The role exerted by income taxation and the international tax-burden

Recent studies on the effect of taxes on Canada’s taxpayers has pondered four central aspects: (1) the income tax, (2) household subsidies and government tax credits, (3) intergenerational transfers and social welfare, and (4) unemployment benefits. Using data from Canada’s consumer finances from 1991, studies by ^{Duclos and Tabi (1996)} and ^{Davidson and Duclos (1997)} estimated an index of progressiveness focusing on tax redistribution. This index evaluates the progressiveness of Canada’s fiscal system as a whole, taking into consideration the effect of taxes and transfers - income being one of the most progressively taxed components in Canada.

^{Perrote, Rodríguez and Salas (2002)} estimated the horizontal inequity of the income tax on Spain’s taxpayers. They found that from 1988 to 1995 the effect of direct taxes and the total redistribute effect of the fiscal system where crucial in the increase of welfare of the citizenry. They illustrate the importance of direct taxation upon redistribution among the different cohorts of taxpayers, as well as the importance of having a more efficient method of tax collection.

Comparative studies among different countries show the effects of income distribution among different types of taxpayers and households (^{OECD, 2005}; ^{Joumard, Pisu and Bloch, 2012}; ^{Paturot, Mellbye and Brys, 2013}; ^{Lustig, Pessino and Scott, 2014}; ^{ECLAC, 2016}; ^{Castañeda, 2016}). Overall, they show that the tax rate along with the system of transfers reduces vertical inequity of income redistribution among most members of the Organization for Economic Cooperation and Development (OECD). These studies claim that a more redistributive tax system leads to higher levels of tax collection. ^{Paturot, Mellbye and Brys (2013)} analyze the progressiveness of payroll taxes for 34 OECD countries and consider the effect of direct taxes upon five intervals of wage income - i.e., between 50 and 200 percent of each country’s mean wage. According to their results tax rates are more progressive among the lower income levels, decreasing along higher levels.

According to ^{Lustig, Pessino and Scott (2014)}, the tax and transfers systems in countries like Argentina, Brazil and Uruguay show a decrease in inequality and poverty. Meanwhile, countries like Mexico, Bolivia and Peru show a decreased redistributive effect as a result of a low collection of income tax with respect to their GDP in spite of their progressive direct taxes. Similarly, ^{ECLAC (2016)} points out that tax revenues from direct taxation are insufficient to have a significant impact on redistribution. Additionally, the personal income tax in Latin-American countries averaged 1.3 percent of the GDP in 2011², less than one fifth of the OECD average (8.5 percent). The effective tax rate of personal income in Latin America is also distinctively low (2.3 percent) with a diminished effect on the reduction of inequality - i.e., -2.1 points of the Gini coefficient. By comparison the average effective tax rate in OECD countries is 13.3 percent, reducing inequality by -11.6 points (^{ECLAC, 2016: 104-105}).

In contrast ^{Castañeda (2016)} claims that for revenues purpose the efficiency of the personal income tax has gone hand-in-hand with Latin America’s opening up to trade as well as a low (personal income) tax burden. Castañeda also provides evidence showing countries in Latin America relied on a greater burden upon indirect taxation - i.e., Value Added Tax - rather than personal income taxes to fund an increase in government spending to afford greater coverage for health and education.

^{Rodriguez (2014)} showed a decrease in the informal sector of Colombia’s economy as a result of a reduction in tax rates - especially in income taxes. He suggests that after 20 years of implementing reduction in rates employment would grow by 2.93 percent. Thus the informal sector would decrease at a rate of 9.02 percent, from 39.1 to 35.57 percent of GDP. Rodriguez proposes an income tax of 24.5 percent, calibrated as a simple average of the effective marginal tax rates in Colombia. This level is similar to that of Mexico circa 2002. Since, there has been a decrease in the effective marginal tax rates in Mexico to 18 percent in recent years (^{Huesca and Araar, 2016}).

According to some studies, Mexico’s tax system as a whole is progressive. Their empirical evidence over three decades (1984-2014) shows that unlike revenue from indirect taxes, revenue from direct taxes bestow progressiveness upon its tax system (^{Vargas, 2010}; ^{Scott, 2014}; ^{Huesca and Araar, 2016}). These studies have also attributed a limited redistributive impact on the tax system due to low tax productivity, multiple exemptions and inefficient collection (^{Scott, 2014}; ^{Huesca and Araar, 2016}).

Relevant research on the labor and income-taxation.

The role of labor and its burden on the fiscal system has been an issue of remarkably relevance. For instance, ^{Keane (2011)} surveys the topic and he focuses on implications for effects of wages and taxes separating the effects by sex. His research study this relationship considering the output of a literature from a three-behavior perspective: static models, lifecycle with savings, as well as life-cycle models with both savings and human capital. When separating the women, variables of fixed costs among demographics like fertility or marriage interacting with human capital.

Considerable controversy can be found in the literature over labor supply and its response to changes in wages and taxes, where at least for males, ^{Keane (2011)} confirms small estimates for labor supply. Elasticities, once corrected for these controls become greater. Then, the previous study explains a faster reaction in the labor market for women, whether to decide work more or to be able to pay a greater share of the income tax-burden than men. Finally, the elasticities for workers become greater for women than for men.

Another important study from ^{Mastrogiacomo et al (2017)} presents how responsive are the jobs in Netherlands with respect to income taxation as well as other benefits. Main findings show differences in labor supply responses between households with and without children are much bigger than those presented in the literature. An efficient tax benefit system should be taken into account for the substantial heterogeneity in behavioral responses to labor market participation. On the contrary, in Germany the study of ^{Hayo and Uhl (2015)} reveals that gender does not have a significant influence on labor market participation when payroll tax change, but it is relevant for self-employed labor supply decisions.

After reviewing several empirical evidence, ^{Meghir and Phillips (2010)} found that there is a consensus on annual labor supply elasticity for women close to 1; however, when weekly hours are considered for estimates, the elasticities tend to be quite small (ranges from 0.0 to 0.3). According to generalized belief, women are more responsive to incentives -which implies that they respond more to tax changes-, the authors concludes that the effect of tax changes on women’s hours worked (particularly married women and lone mothers) is slightly stronger than it is for men.

Also, wage subsidies (such as EITC program) have consequences on labor supply. Empirical literature has documented generally positive impacts such as: growth of net incomes for low-income families and improvement on children wellbeing; also, the decline on labor participation from married women is offset by the rise in participation of single mothers (^{Nichols and Rothstein, 2015}). However, in the United States (US) labor force participation rate in response to the EITC expansions in the US subsidizes make married women to stay at home (^{Eissa and Williamson, 2004}). Despite this, EITC is highly effective as an antipoverty policy, since most of the credit recipients in the United States (US) were either below the poverty line or led to close the poverty gap; hence, EITC is considered as the largest cash or near-cash transfer for low income households in the US (^{Scholz, 1994}). For those reasons, EITC expansions has been of high interest for fiscal policy makers.

For the Mexican case, ^{Alberro (2018)} proposes a negative income tax (NIT) to eradicate chronic poverty and decrease income inequality (which is an EITC-like program). Eligibility criteria for working population aged above 18 years old with total incomes under certain threshold (which considers two parameters: poverty line and income poverty gap). Also, NIT encourages workers to increase their income. According to author’s estimates, the NIT fiscal cost represent 1.25% of GDP.

As a review conclusion in the literature.

The review of the current literature on income taxation and labor participation allows us to have a nicely picture of what is going on with the effort of combining, good or bad jobs linked to the burden exerted through a fiscal system. In general we can conclude that:

3.1. A non-parametric method of analysis

The usual empirical approach analyzes agent behavior through a demand schedule with its corresponding elasticities to estimate the marginal efficiency of any given tax. Instead, we employ non-parametric techniques to allow the data to ‘speak for itself’ and minimize the effect of outliers in the distribution of income upon our observed results (^{Duclos and Araar, 2006, 267}).⁵

Theorem 1. There is a weighted kernel of the form KX-Xth given a differentiable and continuous kernel density function that one may reduce the bias among “neighboring” individuals in the data near the Ti's, inversely proportional to the distance X−X_i such that nh0,5m^(X)-m(X), where nh^−0,2 is the bandwidth between the data and the band h → 0 and nh → ∞ when n→∞, where the conditional variance of the tax burden is given by (See ^{Härdle, 1990}, Proposition 3.1.1).

As follows, in definition 1 we explain why a Gaussian kernel function is preferred in our empirical application.

Definition 1. The definition of the estimator function ƒ^Ti takes into account the properties of continuity and differentiability of the kernel estimator function K(u). For the sake of convenience a kernel function that is symmetric for values near zero is selected, with ∫K(u)du=1, ∫uK(u)du=0 and ∫u2K(u)du=σK2>0 This Gaussian type function meets the inherent properties described in this definition, and has the form:

Proof 1. The “noise” in this Gaussian distribution has the traditional bell-shape and is easily differentiable at any desirable level, with a rate of convergence guaranteed to reach σK2=1. Proof 2. Thus we can differentiate (3) and add h so to rewrite the expression for the Gaussian kernel estimator:

where −0,5⋋_i is a factor of convergence to the real values.

We introduce the non-parametric estimation for the tax burden, T_i, as the predicted response to the income tax in the term m(X_i), such that: ⁶

The function m(X_i) = E[T_i|X_i] overcomes the empirical inability of observing a direct and linear response of the tax rates upon X_i that can explain the functional and deterministic form of the tax burden recorded in T_i. We calculate an estimate using a local non-parametric regression that connects the m points with Theorem 1 and Expression (5), employing ordinary least squares without the need of an a priori specification to imply a functional relation. Thus the tax burden can be estimated using ƒ^Ti as the Gaussian kernel function K_i(X)^0,5T_i such that: ⁷

where α, β are the coefficients for the constant term and the estimated parameter, respectively, X_i represents worker i gross wages and ε_i is a term of randomness. The estimated density function of the tax burden hatƒ(T_i) is represented by the estimator K_i, as described in (6) under the conditions previously established in Proof 2. From the equation in (6) we get the estimators for T = T_i (X)+ε_i as well as T = T_ij (H)+ε_i, with their respective predictions given by:

We use the first partial derivatives to estimate the marginal income tax rates with respect to gross wage levels, a non-parametric conditioning of the form:

Similarly we use first partial derivatives to estimate marginal income tax rates with respect to hours worked. This non-parametric conditioning measure intends to capture the marginal utility of an individual’s wage with respect to tax payments, using hours worked to register this procedure for the worker’s sex:

These conditions let us identify which groups have been somewhat affected in a progressive manner with respect to their tax burdens and hours worked.

3.2. A non-parametric estimate of direct taxes

Using the equation from Expression (3) we identify the direct tax T(X) burden as variable influenced by attributes other than income. Thus we formulate a non-parametric estimate of the tax burden as a stochastic function given a worker’s wages X such that E[T|X + ε_i]dX, where the additional variable behaves as a stochastic determinant of taxes:

After running this non-parametric regression of T(X) with respect to X, we run a similar regression with respect to hours worked (H) such that:

For the sake of robustness of the estimate of the non-parametric relationship between increases in T and X subject to a progressive tax burden structure, we establish the following three axioms:

Axiom 1: Assumptions about preferences in T: The representative worker is rational and orders his or her preferences according to his or her wage level (X). For an individual i that chooses more H such that he or she gets a higher tax burden T, the conditions sufficiently near to H will be reflected upon T, if and only if:

where the necessary condition in equation (13b) stands for regressivity in T as long as partial derivative of ∂(X_i/H_i) decreases and H is higher, when XiHi≡Xi∆1Hi⟹∆0Xi then, T(X_i)≤1 should be negative. In our framework, the necessary condition will apply only for level of incomes in the ranges X_i ∈ [0, z⁺], where z⁺ represents a negative tax (tax credit in our case).

Axiom 2: The representative worker shows monotonicity in his or her decisions: to wit, although a lower tax burden (T) is preferred, upon an increase on income taxes T(X)>0 an increase in working hours will yield a greater tax burden δTijδHij≥1.

Proof. Considering a utility of individual i of gender j as a function of wages and hours worked U_ij (X,H), with a total derivative such that:

Solving for dH/dX gives us a positive slope that will designate an increase in hours worked (H) in order to increase wages (X), yielding a greater tax burden given the condition that dH/dX ≥ 1 such that δTijδHij≥1.

Axiom 3: Consumption goods are normal goods with respect to the tax burden: Consumption for an individual i of gender j, C_ij, increases with an increase in wages, X_ij, and decreases with an increase in tax burdens, T_ij, if and only if the following derivatives hold:

3.3. Methodology and formulating the income tax

We formulate the taxpayers’ payments using Mexico’s National Statistics Institute’s (INEGI, by its initials in Spanish) Household Income and Expenditure Survey (ENIGH, by its initials in Spanish). According to its construction, the information is associated with three units of analysis (dwelling, households and household members); in this research we consider individuals whose source of income are wages. This survey offers a detailed description of the sources of income and characteristics of the participants and has been widely used to analyze Mexico’s fiscal system (^{SHCP, 2016}; ^{Huesca and Araar, 2016}; ^{Lustig et al., 2014}; ^{Vargas, 2010}). Hence, estimation results can be generalized to the population since the survey is probabilistic as well a representative of the country and its 32 states.

We use the ENIGH database for 2016 and purge the data based on sex, hours worked weekly and age.⁸ We limit ourselves to consider only wage earners with their corresponding salaries from their main occupations, expressed in Mexican pesos at current prices - i.e., variable N in Equation (1). Further, we also only consider those individuals that have quoted a social security scheme, guaranteeing their formality and thus corroborating the corresponding income tax payments have been made.⁹ Table A.1 in the annex shows the sources of appointment used in the survey.

We apply the corresponding fiscal rules in the survey according to the reconstruction of the fiscal system described in Equation (1). The unit of analysis is the individual wage earner, not the household. Thus the ENIGH 2016 survey indicates that a total of 44,834 individuals represent 22.1 million wage earners. This data represents 69 percent of the real census of taxpayers enrolled in SHCP, which reaches a grand total of 32.2 million as of august of 2016 (^{SAT, 2017}).¹⁰

The fiscal system features eleven distinct income tax rates for 2016. An annual fee must be paid at each rate according to a corresponding fixed quota denominated in Mexican pesos to the order of [0.00, 114.24, 2,966.76, 7,130.88, 9,438.60, 13,087.44, 39,929.04, 73,703.40, 180,850.80, 260,850.84, 940,850.76]. The marginal income tax rates ranged from 1.92 percent to 35 percent (see Figure 1). ¹¹ Also, according to the income tax law in Mexico, workers with annual wages lower than 88,587.96 per year are entitled to receive a tax credit which is compensated to PIT liability (which is alike to an earned income tax credit programs -EITC- in the United States or other OECD countries).¹² Hereafter, this paper will make reference to Mexican tax credits as EITC. Therefore, the annual amounts of EITC were [4,884.24, 4,881.96, 4,879.44, 4,713.24, 4,589.52, 4,250.76, 3,898.44, 3,535.56, 3,042.48, 2,611.32].¹³

Source: Author’s elaboration using data from ENIGH survey for 2016

Figure 1 Marginal income tax brackets in Mexico, 2016