The role exerted by consumption and excise taxes

According to ^{Bird and Smart (2016}) sales and excise taxes are generally understood as two distinct types of taxes that are common or appraised within the consumption concept. In this research we focus on excise taxes, which are priced in the market for specific goods and services, having regularly negative externalities for different causes, such as health problems, pollution, adulteration, debasement or similar result. Then, its main significance is linked to produce corrections of these externalities, and not only to see them as a potential tax for public revenues.

Although there is separate research done for specific goods accounting for the impact of excise taxes, few literatures estimating the combination of certain sort of groups can be found (as it is intended in this research).

Relevant empirical research on excise taxes

In the international context, excise taxes have been mainly designed for fuels such as gasoline and diesel. For developed countries, those taxes have been regressive (^{Dahl and Sterner 1991}; ^{Chernick and Reschvsky 1997}; ^{Walls and Hanson 1999}; ^{Santos and Catchesides 2005}) and weakly regressive in the case of ^{Poterba (1989}) and Poterba (1991). However, in the case of Spain and by using data from the Family Budget Survey (EPF), ^{Asensio et al. (2003}) estimated a model for gasoline spending, as well as the consumption elasticities and the redistributive effects of a gasoline tax, finding that for low-income groups the tax is progressive and after a certain level of income and for certain localities the tax is regressive, meaning that the application of the consumption tax would tend not to be homogeneous throughout the country. On the other hand, for developing countries gasoline taxes tend to be progressive since most people with lower incomes have a low association to owning a car and their gasoline consumption is much lower, so their income is proportional to the demand for gasoline (^{Antón-Sarabia and Hernández-Trillo, 2014}).

There are other types of products where excise taxes are levied and not less important, such as tobacco, alcohol and soft drinks, as well as other high-calorie food (^{Escario and Molina 2000}; ^{Jimenez et al. 2008}; ^{Waters et al. 2010}; ^{Llamas and Huesca, 2016}; ^{Rodríguez-Iglesias et al. 2017}). ^{Escario and Molina (2000}) analyze tobacco excise taxes in Spain and determine that increasing taxes by 1% will reduce tobacco consumption in less than one percentage point of elasticity, that is, by 0.38% in the short term, which will result in 1.12% reduction in the cancer mortality rate. Meanwhile in the long term, the authors’ estimation of the mortality rate will be reduced in the order of 8.81%. In the case of Argentina, ^{Rodríguez-Iglesias et al. (2017}) is the most recent research up to date on the tobacco issue. The authors reveal how fiscal policy must increase real price on tobacco otherwise, taxation can be neglected; that is, excise taxes could not be effective during times of high levels of inflation. This study found the need to increase tobacco taxes in Argentina in 100 percent, to be able to decrease consumption and maximize public revenues from tobacco.

For Mexico, ^{Jimenez et al. (2008}), estimated tax elasticities for tobacco finding that a 10 percent tax increase would reduce consumption by 6.4 percent along population. According to ^{Waters et al. (2010}), excise taxes on tobacco in Mexico as a percentage of the final price was 62.8% is considerably lower than other Latin-American countries such as Uruguay or Chile, where the equivalent figures were 68% and 76%, respectively. Excise tobacco tax in Mexico would have to increase from 160% up to 350% or at least 17 pesos more (as specific tax) to achieve a tax incidence similar to that of Chile.

For Mexican indirect taxation, ^{Llamas and Huesca (2016}) research follows an analytical approach to decompose both, the incidence and the progressive/regressive effect on household income distribution. Findings reveal that excise taxes are more regressive than VAT, with greater emphasis on high-calorie food. They conclude that families with lower level of income presents a higher tax burden for excise taxes than for VAT.

In general, from the literature revision we conclude that:

Model and determinants of the willingness to work and pay taxes

In order to determine the impact of excise tax payments on consumption, we apply an empirical model based on a simple semi-log quintile specification.

The usual empirical approach analyzes agent behavior through a demand schedule with its corresponding elasticities to estimate the marginal efficiency of any given tax. We consider qth as k subgroups of goods with conditional quantile functions of consumption Cik given a θ vector of x’ exogenous variables as ϱθcikXi´, which implies a simplified equation of the form:

As the error term u _i satisfies ECikX´i=0∀ϱΘ it is analogous to consider the distribution function of ui as Fui to avoid heteroskedastic errors. Then, we write the consumption in terms of the vector of exogenous variables in a quintile semiparametric log linear equation as follows:

Where the log of per capita consumption for the basket of goods is expressed in net (purged from taxes to avoid endogeneity) as C_ik. The log of net per capita monetary income is the log of tax payments to obtain elasticities is represented by log⁡yiθ , while log⁡tikθ is the log of tax payments to obtain elasticities from the excise tax which adds up to five k-subgroups of goods as k= 1: High-calorie food; 2: Alcohol 3: Tobacco; 4: Gasoline & diesel; and 5: Others. The vector of explanatory variables are: Siθ standing for schooling of the household head; a set of dummy variables which includes households with children dchiθ;dsizeiθ taking the value of one for urban households and the error term for each quintile as uiθ. Thus, we likewise replicate this equation to estimate the impact of excise taxes on consumption in both, the north border and the rest of the country.

In order to analyze changes for excise taxes in our model marginal effects (ME) are estimated, as the partial derivative that measures the impact of a change in the tax pressure for each Qθ- quintile on specific k subgroups of goods - from its corresponding excise tax tikθ. In the next equation, we first obtain the effective excise tax as the average for the tax payments on each category as follows:

Taking the partial derivative for any quintile given the effective excise tax rate with respect to the discrete change on the tax pressure on each category for a k- good we obtain:

From the expression (4) we follow the assumption that the marginal change for each quintile in the household i will change by a greater (or lower) magnitude of the average effective excise tax within the quintile distribution. In order to do so, we estimate an ordered logit model where the dependent variable identifies the q-quintiles, and the covariate as a discrete variable that measures the change from a value of cero (no tax pressure) to a value of one (greater tax pressure) when its value from expression (3) is higher than the mean ratio. Remaining in the same quintile of the F distribution upon its corresponding change we formulate the hypothesis stating that for a rational household with ordered preferences on consumption, an increase in income (quantile position) will imply a greater excise tax burden -not necessarily progressive-, and lower quintiles would seem to bare a lower share of this burden. For the purposes of this model, we assume the existence of constant returns to scale between the levels of per capita income along the quintiles (that is, an additional level of income is paid at its corresponding level of per capita income rate) as well as a consumer behavior depending on the subgroups of goods.

Methodology and formulating the excise tax

We use the National Survey of Households’ Income and Expenditures (ENIGH by its initials in Spanish) for the year 2016. Excise duties were calculated according to Mexican tax regulations and five commodities regrouped were made to facilitate estimates comprehension in section V.¹ Households’ incomes, expenditures and taxes are expressed in Mexican per capita pesos at current prices. Expenditures made at the informal market were not taken in consideration; therefore, it was necessary to distinguish between formal and informal expenditures by using the place of purchase provided in the survey.

The attributes of the household described in equation 2, were taken from the survey. Also, we distinguish the states located at the northern border from the rest of the country (that is, Baja California, Chihuahua, Coahuila, Nuevo León, Sonora and Tamaulipas) due to a major proximity to United States that could exert an influence in consumption patterns. Hereafter, to ease further explanations this paper will make reference to the northern border and the rest of the country, as region 1 and region 2, correspondingly.

Hence, the ^{ENIGH 2016} survey indicates a sample of 70 187 representing 33.4 million of households, from which 18.7 percent are located at region 1 and 81.3 percent in region 2 (see table 2). Therefore, quintiles were constructed according to each corresponding weigh by each subsample, using households’ monetary income (disposable income). It can be seen how per capita income for the country as a whole is somehow in the middle level with a yearly amount of $45 462 pesos, meanwhile region 1 shows a higher income level ($54 054 pesos) than region 2 ($42 722 pesos). Also, standard deviation is higher in the former region as a result of greater inequality of earnings amongst its population.

Figure 1 presents the per capita annual excise tax payments for all the five categories of goods. The greater share of revenues in general, comes from the upper two quantiles (4 and 5). Tobacco, gasoline and alcohol are the products more purchased by the population in both regions, and in a lesser extent are those of high calorie intake of food as well as other products. Per capita tax payments are somehow dispersing, as a result of different consumption patterns among the tax-payers in the distribution.

Note: Silverman Gaussian kernel bandwidth for k-categories.

Source: Authors’ estimations using ENIGH, 2016.

Figure 1.  Densities of excise taxes in Mexico by category, 2016 (mxn yearly pesos)  

Econometric results

We run quintile regressions using equation (2) from the previous section modeling the level of consumption in log terms according to the corresponding selected exogenous variables. In the Appendix we include tables A1 and A2 which shows the full set of coefficients from the models estimated for both regions. For a sake of space, figure 2 illustrates the coefficients of excise tax elasticities included in our models, according to each category except that for others. Also, confidence intervals were included for statistical robustness. Narrower confidence intervals are shown for a statistical significance above 95 percent and wider confidence intervals for a statistical significance above 90 percent (see quintile 5 for alcohol and tobacco). It is important to recall that elasticity in our model corresponds to a proportion of the price for the good, acting as a price elasticity of the demand.

Source: Authors’ estimations using ENIGH, 2016.

Figure 2.  Excise tax elasticities in Mexico, 2016.  

In general, all elasticities are partially inelastic (coefficients range -1<log⁡tikθ) which tells how this sort of taxes reduces consumption for each category. According to ^{Peck (2013}) in developing countries, the poor are comprised in quintiles 1 and 2 while the rich are in the quintile 5. By comparing both regions, similar trend were found across quantiles for high calorie food and alcohol with small differences. For both regions, the increasing trend of elasticities shows a lower response of consumption as we move towards upper quintiles when the excise tax rises (except for tobacco). Region 2 depicts less inelastic coefficients along the quintiles for gasoline and diesel.

Now we proceed to a detailed analysis for each category. For high-calorie food, an excise tax increase equivalent to 10 percent, would reduce consumption by 2.7 percent for quintile 1 and 1.06 percent for quintile 5 in region 1, -i.e. quintile 1 decreases it consumption 2.5 times more than quintile 5-; for the rest of the country, reductions are 3.0 and 0.7 percent, respectively (difference between quintiles is 4.2 times greater).

Recall that excise taxes are useful to reduce negative externalities from consumption. The lesser response from quintile 5 implies that consumption on high-calorie food will not be reduced significantly even when the tax burden is higher. These results are useful for alternative studies which examine the association between the socioeconomic status with the prevalence of metabolic syndrome, diabetes or obesity, observing a higher risk of metabolic syndrome for low socioeconomic strata (quintile 1) in Brazil and Mexico, while findings for India goes in the opposite way, where a direct association between high socioeconomic position (quintile 5) and metabolic syndrome has been found (^{Alemán-Mateo} et al. 2018; ^{Marquezine et al. 2008}, respectively).

Also, higher risk of diseases from alcohol and tobacco consumption are associated to low income strata (^{Warner, 2014}; ^{Sesma et al., 2002}); for that reason, greater elasticities from the poor in such products would be a positive finding. In regards to alcohol, estimates indicate a higher response of consumption from the poor in comparison to the rich for both regions. To be more precise, tax elasticities for quintile 1 depicts that consumption would be reduced by 3.0 and 3.4 percent for both, regions 1 and 2 respectively, as a consequence of a 10 percent tax increase; while elasticities for the rich are 1.8 and 1.2, respectively.

Tax elasticities for tobacco shifted in opposite direction for region 1. Indeed, the upper quintile tends to respond more than the poor, indicating that it discourages much more consumption of the richest groups. An excise tax increase equivalent to 10 percent would reduce consumption by 2.8 percent in such quintile. Only in region 2 tax elasticities for quintile 1 are slightly higher than quintile 5 (2.7 versus 1.8).

Since it would be expected to discourage the consumption of lower quintiles, the results allow to shed light on other factors that may affect the low impact on elasticity. Indeed, the policy of increasing the excise tax on tobacco would imply that revenues will increase at the expense of the poorest, hence it can be seen the importance of targeting its revenues for a transfer in order to deal with externalities as a matter of importance within this category. For tobacco, we can conclude that the less responsive to its excise tax along lower quintiles, gives insights for the presence of tobacco as normal goods, at least in region 1.

Furthermore, empirical results for tobacco can be compared with international evidence. Price inelasticity for tobacco ranges reduction in consumption from 0.25 to 0.50 in high-income countries, while in middle and low income countries is around 0.80; in countries such as Argentina, Brazil, Chile and Uruguay it ranges from 0.27 to 0.55 (^{Waters et al., 2010}). For comparative purposes, recall that our results should be interpreted as a proportion of the price elasticity for tobacco. In such case, our estimations remain lower than price elasticities referred in the literature. However, our partial-price inelasticity is similar to those estimated for high-income countries, where quintile 5 displays a coefficient of 0.28 in region 1 and quintile 1 a coefficient of 0.27 in region 2. A special case is found in ^{Jimenez et al. (2008}), where his estimates for elasticities in tobacco are higher. For instance, a 10 percent tax increase would reduce consumption by 6.4 percent in Mexico (in average).

Another category with greater importance on public revenues is gasoline and diesel. Both goods presents an excise tax elasticities with consumption as being more sensitive to an increase on its excise tax than in the rest of the categories. So, also this sort of excise tax exerts a greater effect reducing consumption in the gasoline. Along the quintiles, it can be noticed how the reduction of consumption upon an increase in the excise tax on gasoline and diesel is higher in region 1 than region 2. If the excise tax increase by 10 percent, reductions in consumption for the poor would be by 7.9 and 7.4 percent -at regions 1 and 2, respectively-; while for the rich, consumption would decrease by 3.8 and 3.4 percent, in each area. With significance confidence, tax effect in both regions on the poor are greater than its counterparts, as long as the elasticity intervals crosses each other.

Last but not least, the fifth category of goods refers to telecommunication and gambling services and those presented an estimation of elasticities to close to alcohol and coefficients of consumption propensity slightly higher to the later product (see tables A1 and A2 in Appendix). The trend of elasticities over this category of products presents statistical significance and respond more for the region 2 as a result of the lower economic capacity in comparison with the border region.

Summarizing, as expected excise taxes exert a greater effect on decreasing the consumption of the poor for high-calorie food, alcohol and gasoline and diesel; on the other hand, tobacco has shown a similar pattern regardless of the socioeconomic strata. Between regions, the most important difference is found for gasoline and diesel, averaging one point less in elasticity.

We also proceed to compute a test for equality of coefficients from excise taxes and presents the results in table 5. It can be seen that the null hypothesis is rejected for all categories; however, variations in coefficients across the quintiles, is not statistically significant for tobacco at region 1.

Now, we end up with calculations to obtain marginal effects from equations (3) and (4) and presents the output in figure 3. Recall that ME measures the impact of a unit of change in the tax pressure for each quintile on specific subgroups of goods. A positive (negative) trend implies a greater (lesser) tax pressure across the quintiles. It should be expected that trends on ME would become more positive as we move to a higher quintile because of a greater tax burden and a more progressive taxation scheme.

Figure 3 Marginal effect of excise tax pressure by quintiles in Mexico, 2016.  

Main findings shows that lower quintiles bare a higher tax burden than its counterpart, with the exception for the groups of high calorie food and gasoline. Ranking the products, the observed ME are higher for tobacco and alcohol in detriment of low income groups (quintiles 1 and 2).

Differences along the quintiles for high-calorie food as well as gasoline and diesel are lower by more than 5 points of change. Furthermore, regional analysis shows how lower quintiles in the region 2 faces a greater tax incidence than region 1, but it is relevant to remark how region 2 seems to start having less pressure for high-calorie food where marginal changes are near cero. We confirm the hypothesis that higher quintiles do not necessarily bare a greater tax burden, at least for two products clearly: tobacco and alcohol. In this regard, a rise of government tax revenues derived from an increase in excise duties would be at the expense of putting more pressure on the poor for tobacco and alcohol.

Finally, some fiscal policy recommendations can be derived from this study. First, gasoline and diesel excise taxation exerts a greater effect reducing consumption of these goods. Also, as long as there is a higher density of population in region 2, the slightly lower tax elasticity would imply that a greater tax collection from the rest of the country should be targeted to fight pollution other environmental positive proposals. Second, there is a higher margin to increase excise taxes for high-calorie food for region 2 due to its lower marginal effect from a higher tax pressure. Third, excise tax revenues should be earmarked for budgetary purposes, therefore revenues must be allocated to alleviate negative externalities. In the present those revenues are not earmarked for destination on the use to fight health problems (diabetes, lung cancer, addictions, among others).