic .309686 Avg. log likelihood - .814721
Prob(LR statistic) .577873
This value shows that using Ordered Probit model for tax rate with rental income
as explanatory variable has better goodness of fit (LR statistic=3.11). P-value for the <Table 6> shows that there is negative correlation between sentiment for tax
LR statistic shows that the overall significance for the regression approach of burden and desirable tax rate. It coincides with intuition. In questionaire, the
ordered probit model(<Table 4>). question for positioning between satisfaction and desire for tax burden(<Figure 4>).

2. Sample Selection
Welafare(W)

This study estimated selection equation by (maximum likelihood) Heckman


selection method. This model is a kind of incidentally truncated model(Greene 2008). H LT, HW MT, HW HT, HW

Prob(ARENT01=1)=Φ(β'x)=Φ[β0+β1 ln(AINCOME)] (2-3) M LT, MW MT, MW HT, MW


ARENT01=1: taxpayer has rental income.

L LT, LW MT, LW HT, LW


<Table 5> shows that the estimated coefficients change when I consider the
taxpayer has any rental income. If I ignore the fact that the respondent have a rental L M H Burden of Tax(T)

income, the coefficient can be underestimated or overestimated. Therefore, the <Figure 4> Tax Burden vs. Welfare Level (Current)
hypothesis of a kind of “welath effect” on attitude is not supported in NasTab Data.

- 30 -
Estimation of Latent Preference for Desirable Personal Income Tax Rate

<Table 7> Sample Correlation and Effects of Financial Income (Income 30MN) 4 Discussion
Coding Level of Burden→
9 8 7 Knoll (2003) analyzed the tax efficiency of stock-based worker compensation. This
Level of Welfare 6 5 4
↑ 3 2 1 <Table 9> shows the difference from system of (labor) income tax. This explains the
difference income taz rate applied each bracket in that different tax rates(   : grant of

Coded answers are collected and displayed as empirical histogram(<Figure 5>). stock,   : exercising of stock,   : sale of stock) are applied to employees.

6,000 <Table 9> Tax Treatment of Stock-based Compensation to Employees and Employers per
Series: CURRENT2
Sample 1 14759
Share of Stock Covered (Knoll 2003)
5,000
Observations 14204
Date Grant(g) Exercise(e) Sold(s)
4,000
Mean 4.894678 Stock Price P(g) P(e) P(s)
Median 5.000000
3,000
Maximum 19.00000
Nonqualified Options
2,000
Minimum 1.000000
Employee 0 -  X[(P(e)-P(g)] -g X[(P(s)-P(e)]
Std. Dev. 1.968892
Skewness 0.113462
Employer 0  X[(P(e)-P(g)] 0
1,000
Kurtosis 2.996141
Incentive Stock Options
0
2 4 6 8 10 12 14 16 18 Jarque-Bera 30.48472
Employee 0 0 -g X[(P(s)-P(g)]
Probability 0.000000�
Employer 0 0 0
6,000
Series: FUTURE2
5,000 Sample 1 14759
Observations 14204 Future study can extend our analysis to estimation of latent tax bracket in
4,000
Mean 6.257251 option-based income. That is, I can estimate different unobserved reservation tax
Median 6.000000
3,000
Maximum 19.00000
Minimum 1.000000
rates of taxpayers.
2,000
Std. Dev. 1.694427
Skewness -0.066033
1,000
Kurtosis 2.553087

0
2 4 6 8 10 12 14 16 18 Jarque-Bera 128.5303
Probability 0.000000 �
III. Summary and Limitations
<Figure 5> Tax Burden vs. Welfare Level (Current vs. Future)

This study developed and estimated the model of ordered choices that find the
<Table 8> shows that there is negative effect of income on sentiment for current characteristics of households affecting the choice for desirable tax rates. I also examine
tax burden and desirable degree(rating) of future tax rate. It also coincides with how the the level of factors such as rental income level affect the choice probabilities.
intuition. Members in high income household felt dissatisfaction for current burden. This study’s empirical research have many interesting aspects which can be
extended to the future research.6) This can give high incentive for policy making for
<Table 8> Effects of Income on Degree of Satisfaction of Tax Burden taxpaers and give implication tax policy. First of all, future study can use full panel
Dependent Variable: CURRENT, Method: ML - Ordered Probit (Newton-Raphson/Marquardt steps) data of this survey.
Variable Coefficient Prob.
AINCOME -7.11E-06 .0226** I can also estimate the marginal (partial) effect of a change, for example, in rental
Limit Points income change on the probability that an individual chooses a specific bracket for
LIMIT_2:C(2) -1.806889 .0000**
LIMIT_3:C(3) - .948442 .0000** income tax.
LIMIT_4:C(4) - .744884 .0000**
LIMIT_5:C(5) - .464211 .0000** Finally, Bordoff and Kaufman (2018) proposed carbon tax rates as $50/ton in
LIMIT_6:C(6) .604924 .0000**
LIMIT_7:C(7) .941359 .0000** 2020. I need to provide empirical sentiment(willingness) of taxpayers by limited
LIMIT_8:C(8) 1.058611 .0000**
LIMIT_9:C(9) 1.809421 .0000** dependent variable econometric model.
LR statistic 5.213927 Avg. log likelihood -1.829759
Prob(LR statistic) .022407**

Dependent Variable: FUTURE, Method: ML - Ordered Probit


Variable Coefficient Prob.
AINCOME -7.36E-06 .0198** Reference
CURRENT2 .061805 .0000**
Limit Points
LIMIT_2:C(3) -2.577068 .0000**
Korea Institute of Public Finance (2021), 13th Questionnaire for Members in a
LIMIT_3:C(4) -1.774042 .0000** Household(Korean).
LIMIT_4:C(5) -1.459626 .0000**
LIMIT_5:C(6) -1.270091 .0000** Allenby, G. and P. Rossi (1999), “Marketing Models of Consumer Heterogeneity,”
LIMIT_6:C(7) .193434 .0000**
LIMIT_7:C(8) .466529 .0000** Journal of Econometrica, 89, 57-78.
LIMIT_8:C(9) .630126 .0000**
LIMIT_9:C(10) 1.697823 .0000** Alm, J. and L. Whittongton (2003), “Shocking up or Shelling out: Income Taxes,
LR statistic 118.9545 Avg. log likelihood -1.579683
Prob(LR statistic) .000000** Marriage, and Cohabitation,” Review of Economics of the Household, 1-24.
Note: Two tables are different in explanatory variables. 2. Upper table has CURRENT [sentiment Ann, J., G. Jeong and Y. Kim (2008), “A Forecast of Household Ownership and Use of
(opinion) and FUTURE (desirable burden) for tax burden] as a dependent variable Fuel Vehicles,” Energy Economics, 30, 2019-2104.

6) I can also use Multinomial Probit Model (MNP). Recently, simulation based methods are popular. You can refer to Manski (1975a, 1985b).

- 31 -
Journal of Business Convergence, Vol.7, No.1(February, 2022), 27~32

Baek, T. H., J, Kim and J. H. Yu (2009), “The Differential Roles of Brand Credibility and Pennsylvania Carey Law School.
Brand Prestige in Consumer Brand Choice,” Psychology and Marketing, McFadden, D. (1974), "Conditional Logit Analysis of Qualitative Choice Behavior,” in
27(7), 662-678. Frontiers in Econometrics. Ny: Academic Press.
Berndt, E., B. H. Hall and R. E. Hall (1974), “Estimation and Inference in Nonlinear Makoto, T, M. Ida, K. Murakami and L. Friedman (2013), “Consumers' Willingness to
Structural Models,” Annals of Economic and Social Measurement, 3(4), 653-665. Pay for Alternative Fuel Vehicles: Comparative Analysis between US and
Berry, S., J. Levinsohn and A. Pakes (1995), “Automobile Prices in Market Equilibrium,” Japan,” working paper series GSPP13-001, University of California Berkely.
Econometrica, 63(4), 841-890. Manski, C. (1975a), "The Maximum Score Estimator of the Stochastic Utility Model of
Bordoff, J. and N. Kaufman (2018), A Federal US Carbon Tax: Major Design Decisions Choice,” Journal of Econometrics, 3, 205-228.
and Implications. Cell Press. Manski, C. (1985b), "Semiparametric Analysis of Discrete Response: Asymptotic Properties
Greene, W. (2015), Econometric Analysis. 7th., NJ: Prentice-Hall. of the Maximum Score Estimator,” Journal of Econometrics, 27, 313-333.
Greene, W. and D. Hensher (2010), "Does Scale Heterogeneity across Individuals Manski, C. (1986c), "Operational Characteristics of the Maximum Score Estimator,” Journal
Matter? An Empirical Assessment of Different Logit Models,” Transportation, of Econometrics, 32, 85-100.
37(3), 413-428. Xia, W., J. Choi, K. Wang and A. Zhang (2019), "Estimating Airport Ground Transport
Hansen, B. (2001), “The New Econometrics of Structural Change: Dating Breaks in Choice Using Market Aggregate Data,” WEAI, 1-37.
U.S. Labor Productivity,” Journal of Economic Perspectives, 15, 117–128.
Hausman, P. and A. Ruud (1986), “Specifying and Testing Econometric Models for
Rank-ordered Data,“ Journal of Econometrics, 34, 83-104. Appendix
Hensher, D. and W. Greene (2002), "The Mixed Logit Model : The State of Practice,”
working paper, Institute of Transport Studies. <Table 1> Code Table
Honore, B. and E. Kyriazidou (2000) "Estimation of Tobit-Type Models with Individual
Specific Effects," Econometric Reviews, 19(3), 341-367.
Kim, B. (2015), "Knowledge Protection Selection: Multinomial Logit Model,” Journal of
Product Research, 33(6), 55-62.
Kim, B. (2020), "Vehicle Brand Selection: Multinomial Logit Model,” Journal of
Product Research, 38(3), 129-139.
Kim, Y., G. Jeong, J. Ann and J. Lee (2007), "Consumer Preferences for Alternative Fuel
Vehicles in South Korea,” International Journal of Automotive Technology
and Management, 7(4), 327-342.
Koo, Y., C. Kim, J. Hong, I. Choi and J. Lee (2012), "Consumer Preferences for
Auromobile Energy-Efficiency Grades,” Energy Economics, 34, 446-451.
Kroll, S. (2003), “The Tax Efficiency of St ax Efficiency of Stock-Based Compensation
ock-Based Compensation,“ in Faculty Schologhip at Penn Law, University of

- 32 -

You might also like