GOODS AND SERVICES TAX DYNAMICS
Patricia Ferreira Motta Café!, Luciano Neves Fonseca”, Ricardo Chaim?
pmottacafe@ gmail.com; luciano.unb@ gmail.com; rmchaim.unb@ gmail.com
‘Finance Secretariat (DF), Major in Civil Engineering (UnB), MS in Civil Engineering
(UC Davis) and Economics (UnB)
"University of Brasilia (UnB-FGA), Major in Electrical Engineering (UnB), MS in
Electrical Engineering (UNICAMP), PhD in Oceanic Engineering (UNH)
°University of Brasilia (UnB-FGA), Major in Management and Computer Science, MS
and PhD in Information Science
ABSTRACT
This paper aims to model the dynamics and the characteristics of the tax over the
circulation of goods and services (ICMS) and thus estimate the potential ICMS tax
capacity for Brazilian states from a set of socioeconomic variables. As ICMS in Brazil
is the main source of public resources of tax origin for Brazilian states, it also defines
the maximum capacity of tax collecting by tax authority, given the economic and social
characteristics of each state.
First, the ICMS behavior was analyzed and econometric models based on multiple
linear regressions using the Ordinary Least Squares Method were built. Statistical
criteria were used in the selection of the most appropriate estimation model to estimate
the potential ICMS revenue of all Brazilian states. The principle of parsimony was also
taken into account to select the simplest model, which still complies with the chosen
criteria. The Tax Effort Index for each state was calculated from the ratio between the
effective and the potential ICMS revenue, which reveals a valuable tool for revenue
performance analysis on this kind of policy making processes. Finally, this study also
produced a SD model of the Brazilian good and services tax dynamics to enhance the
econometric capability to explain the tax behavior and its interaction with
socioeconomic factors.
Keywords: Goods and services tax (ICMS), potential revenue, explanatory variables,
method of ordinary least squares, multiple linear regressions, fiscal effort index, tax
dynamics.
1) INTRODUCTION
Taxes are the main provision source to finance governments in order to offer
public services to society. In Brazil, the main taxes are levied on classic basis such as
income, consumption and property. Since Brazil has a federalist system, taxes are
collected at the three levels of government: federal, state and municipal.
In general, the Federal Government is in charge of social contributions and
income tax, the States for tax on goods, and the Municipalities for tax on services and
tax on urban property. Based on this distribution of competencies, in 2014, federal
government was responsible for 68.5% of total revenue, while states and municipalities
for 25.3% and 6.2%, respectively. The Federal District is the only federated unit that
collects both state and local taxes.
In Brazilian tax system, taxes on goods and services account for 51.0% of total
tax revenue. Among these taxes, the ICMS state tax is the one that most raises revenues,
corresponding to 20.9% of total and 82.8% of state revenues. Therefore, ICMS is the
main source of public resource of tax origin for Brazilian States and it will be the focus
of this work.
According to Prado (2009), despite the central role of ICMS in the Brazilian
taxation system, States are still in a fragile situation. Most of them are in debt, working
with limited budget. Additionally, horizontal cooperation among States is precarious,
and the "fiscal war" is an illustration of this fact. Indeed, States were the main losers in
terms of federal transfers after the Constitution of 1988, due to the expansion of social
contributions to finance municipal programs, not shared with States. Over time, state
governments concentrated about 40% of its ICMS revenues on the known blue chips -
electricity, telecommunications and fuel, with little room for growth nowadays.
Therefore, state tax administrations has no other option then to collect ICMS as
efficiently as possible, considering that this tax is the main support of their budgets.
Every month, state technicians evaluate tax collection by gathering modality and
economic activity, establishing monthly and annual comparisons. However, the
assessment of state capacity of collecting ICMS should go beyond the simple analysis
of historical collection series, because theses series obviously do not include
uncollected components due to the effect of tax expenditures, administrative and
judicial litigations, elision and/or evasion. These uncollected taxes constitute the so-
called tax gap, which is an object of many tax administration studies. Consequently,
knowing the socioeconomic factors that affect ICMS revenue, estimating its maximum
tax capacity and how much the effective collection represents in relation to this
potential constitute an important management tool for tax administration. This ratio
between actual and potential ICMS revenue will be here called Tax Effort Index (TEI).
This study focuses on exploring a set of socioeconomic variables of the 26
Brazilian States and the Federal District, called States from now on, which might
explain their ICMS potential revenue, from the structural point of view. Since Brazil is
a federation, a good understanding of the ICMS system as a whole and the diversity
among its members is vital to develop better policies and orientations.
First, an econometric study will be undertaken by regressions of the ICMS
variable on explanatory variables of social and economic nature for the set of States.
Thus, potential revenue models using an econometric tool with cross section data with
all States (as opposed to a simple temporal analysis) will be constituted for the most
recent year in which the variables collected are available, that is 2012. Then, potential
revenue will be compared with effective revenue of ICMS to calculate the fiscal effort
index of each State.
Knowing that econometric models don’t capture the feedback relations between
factors, a combination of the econometric study with a system dynamic model will
enhance the econometric capability to explain the ICMS behavior and its interaction
with the socioeconomic variables studied previously. Therefore, this paper will also
explore a system dynamic model based on the relations identified by the econometric
study.
Regarding this paper organization, Item 2 will present a discussion, including a
literature review of other studies that deal with the same theme. Item 3 will discuss the
methodology to be used in the SD and statistical models. Item 4 will analyze the
collected variables that can be used as explanatory variables in the SD and estimation
models of potential ICMS tax capacity as well as the behavior analysis of explanatory
variables in relation to ICMS. Item 5 will test tax capacity models using the
econometric package Gretl (Cottrell and Lucchetti, 2016), and the most appropriated
model will be selected. Item 6 will discuss results obtained for the potential ICMS
revenue using the selected model as well as results for the fiscal effort of each State.
Item 7 will explore a SD model of the ICMS tax capacity and finally, Item 7 will
present final conclusions.
2) DISCUSSION
The potential revenue of a particular country or state is the maximum revenue
that the government would be able to raise, given their socioeconomic conditions, as
well as the legal framework of their taxes. Hence, there are two concepts of potential tax
capacity, one from the legal and the other from the structural point of view, according to
Viol (2006).
The legal potential tax capacity is related to what the government demands from
taxpayers based on current tax legislation. The potential revenue would then be that
maximum possible revenue resulting from the complete application of the current tax
system. The taxable basis predicted in the legislation and current rates to be applied
should be considered to measure the legal potential. This is the deterministic method of
measuring tax capacity introduced by Carvalho et al (2008).
As for the structural potential, there is less clarity in their outlines and greater
difficulty in their measurement. Its estimate is made using econometric models, where
the tax becomes the dependent variable of other explanatory variables that reflect the
socioeconomic characteristics of a given country or state. There are several literature
works that estimate the potential tax capacity considering this structural approach, both
internationally and in Brazil.
From the concept of potential tax capacity, one can derive the concept of Tax
Effort Index (TEI), or degree of effectiveness, according to some authors. This index is
calculated from the ratio of tax revenue, which effectively enters in the public coffers,
and the potential revenue, which is estimated by an appropriate structural econometric
model. The TEI is used to make comparisons of fiscal effort among countries, as well as
among federal units of a given country.
At the international level, many authors have studied variables and tax capacity
models of countries. Using cross-section data in 1964, Lotz and Morss (1969) were the
first authors to confirm the positive influence of per capita income and degree of
openness of economy. Shin (1969) discussed the significance of per capita income,
agricultural product and population growth variables in the analysis of cross-section
data. Chelliah (1971) showed that ratio of extractive industry product variable was
highly significant, degree of openness was significant and per capita income was not
significant. Bahl (1971) confirmed the significance of the agricultural product and the
mining industry product, and the tax capacity related negatively with the first and
positively with the second. Tait, Gratz and Eichengreen (1979) updated the results of
Lotz and Morss, as well as Chelliah using cross section data in 1974, and they
concluded that the variables of the most explanatory power were mining industry
product and degree of openness. Mann (1980) studied the tax capacity of Mexico, using
time series, and he concluded that degree of openness, per capita income and
agricultural products were significant at certain periods of time while only per capita
income was significant and inversely related to tax capacity at more recent time.
Piancastelli (2001) used both cross section data with the average for the period 1985-95,
as well as panel data, concluding that for the total sample studied, per capita income and
degree of openness of economy were significant. However, when the sample was
divided into low and middle income countries, he found that only degree of openness
became significant for the low income group while agricultural and industry products
influenced tax capacity negatively and positively, respectively, for the middle income
group. Cafe (2003) estimated tax capacity of industrialized countries and Latin America
countries, concluding that per capita income and degree of openness were significant
and positively related to tax capacity of the full sample of countries, while for separated
groups, there was an improvement in the linear adjustment when the agricultural
product variable was added to the model.
Several studies establish comparisons among the tax capacity of the Brazilian
States. Reis and Bianco (1996) used production function models with panel data for the
years 1970, 1975, 1980, 1985 and 1990, obtaining expected results for GDP, urban
population and inflation. Marinho and Moreira (2000) estimated the potential tax
capacity of the Northeast Brazilian States for various taxes in the period between 1991
and 1996, also using models of production function with panel data, obtaining
significant and direct relationships between ICMS and per capita income, urban
population and degree of urbanization, and negative relations with exports and inflation.
Vasconcelos et al (2006) used panel data from 1986 to 1999 to estimate the potential tax
burden of Brazilian States, concluding that industry and service products and GDP per
capita were significant and they had positive signs as expected. Carvalho et al (2008)
estimated the Amazon States tax capacity between 1970 and 2000, in the census years,
also using production function models with panel data and they concluded that the
economic and demographic variables used in the model were important to access
potential revenues of States. However, they obtained a negative not expected sign for
industry] product and a not significant relationship. Cafe (2011) estimated the potential
tax capacity of Brazilian States in the 2003-2007 period, using linear regression models,
and she concluded that GDP, population and industry variables were positive and
significant.
3) SYSTEM DYNAMICS MODELING COMBINED WITH STATISTICAL
METHODS
The study considered many variables found in the literature that could affect the
tax capacity like the Gross Domestic Product (GDP), which measured economic
development stage; Exports and Imports, which measured degree of openness; Sector
Products (added-value indices) that measure the degree of industrialization and
urbanization; and population size. Many other variables such as level of economic
inequality (Gini Index), debt (Default Rate), employment (Formal Jobs) and size of the
private sector, may also influence state tax capacity and they will be also considered in
this study. Figure 1 shows a preliminary causal loop diagram that represents the
relations between state tax capacity and its explanatory variables.
Population
Foreign Trade
Fuel Sales
State Tax Capacity
Provision of Publ
ods
Prvate Sector e
Inequality of
Household Income
Production ,
+
Consumption
WA. Y~agicutual
Production
Industrial
Services :
Production Pryucton
Degree of
Urbanization
Figure 1: ICMS State Capacity
Preliminary Causal Loop Diagram
The analysis of variables took into account interventions and expected signs, as
well as level of correlation among them, always considering specific economic and tax
aspects of Brazilian states. An explanation of the methods, besides the analysis of each
variable and concems about their interrelation are shown below.
Multiple linear regression models and the method of Ordinary Least Squares
(OLS) were used to estimate the potential ICMS tax capacity of Brazilian States. For
that, the potential revenue y; of each State was estimated by the following equation,
according to Wooldridge (2010):
F= Bo + Bi Xai + Bo Xait... +P XKi + wi
onde,
¥, ~- Estimated value for ICMS revenue of each State;
i - Index which represents each State (from 1 to 27);
K_ - Index which represents the number of explanatory variables;
xxi - Explanatory variable K of State i;
Bx - Parameter to be estimated for each explanatory variable K;
Bo - Intercept.
The value f;,, calculated by this equation, estimates the actual value y; of
potential ICMS revenue for each State. The difference between the estimated and the
actual value is represented by a residue i1;. Therefore:
wi VK
Where,
¥, - Estimated value of potential ICMS revenue of each State;
y; - Actual value of ICMS revenue of each State;
ui - Residue of the State index i.
A vector of B parameter (one parameter Bx for each explanatory variable) must
be chosen in order to make the smallest possible error in the estimation of the ICMS
potential revenue y;. The estimations of Bis accomplished by solving a set of
overdetermined normal equations, which have the following solution:
B= (TY) "XTY
Where,
X27 x K +1) is the design matrix with all measured values of the explanatory
variables. The lines correspond to the index of each State, and the columns correspond
to the index of each explanatory variable.
Y27x1is the column vector with measured values of ICMS revenue of each State.
yl
After the estimation of B parameters, the following tests were applied:
1) Reset Test of Ramsey regression specification error, as in Wooldridge
(2010), especially for omission of variables. This test includes quadratic and cubic
terms in the model, and it verifies if the coefficients of these terms are significant, via F
test:
Ho: Bs =Bs =0
H1: 8540 or Bg #0
If Ho is rejected, the model is poorly specified.
2) Breusch-Pagan Test of heterocedasticity, as in Wooldridge (2010). This test
checks if the variance is affected by some independent variables (u? x B) via F test:
Ho : Bi = By =B3 =... =0
Ho is the null hypothesis of homocedasticity.
3) Variance Inflation Factors (VIF) where values above 10.0 may indicate a
multicolinearity problem, as pointed by Miloca S. and Conjo P. (2011). If a variable is a
linear combination of others, the fit degree R? tends to 1.0, since VIF is given by:
FIV =1/(1-R’)
if R’1, FIV
In this study, Gretl and V ensim softwares were used for analysis and selection of
the most appropriate model in statistical terms, as well as for running the tests presented
above.
4) ANALYSIS OF SOCIOECONOMIC VARIABLES
Several socioeconomic variables were collected for all Brazilian States, which
may serve as explanatory variables to estimate the potential ICMS tax capacity for the
year 2012. Table 1 consolidates this information.
Figure 2 presents a preliminary graphic analysis of ICMS, the dependent
variable of the model. A compensation factor of 1/3 was applied to ICMS data from Sao
Paulo State, in order to avoid distortions. Sao Paulo is clearly an outlier, with
production index and tax revenues way beyond the Brazilian average. The same
compensation factor will be applied to the variable population and to all those variables
related to the economic performance of Sao Paulo (GDP, Ag P, Ind P, Serv _P, Fuel, X
and M).
R=0,9508 2
ICMS (R$ Thous) Millions
ie
15 =
‘ Sa =
§ a2
a ee
RR AC AP TO SE PI AL RO PB RNMA DF MSAMMT PA CE ES PE GO SC BA PR _RS RJ MG SP
Figure 2 - ICMS revenue in R$ billions versus Brazilian States ordered by
ICMS revenue, considering one third of Sao Paulo ICMS revenue.
From Figure 2, the ICMS revenue curve of Brazilian States follows an
exponential trend with degree of adjustment R’ of 0.95. Then the variable logarithm will
be used for the linearization of the model.
‘Table 1 - Socioeconomic variables to estimate potential ICMS tax capacity.
Variables Initials Units Source Explanatory Notes
‘Tax over the Circulation of Goods ICMS. R$ (thousands) CONFAZ Current market prices
Gross Domestic Product GDP R$ (millions) IPEA/IBGE Current market prices
Economically Active Population EAP Individuals IBGE Projections (1992 to 2009 data)
Agricultural Product Ag_P R$ (millions) IBGE Gross Added Value -current market prices
Industry product Ind_P R$ (millions) IBGE Gross Added Value -current market prices
Service Product Ser P R$ (millions) IBGE Gross Added Value -current market prices
Gini Index Gini Index IBGE Inequality income distribution
Fuel Sales Fuel mm ANP Distributor sale
Exports x USS (thousands) |BCB/MDIC/Secex Foreign trade
Imports M USS (thousands) |BCB/MDIC/Secex Foreign trade
Formal J obs FJ Units BCB/MTE (Admissions - layoffs)
Default Rate DR Percentage BCB In credit operations
Proxi for Private Sector PP Proportion RFB (Private sector contributions)/income tax)
The correlation level between the dependent variable (ICMS) and each
explanatory variable was studied. Table 2 shows the values of linear correlation (LC),
which serves as an indication of what variables should be included in potential tax
capacity models.
Table 2 - Linear correlation between ICMS and explanatory variables
represented by their initials according to Table 1.
Variables Lc
GDP 0,959
EAP 0,920
Ag P 0,587
Ind_P 0,891
Ser P 0,979
Gini -0,460
Fuel 0,960
xX 0,873
M 0,867
F 0,624
DR -0,140
PP 0,035
The linear correlation values vary in magnitude between 0.035 and 0.979. The
positive sign points to a direct linear correlation while the negative sign for an inverse
linear correlation. All variables will be regarded as potential explanatory variables in the
ICMS tax capacity models.
Table 2 shows that the relationship between ICMS and variables which indicate
economic performance were positive and high, above 0.85, except for Agriculture Product
(Ag_P). For illustration purpose, Figure 3 shows the relationship between ICMS and
industry product (Ind_P) with positive linear correlation of 0.89. This positive relationship
was expected, as pointed out by Varsano et al (1998).
4, 120
Fa
s Re
ig, 100 R?=0,793
z 80
2
= +
5 60
2
= 40 = *
a e ¥. 2
= 20 ee —
—-
sa oo ad T T T T T
5 10 15 20 25 30 35 40
0) Millions
ICMS (R$ Thousands)
Figure 3 - Industry product (Ind_P) in R$ trillions versus ICMS in R$
billions of Brazilian States. The blue points are observations for each State. The
continuous line represents a linear model adjusted to the observations with R? =
0,793. Notice that Rio de J aneiro State is an outlier.
10
Rio de Janeiro State is an outlier in the relationship between ICMS and Industry
Product, due to the oil production industry. Although the State is a major oil producer,
taxation is at destination in interstate operations with lubricants and oil fuels, according to
the Brazilian Federal Constitution. Thus, ICMS is charged where consumption occurs,
which induces a gap between industry product and ICMS revenue in Rio de Janeiro.
The curves of variables related to economic activity plotted against Brazilian
States ordered by ICMS revenue follow an exponential trend, which suggests the use of
logarithm of these variables in order to obtain linear relationships in the estimation
models of potential ICMS tax capacity. All of these variables, with the exception of
Agriculture Product, have good explanatory power of the dependent variable ICMS, since
the values of fit degree R? are above 0.75.
The level of linear correlation among the variables that are indicative of economic
activity was also studied and a high correlation was found, as expected. Thus, the models
will contain only one indicative variable of economic activity to avoid colinearity
between explanatory variables, respecting this classic hypothesis of the method.
The Economically Active Population (EAP) also has a strong correlation to
ICMS, with correlation index of 0.92. The positive relationship is obviously expected: the
larger the population, the greater the tax capacity, as cited by Varsano et al (1998). The
EAP curve has also an exponential behavior, shown in Figure 4, also indicating the use of
its logarithm for the linearization of the estimation model.
14,000
12.000 *
R?=0,8457
10.000 LN
8,000 * =
WF
6.000
= a
4.000
2.000 - °
an + ra ae
EAP (thousand)
=
RR AC AP TO SE PI AL RO PB RN MA DF MS AM MT PA CE ES PE GO SC BA PR RS RI MG SP
Figure 4 - Economically Active Population (EAP) in thousands of
individuals versus Brazilian States ordered by ICMS revenue. The blue points
represent the observations for each State. The continuous line represents an
exponential model adjusted to the observations with R? = 0,846.
The degree of adjustment R? of 0.846 of ICMS versus EAP indicates a good
explanation power of EAP variable in the ICMS potential estimation model.
11
~i4
=
g 5x ==
8 3 R2=0,846
S F 10
S
£ 8 + 2 o
z =——
5 6 &
3 °
E 4 ~—s
= + °
3 2 SZ oo
0 - -
0 10 20 30 40
Millions
ICMS (R$ Thousand)
Figure 5 - Economically Active People (EAP) in millions of individuals
versus ICMS in R$ billions of Brazilian States. The blue points are the
observations for each State. The continuous line represents a linear model adjusted
to the observations with R? = 0,846.
Gini index measures the degree of inequality in the distribution of per capita
household income among individuals. Its value can vary theoretically from zero, when
there is no inequality, to one when inequality is maximum. The correlation coefficient
found between ICMS and Gini index is -0.46, indicating an inverse relationship
between these two variables.
There is a positive linear correlation of 0.87 between ICMS and both export (X)
and import (M) variables, as shown in Table 1, despite the Complementary Law No.
87/1996, known as Kandir Law. Although this Law exempts from taxation goods and
services for export, increase in ICMS revenue following both exports and imports was
observed. This positive correlation is probably due to the direct effects of imports, since
ICMS is levied on imported goods, and secondary effects of exports, as they move the
economy, creating jobs and increasing income to purchase goods.
Varsano et al (1989) argue in favor of an inverse relationship between potential
ICMS revenue and trade balance, and consequently exports. On the other hand,
Vasconcelos et al (2006) point out that intemational trade is an important source of
income, especially in developing countries.
There is a good degree of adjustment between ICMS and foreign trade variables,
with R? greater than 0.75. In addition, X and Y curves tend to a 2"¢ degree polynomial
equation, suggesting the use of square root of these variables in order to linearize the
models. Figure 6 illustrates the curve of Imports (M).
12
w
38
Millions
N
a
°
R?=0,7639
A
20 Kd
M (US$)
7"
a
o
* ©.
5
°
« die!
0 vm
RR AC AP TO SE PI AL RO PB RNMA DF MSAMMT PA CE ES PE GO SC BA PR RS RJ MG SP
Figure 6 - Imports of goods (M) versus Brazilian States ordered by ICMS
revenue. The blue points are the observations for each State. The continuous line
represents a 2"! degree polynomial model adjusted to the observations with R? =
0,764.
The variable of formal jobs generated in the year (FJ) is the difference between
admissions and layoffs in 2012. This variable is an indirect indicator of economic
performance and therefore establishes a positive correlation with ICMS of 0.62.
However, the ICMS versus FJ showed a degree of adjustment R’ of less than 0.5,
indicating that FJ has a low explanation power in the ICMS estimation model.
According to Table 1, the variable default rate on credit operations (DR) has a
negative correlation with ICMS of -0.14, indicating an inverse relation but probably,
DR has a low explanatory power of the variable ICMS, confirmed by the low degree of
adjustment R? of 0.02. The negative sign of this correlation was expected, since the
default rate is indicative of debt, which is related to low family income and low
consumption power.
The proxy of private sector (PP) used is the ratio between private sector
contributions and income tax. This variable has a low correlation of 0.03 with ICMS,
which induced an extremely low degree of adjustment R? of 0,001 in the linear relation
between ICMS and PP. Despite the expected positive signal at first, as services offered
by the public sector are not in the tax incidence field, Varsano et al (1989) argued for a
negative sign. According to them, a greater participation of the States in the provision of
services, such as education and health, will induce a greater tax capacity, since this
provision would replace purchase of such services in the market, freeing up more
resources for private consumption.
5) SELECTION OF ICMS TAX CAPACITY MODEL
Several models have been tried with Gretl program (Cottrell and Lucchetti,
2016), including the explanatory variables described in the previous section, and
containing only one variable of economic performance. The goal in any attempt was to
obtain a model that contains all significant variables to at least 10%, expected signs and
satisfactory behavior in Reset, Breush-Pagan and colinearity tests.
The F statistic showed significance at 1% of the set of variables in all tested
models, indicating that the set of explanatory variables can effectively be used to
13
estimate the dependent variable (ICMS). However, only one model was able to meet the
conditions placed above and therefore, this one was selected.
The selected model includes industry product (Ind_P), population (EAP) and
import (M) variables, significant to 1%, 1% and 10%, respectively, with expected
signals, and intercept at 1%. The model showed good fit and good results in all tests.
However, considering that the coefficient obtained for M variable was very small
(0.0001), and using the principle of parsimony, this variable was removed from the
model. Thus, the multiple regression model finally selected for the estimation of
potential ICMS tax capacity includes industry product and population variables. Table 3
presents the selected model.
Table 3 - Selected Model for Potential ICMS Tax Capacity Estimation
In(ICMSS) = 5,840 + 0,405In(Ind_P) + 0,433In(EAP)
n = 27 observations (Brazilian states)
coeffici: Jard error | t isti p-value
const 5,83989} 0,458635 12,73}3,63e-012 ***
_Ind_P 0,404767 0,0720932, 5,614] 8,85e-06 ***
|_EAP 0,432982 0,124151 3,488] 0,0019 _***
R2 adjusted 0,950665,
F(2, 24) 251,5047|F (p-value) | _7,96E-17
The estimated coefficients are indeed measurements of elasticity, since it is a
double logarithmic model. Thus, controlled for the EAP variable, 1% of Ind_P increase
represents an ICMS increase of 0.405% for the set of States. Similarly, controlled for
the Ind_P variable, 1% of EAP increase represents an ICMS increase of 0.433%.
6) RESULTS
Figure 7 shows the comparison between logarithm values of effective ICMS and
ICMS values estimated by the selected model. The estimation has an average deviation
of 0.004 or 0.05% of logarithm of effective ICMS.
RR AC AP TO SE PI AL RO PB RN MA DF MS AM MT PA CE ES PE GO SC BA PR RS RI MG SP
—*-LICMS —@|_ICMS estimated
Figure 7 - Comparison between logarithm of effective ICMS (in blue) and
logarithm of ICMS values estimated by the model (in red).
14
The Tax Effort Index (TEI) was calculated from the estimated ICMS tax
capacity. The largest deviations obtained between actual and estimated ICMS are
reflected in this index since it is defined as the ratio between effective and potential
ICMS. Figure 8 shows TEI curve for Brazilian States, whose index values range
between 0.61 and 1.86. Index values below 1.0 indicate that States can increase its
ICMS revenue, while values above 1.0 indicate that States collect higher revenue than
what would be expected from their bases.
Figure 8 - Tax Effort Index of Brazilian States (base year 2012).
From Figure 8, 14 States obtained TEI equal to or above 1.0, and 13 States
below 1.0. Distrito Federal, Sao Paulo and Mato Grosso do Sul presented the highest
TEI of 1.86, 1.48 and 1.38, respectively. On the other hand, Para, Sergipe and Acre TEI
were the smallest of 0.68, 0.70 and 0.72 respectively.
There is a natural difficulty to accept higher rates than 1.0. Those values can be
explained by the fact that the method (OLS) gives an average curve that minimizes the
square sum of the errors, separating points above and below the average curve.
Consequently, for some States, this may lead to an overestimation of their potential
revenue. A second possible explanation is related to the State law that provides effective
revenue greater than the one derived from the State economic bases. Finally, there are
situations that can increase exogenously the effective collection.
In 2012, the base year of the study, there were exogenous factors such as an
anticipated ICMS on electricity, as well as, incremental revenue arising from a credit
recovery program that may explain the high TEI of 1.86 obtained by Distrito Federal.
The case of Mato Grosso do Sul can also be explained by an exogenous factor to its tax
capacity, which granted an injunction that guarantees the ICMS tax on the Bolivian
natural gas import operations to the State.
On the other hand, Para had the lowest TEI among the Brazilian States, related
to the fact that Para is a major producer and exporter of iron ore and aluminum, which
increases its industry product, but this activity is not taxable by ICMS. Similar case
occurs with Rio de Janeiro and Sergipe. Both States are oil producers, which increase
their industry product, but ICMS taxation on oil and its derivatives at interstate
operations occurs in the destination State, not in the producer State.
7) GST TAX DYNAMICS
These econometric analyses above gave insights to identify the variables that
work as stocks and to produce the following stock and flow diagrams to represent the
GST dynamic model and its interrelations. This diagram considers the actual structure
of the model, including stocks, flows and external inputs (Sterman, 2000).
In the GST tax dynamic model, there are three majors stocks identified by the
selected econometric model: state tax reserve, originated by the state tax revenue
15
collected - the dependent variable of the econometric model, industry product and
population, both significant explanatory variables. The following topics will present the
step by step construction of GST tax dynamics model.
7.1) Industry Product Stock
Figure 9 shows the relations among the industry product stock and relevant
variables, and the interactions with the ICMS tax collection. The industry product stock
is subjected to the interest rate/consumption/investments flow. The higher the interests
rate of the economy, the lower the investments and the lower the production.
Additionally, the higher the interests rate, the lower the consumption, which implies the
lower the demand rate of production and consequently, the lower the production itself.
More supply explains the production of goods and services, which induces an
increase in consumption (sales) and in state tax capacity.
Interests Rate
oO SS -——%
IC}S'tax revenue er fi ave G overnment
Teas + rate expenditures rate
Figure 9 - State Tax Reserve / Industry Product relations.
7.2) Population Stock
Consumption is also based on size of population and on employment (number of
jobs). Figure 10 includes the population stock and its positive relation with consumption
and negative with employment. In other words, as the size of population increases, less
job positions are available, which in turn causes debt. Both employment and debt affect
consumption, the first directly and the last inversely.
Besides, consumption is reduced when inequality of household income
increases, measured by the Gini index in previous econometric analyses, which also
increases when employment decreases.
o
16
Interests Rate
oO as STATE TAX ->———2—O
- ToMSax reveme | “RESERVE Govemment
. + rate expenditures rate
Investments
if 7” Debt
4 Inequality of
—_ nequality of 4 —* Employment
qemei ae Household Income Y
INDUSTRY
PRODUCT
supply
POPULATION}
bithrte = LI] mortality rate
Figure 10 - Addition of population stock to the GST model.
7.3) OPENNESS OF THE ECONOMY - Goods and Services Export and
Import
This entire picture is influenced by the openness of the economy, as shown in
Figure 11 below. First, the number of jobs increases with the openness of the economy
due to the increase of labor mobility, and consumption in general tends to increase,
which in turn speeds up ICMS tax revenue collection.
Goods export increases the demand rate for products, which in turn leads to an
increase in production. The increase in production also tends to increase goods export.
Goods export doesn’t affect ICMS tax revenue rate directly because, as mentioned
before, exports of goods and services were exempted from ICMS tax.
Goods import tends to reduce consumption of national products, which leads to a
decrease in ICMS tax revenue rate. On the other hand, goods import increases ICMS tax
revenue rate directly because ICMS tax is levied on imported goods.
STATE TAX
ICMS taxrevemue| RESERVE
re a expenditures rate
Interests Rate
0 9f the
+ Goods Import poets
Debt
INDUSTRY
PRODUCT
supply rate
Goods Export
birth rate mortality rate
Figure 11 - GST dynamic model, including openness of the economy variables
(foreign trade).
17
7) CONCLUSION
Measuring the structural potential tax capacity, based on the socioeconomic
characteristics of state or country is not a trivial task. It is necessary to characterize the
tax base by capturing effects of capacity of tax contribution (GDP and population),
composition of the economy (sector products), foreign trade, degree of urbanization and
others to construct models to estimate state or country tax capacity.
This paper studied a large set of variables that might explain tax capacity,
including GDP, sector products and fuel sales variables used to measure economic
performance. Exports and imports of goods were used to measure the degree of
openness of the economy. As far as socioeconomic parameters, besides size of
population widely used, number of formal jobs, default rate on credit operations which
indicates debt, Gini index of income inequality and finally, a proxy of private sector
were adopted.
The preliminary study of the dependent variable (ICMS) and the behavior of
potential explanatory variables were very useful for building models. This study
consisted of a graphical analysis of variables for the set of States, verifying trends and
presence of outliers. In the case of ICMS variable, for example, exponential trend was
observed, as well as presence of an outlier that would be Sao Paulo revenue, which is
far higher than the other state revenues. Thus, measures were taken to correct possible
distortions that could happen, such as using ICMS logarithm aimed the variable
linearization, and employing a third of Sao Paulo ICMS revenue. It was also found
exponential behavior of GDP, population, sector products and fuel sales, indicating the
use of these variables logarithm as well. Regarding trade variables, it was found
quadratic trend, indicating the use of these variables square root.
Additionally, analysis of correlation level and signs between ICMS and
explanatory variables was important to select which variables had power to explain it as
well as the expected signs. Thus, positive and over 0.8 correlations were found between
ICMS and GDP, population, industry and service products, fuel sales, exports and
imports. Negative and between 0.5 and 0.8 correlations were found between ICMS and
Gini index, and positive between ICMS and formal jobs and agriculture product.
Finally, low correlations below 0.2, negative and positive, were found between ICMS
and default rate and between ICMS and private sector proxy, respectively.
On the other hand, high levels of correlation between independent variables
possibly introduced colinearity issues in models. In this study, it was observed a high
degree of correlation among the variables that were indicative of economic activity,
leading the use of only one variable of this group in the estimation models.
Several models using the method of OLS in Gretl were tested. In general, it was
obtained a good degree of fit in almost all models tested and they all showed
significance of the set of variables using the F test. However, few models presented all
variables significant at least 10%, in addition to be simultaneously successful in Reset
test for good model specification, Breusch-Pagan test for heteroscedasticity and
colinearity test.
Choosing the best model should be based on objective criteria, for example,
containing all statistically significant variables, expected signs of coefficients and good
degree of adjustment, besides meeting the requirement tests and the classic assumptions
of the OLS method. It should be taken into account the principle of parsimony in all
cases to choose the simplest model that still meets the requirements. According to that,
the selected model was one that explains ICMS logarithm by industry product logarithm
and population logarithm.
18
The potential tax capacity results were used to calculate the Tax Effort Index
(TEI) of Brazilian States. The highest TEI obtained were 1.86, 1.48 and 1.39 for Distrito
Federal, Sao Paulo and Mato Grosso do Sul, respectively. Exogenous situations to the
economic base of these States discussed before raised their index. On the other hand, the
lowest results of IEF of 0.61, 0.70 and 0.72 were observed in the States of Para, Sergipe
and Acre, respectively, because some major economic activities of these States,
especially the first two, are not reached by ICMS legislation, already discussed.
The Tax Effort Index is a useful tool for analyzing fiscal performance, which
allows comparisons between countries or states. It can even be considered for feasibility
studies of tax burden raising or even as a guide to tax enforcement actions. However,
TEI should not be used mechanically as an absolute truth. Its calculation is linked to
econometric models estimation, which always requires additional analysis and
verification of results.
Econometric models don’t capture the feedback relations between factors.
Because of that, a combination of econometric studies with system dynamic models
seems to be a promising way to explain ICMS behavior and its interaction with
socioeconomic variables. Therefore, a SD model was built step by step, which helped to
have a comprehensive view of the GST dynamics, quite important for the Brazilian
States in their current economic situation.
Further studies will include time dimension in the econometric model by using
panel data. The obtained relations between GST and explanatory variables will be added
in the SD model in order to obtain feedback responses and improve the understanding
of the GST dynamics.
REFERENCES
BANCO CENTRAL DO BRASIL. Economia Regional. Available in:
<https://www3.bcb.gov.br/sgspub/localizarseries/localizarSeries.do?method>
pararT elaLocalizarSeries >. Access in: 10 Nov. 2015.
BAHL, R. W. A Regression Approach to Tax Effort and Tax Ratio Analysis.
IMF Staff Papers, Washington, v. 18, n. 3, p. 570-612, Nov. 1971.
CAFE, P. F. M. Reforma Tributaria: Panorama Internacional e Aspectos
Quantitativos. 2003. 111 f. Dissertatio (Master's Degree in Economcs) -
Economy Departament, Universidade de Brasilia, 2003.
CAFE, P. F. M. Um Estudo do Potencial de Arrecadacao do ICMS e do Indice
de Esforgo Fiscal dos Estados Brasileiros. Caderno de Financas Publicas,
Brasilia, n. 11, p. 5-296, Dec. 2011.
CARVALHO, D. F.; OLIVEIRA, C. C. R. de; CARVALHO, A. C..
Desigualdades Inter-Regionais, Capacidade Tributaria e Esforco Fiscal dos
Estados da AmazOnia (1970-2000): uma abordagem econométrica de fronteira
estocastica. Novos Cadernos NAEA, [S.I.], v. 10, n. 2, p. 5-44, Dec. 2007.
19
CHELLIAH, R. J. Trends in Taxation in Developing Countries. IMF Staff
Papers, Washington, vol. 18, n. 2, p. 254-331, Jul. 1971.
CONFAZ - CONSELHO NACIONAL DE POLITICA FAZENDARIA.
Boletim do ICMS. Available in:
<https://www.confaz.fazenda.gov.br/legislacao/boletim-do-icms>. Access in: 25
Jun. 2015.
COTTRELL, A and LUCCHETTI, R, “Gretl User's Guide: Gnu Regression,
Econometrics and Time-series Library’; <http://gretl.sourceforge.net/gretl-
help/gretl-guide.pdf>, Access in: 1 Mar., 2016
IBGE - INSTITUTO BRASILEIRO DE GEOGRAFIA E ESTATISTICA.
Sistema de Contas Regionais. Available in: <
http://www. ibge.gov.br/home/estatistica/economia/contasregionais/2013/default.
shtm>. Access in: 25 Jun. 2015.
oe . Sistema IBGE de Recuperacao Automatica - SIDRA. Populacao.
Disponivel em: <
http://www.sidra.ibge.gov.br/bda/tabela/listabl.asp?z=t&0=25&i=P &c=3153>.
Access in: 8 Nov. 2015.
IPEA - INSTITUTO DE PESQUISA ECONOMICA APLICADA. Base de
dados Regional. Contas Regionais. Available in: http://www.ipeadata.gov.br/.
Access in: 20 Jan. 2016.
-Base de dados Social. Renda Nivel: Estados. Disponivel em:
http://www. ipeadata.gov.br/. Access in: 25 Fev. 2016.
MANN, A. J. Economic Development and Tax Structure Change: Mexico,
1895-1975. Public Finance Review, [S.1.], v. 8, n. 3, p. 291-306, Jul. 1980.
MARINHO, E. L. L.; MOREIRA, A. F. Esforgo Fiscal e Carga Tributaria
Potencial dos Estados do Nordeste. Revista Econémica do Nordeste, Fortaleza,
v. 30, special n., p. 634-651, Dez. 1999.
MILOCA, S. A.; CONEJO P. D. Multicolineariedade em Modelos de Regressao.
In: XXII SEMANA ACADEMICA DA MATEMATICA, 2008, Cascavel.
Anais eletrénicos... Cascavel: Unioeste, 2008. Available in:
<https://www.ime.usp.br/~yambar/M1404Metodos% 20Estatisticos/A ula%208-
9% 20Regress% E30% 20mult% 20dim/inete% 20adicional% 20-
%20multicolinearidade% 20em% 20modelos%20de%20regressao.pdf> Access
in: 8 Fev. 2016
20
PIANCASTELLI, M. Measuring the Tax Effort of Developed and Developing
Countries. Cross Country Panel Data Analysis, 1985/95. Texto para Discussao
IPEA, Rio de Janeiro, n. 818, p. 1-18, Set. 2001. Available in:
<http://www.ipea.gov.br/portal/images/stories/PD Fs/TDs/ingles/dp_103.pdf>
Access in: 25 Sep. 2015.
PRADO, S. R. R. do. A "questao federativa" no Brasil: ampliando a agenda. In:
PISCITELLI, R. B. et al. Reforma tributaria: a costura de um grande acordo
nacional. Sao Paulo, Editora Atlas, 2009. chap. 5, p. 170-181.
REIS, E. J.; BLANCO, F. A. Capacidade Tributaria dos Estados Brasileiros,
1970/90. Texto para Discussao IPEA, Rio de Janeiro, n. 404, p. 1-26, Fev.
1996. Available in:
<http://www.ipea.gov.br/portal/images/stories/PD Fs/TDs/td_0404.pdf>. Access
in: 30 Jan. 2016.
SECRETARIA DE FAZENDA DE ESTADO DO DISTRITO FEDERAL.
Arrecadacao de Origem Tributaria do Distrito Federal. Gerencia de Estudos
Econ6micos e Politica Fiscal, Dec. 2012.
SHIN, K. International Difference in Tax Ratio. The Review of Economics and
Statistics, Massachusets, v. 51, n. 2, p. 213-220, May 1969.
STERMAN, J. D. Busi Dynamics: Thinking and Modeling for a
Complex World. Irwin McGraw-Hill. 2000, 982 p.
TAIT, A. A; GRATZ W. L. M.; EICHENGREEN, B. J.. International
Comparisons of Taxation for Selected Developing Countries, 1972-76. IMF
Staff Papers, Washington, v. 26, n. 01, p. 123-156, Mar. 1979.
VASCONCELOS, J. R. de; PIANCASTELLI, M.;MIRANDA, R. B. Esforco
Fiscal dos Estados Brasileiros. Texto para Discussao IPEA, Brasilia, n. 1045,
p. 1-39, Set. 2004. Available in:
<http://www.ipea.gov.br/portal/images/stories/PD Fs/TDs/td_1045.pdf>. Access
in: 30 Set. 2015.
VARSANO, R. et al. Uma Anilise da Carga Tributaria no Brasil. Texto para
Discussao IPEA, Rio de Janeiro, n. 583, p. 1-55, Aug.. 1998. Disponivel em:
http://www. ipea.gov.br/portal/images/stories/PD Fs/TDs/td_0583.pdf. Access in:
07 Fev. 2016.
Ventana Systems. (2015). Manual of Vensim version 5. Retrieved from
http://www.vensim.com/
24
VIOL, A. A Definicao e o Cémputo da Arrecadacao Potencial. Analise do
Potencial Econémico-Tributario e de seus Condicionantes. In: Assembleia
Geral do Centro Interamericano de Administragées Tributarias (CIAT), 40.:
Florianopolis, Apr. 2006.
WOOLDRIDGE, J. M. Introdugao a Econometria: Uma Abordagem
Moderna. 4 ed. Sao Paulo: Cencage Leaming, 2014. 701 p.
