Consumer Behavior and Public Policies: Empirical Evidence through vec model on Brazil’s Automotive Industry

Source: Prepared by authors using data from Economic Department of the BCB (Central Bank of Brazil).

Brazilian federal government, led by President Lula, adopted the policy of reducing the tax on industrialized products (IPI) on new cars through the MP (provisional measure) no. 451/08 as of 12 December 2008, (BRASIL, 2008). This reduction, despite having as limit March 2009, was extended a few times and lasted until the end of March 2010. Tax on Financial Transactions (IOF) for credit to individuals was also reduced from 3% to 1.5% per year, which ended up increasing the credit facilities for the purchase of vehicles.

Alves and Wilbert (2014) reported that the purpose of the IPI reduction was to encourage domestic consumption after the global financial crisis of 2008. In addition, it had the objective of reducing the stock of domestic automakers, allow an increase workers' purchasing power and avoid layoffs in the auto industry.

The IPI reduction on vehicles had different levels according to the cubic centimeters(cc) and the used fuel. Vehicles up to a thousand cc (1.0) had the IPI reduced IPI from 7% to zero. Vehicles between one and two thousand cc and gasoline-powered had an IPI reduction from 13% to 6.5%, but flex vehicles between one and two thousand cc had and IPI reduction rate from 11% to 5.5%. For imported vehicles and cars of more than two thousand cubic centimeters, the rates had not changed. Commercial vehicles (light commercial) had a reduced IPI rate from 4% to 1%.

In May 2012, government led by Dilma Rousseff reduced again the IPI rates on national vehicles, and the IOF (financial operations tax) for credit to individuals. Including auto loans rates dropped from 2.5% to 1.5% per year, according to Decree No. 7726/12 (BRASIL, 2012). The argument used to adopt the policies was to stimulate economic activity to fight the worsening of the international financial crisis and to avoid layoffs in the auto industry.

Between May and December 2012, the IPI rates on vehicles nationally manufactured had the same reduction as the ones used before, on December 2008: from 7% to 0% (to 1.0 cc vehicles), from 11% to 5.5% (to flex vehicles of more than 1.0 cc and up to 2.0 cc), from 13% to 6.5% (gasoline-powered vehicles of more than 1.0 cc and up to 2.0 cc) and from 4% to 1% to commercial vehicles.

In 2013, the IPI rates on national vehicles were, according to Alves and Wilbert (2014), 2% for cars of 1.0 cc, 7% to flex vehicles of more than 1.0 cc and up to 2.0 cc, 8% to gasoline-powered vehicles of more than 1.0 cc and up to 2.0 cc and 2% for commercial vehicles.
3. Literature Review

Market interventions through public policies are a result of pressure from business groups as stated in Becker (1983), Vieira and Gomes (2014) and Xavier, Bandeira-de-Melo and Marcon (2014) and the efficiency of such policies has been a question addressed by studies such as Bullock (2005).

In fact, public policies have been the objective of researches focusing the reasons and consequences of policies on markets. Shaffer (1995) shows business-government relations developed within a managerial framework to discusses the consequences of public policies for the competitive environment of the firm that determines firm-level responses which include both strategic adaptation and attempts to influence government.

Ball (1995), examines the social costs and benefits of interest groups with political influence. The study concludes that interest groups may not be that bad as the information those groups have may enable governments to choose better policies and this way lobbying can enhance welfare.

Lazzarini (2011), explaining Brazil’s capitalism dynamics, open the discussions for and against government intervention on markets and shows the reasons for the adoption of public policies based on the power of relationships among Brazilian enterprises owners and the government.

Public policies in fact tend to favor powerful groups that pressures government to apply policies arguing in general high costs or economic environment and others reasons that may lead to massive unemployment. This is the case of the vehicles market in the country.

According to Zimmermann, Egan and Poli (2011), Brazil’s strong domestic demand, along with the tax reduction policy made possible the country’s automobile sector to grow compared to other sectors.

The adoption of a policy of tax reduction will have different effects on consumers and producers. On consumers the tax reduction leads to two different effects, the income effect and the substitution effect (Varian, 2006). The income effect happens because the price of cars (considered here as a study case) relative to the price of other products became cheaper with the reduction of tax which represents a change in real income, and for that reason consumers may decide to buy a car instead other products using the opportunity faced which is the substitution effect.

On the producers’ side, the tax reduction will change producers decisions of reducing the number of employees or closing the facture which probably were the main arguments used to make the government apply the policy of tax reduction and lower the credit rates. The response will be the sales increase which will have an economic impact (Lersh, 2004).

The impact of tax reduction on the vehicles industry was first measured by the Board of macroeconomic studies of the Applied Economic Research Institute (IPEA), that using a simple linear regression model between June 2003 and June 2009, considering the number of cars sold as a function of their prices, income and loans obtained that 13.4% of the cars sold were due to the reduced IPI (DIMAC, 2009).

Alvarenga et al. (2010a, 2010b) also analyzed the IPI reduction and the credit impacts on the sales of vehicles. An econometric time series model was used to analyze the co-integration of series seasonally adjusted of automobiles and light commercial, whole sale price index (IPA) for industrial products, income as measured by GDP at current prices and credit measured by the consolidated concessions of loans for purchase of vehicles. Prices, GDP and credit were deflated by the consumer price index (IPC). The study considered different scenarios: if there had been no IPI reduction in the IPI, if there had been IPI reduction and 5% increase in credit extensions, if there had been no IPI reduction and there was an increase of 5% on credit.

The authors found that 20,7% of the increase in sales of cars between January and November 2009 was due to the IPI reduction. The effect of credit was not negligible and increased with the IPI reduction.

Alves and Wilbert (2014) estimated linear regression model, considering the sale of cars a function of average income, available credit to individuals, a time trend variable and IPI as a dummy variable. The authors divided the regression into two periods: from January 2006 to March 2010 and from April 2010 to August 2013. The study found no significant impact of IPI reduction on sales in both periods in contradiction to the results obtained by Alvarenga et al. (2010a, 2010b) and DIMAC (2009).
4. Methodology

To measure the impacts of a policy we tested two policies applied by the Brazilian government on the vehicles sector. So, quantitative method is used to test IPI and IOF rate reduction effects on cars and light commercial vehicles produced in Brazil. This section is divided into two subsections. In the first subsection it is shown the model to be used. The second subsection presents the methodological procedures used to define the appropriate model for the variables of interest and used in the simulations.

This study adopts a time series model in which the number of vehicles is a function of price, income and credit concessions, s = f (p, i, c),where:

a) s = Sales: car count and new licensees national light commercial (in units). This variable is used as a proxy for wholesale sales in the domestic market. Source: National Association of Vehicle Manufacturers (ANFAVEA, 2012);

b) p =Price: Wholesale Price Index (IPA) origin - industrial products (motor vehicles, trailers, truck bodies and parts) - monthly) .Source: Getúlio Vargas Foundation (FGV,2015);

c) i= Income: Average nominal income of the main job, actually received in the reference month, for persons 10 years or older, employed during the reference week, by regions. Source: Brazilian Institute of Geography and Statistics (IBGE,2015).

d) c= Credit: consolidated concessions of loans with referential free resources for the purchase of goods vehicles (R$ million monthly). Source: Economic Department of the Brazilian Central Bank (Banco Central do Brasil,2014);

e) IPC - general: consumer price index, which will be used to deflate the price series, income and credit. Source: Getúlio Vargas Foundation (FGV,2015).

The data is monthly and the period extending from June 2002 to December 2012. The option to deflate by the IPC the price, income and credit series is justified by developments in prices that is understood by the consumer. The statistical software EViews in version 5.0 was used and the series were considered in terms of the logarithm (natural), that is, LSALES, LPRICES, LCREDIT and LINCOME to facilitate the estimation of these variables as that Alvarenga et al. (2010a, 2010b) carried out in his study.

After the seasonal adjustment using the method developed by the US Census Bureau, X-12 ARIMA contained in EViews 5.0 and putting the series in natural logarithm, the series have the following specification: LSALES_SA, LPRICES_SA, LCREDIT_SA , LINCOME_SA.
4.2 Methodological procedures

According to Bueno (2011) and Enders (2004), a time series is called weakly stationary or stationary if:

a) , , that is, the second off-center point must be finite;

b) , that is, the average should be constant over time;

c),that is, the variance should be the same in all periods of time and the autocovariance should depend solely on temporal distance between observations.

One of the most popular tests for the presence or absence of unit root is the Augmented Dickey-Fuller test (ADF). In this test, the null hypothesis is that there is a unit root and the process is not stationary. The alternative hypothesis is that there is no unit root and thus the process would not be stationary. So, the decision rule for the ADF test is:

a) p-value >, do not reject . There would be a unit root and the process would not be stationary;

b) p-value <, rejects . Therefore, there would be no unit root and the process would be stationary.

This investigation does not sought to determine the order of integration based on the ADF test considering the presence of structural breaks in the series due to the 2008 crisis. The justification is that the tests of Dickey and Fuller (DF), Dickey Fuller (ADF) and Phillips and Perron (PP) have low power, for "when there is a structural break tests [DF, ADF and PP] lead to biased results in not rejecting the null hypothesis of a unit root when in fact the series is stationary "(Margarido and Medeiros Junior, 2006, p.151). Thus, DF , ADF and PP tests have low power because there is a high probability of committing the error of type 2 (not reject when is false).

Here, we used only the KPSS test developed by Kwiatkowski, Phillips, Schmidt and Shin. The KPSS test reverses the null hypothesis and ADF test alternative and may, therefore, "distinguish the unit root series whose data are not conclusive enough" (Bueno, 2011, p.129).

The hypotheses of the KPSS test are:

a):= 0, there is not a unit root and the process is stationary;

b) :>0, there is a unit root and the process is not stationary.

Still according to Bueno (2011) the KPSS test statistic is based on Lagrange multiplier (LM) and formalized as follows:

KPSS=

where is the partial sum of the residues, T is the number of observations and is the long-term variance. Since the decision rule is the following:

a) KPSS * < KPSSc (α) does not reject . Therefore, there would be no unit root and the process would be stationary;

b) KPSS *> KPSSc (α) rejects . There would be a unit root and the process would not be stationary.

In this study the series sales, prices, credit and income are all integrated of order 1, according to the KPSS test performed for the period between June 2002 and December 2012.

Because of the presence of series with a unit root, it is important to check for a long stationary run equilibrium relationship. According to Engle and Granger (1987) series that form a vector order (n x 1) are cointegrated of order (d,b) and called for ~ CI(d, b) if:

a) all elements are I (d), integrated order d; and

According to Bueno (2011), it can be stated that there is a long-term balance between two variables if = 0, this is, if the vector establish a linear combination of the variables , it follows a common trend, without deviation. Short-term deviations are represented by . When two series are cointegrated, the residuals of the regression involving variables is stationary, that is, order 0 and there is a long-term relationship between sets.

From (1) it can be pre-multiplied by the inverse of A, that is, by , in order to obtain the reduced form:

+

+ (2),

Where

On the other hand, a vector error correction model (VECM) would be a model to correct a problem of VAR. This problem is the fact that the VAR model only considers differentiated variables I (0) or non-stationary variables. According to Bueno (2011), a VEC( ) model can be written as follows³:

The terms and are, respectively, the adjustment cointegration matrix and matrix. While is related to short-term adjustment, is related to long-term relationship between the variables.

In this study, as the variables were I (1), it was performed a procedure described by Johansen and Juselius (1990) to detect the presence or absence of cointegration, that is, the presence or absence of long-term relationship between the variables.

Before carrying out the Johansen and Juselius procedure, the number of lags in the VAR model was determined, which was obtained with the Akaike information criterion (AIC), Hannan Quinn (HQ) and Schwarz (SBC or BIC). This study sought to use the criterion of Schwarz, because AIC tends to overestimate the asymptotic order of the VAR.

For all criteria, the ideal lags number is two and that was that was the number of lags considered for cointegration and considered in the Johansen and Juselius (1990) procedure which is composed of two tests based on a restricted maximum likelihood estimation. The first is the trace test, where hypotheses are:

a): r = r*⁴

b):r>r*.

The second test used is the maximum eigenvalue test, which also has an unconventional distribution, and the hypotheses are:

a): r=r*;

b): r=r*+1