Location of Sugar Cane Production and Fuel Demand in Brazil

Transcription

1 Location of Sugar Cane Production and Fuel Demand in Brazil Gervásio F. Santos 1 The objective of this paper is to evaluate the regional differences in the fuel consumer s behavior induced by regional production of ethanol in Brazil. In the same way that other economic activities in Brazil, the sugar cane plantations and consequently the ethanol production, are spatially concentrated. The hypothesis of the paper is that regional differences in the productivity in the ethanol production and transport costs may affect competitiveness of the fuels among Brazilian states. For this reason, the knowledge of how the location of ethanol production affects consumer s behavior might bring some insights about the possible impacts of the national energy policy regarding the fuel diversification. Econometric estimation of demand equations for ethanol and gasoline will be carried using a dynamic panel model. After that, binary interaction variables will be used to evaluate how the demand equations diverge between the largest and smallest ethanol producers regions in Brazil. The results show that there might be considerable gains for the ethanol and gasoline consumers in largest ethanol producer regions. Key words: sugar cane, ethanol, fuel demand. 1. Introduction The Brazilian fuel market is considerably different from other markets due to diversity of alternative fuels. There are four main fuels in this market: gasoline, ethanol, vehicular natural gas (GNV), and diesel. The gasoline still remains as the main fuel, but it strongly competes with two substitutes, ethanol and GNV. The features of the Brazilian automobile fleet imply that diesel is just used by large road vehicles, not competing with other fuels in the short-run. Ethanol has a historic role in the national energy policy, being an important alternative in period of high oil prices or to face environmental enforcements. On the other hand, GNV has recently been introduced in the market and already competes with ethanol and gasoline. In this diversified marked, new market rules, and technological advances in the automobile industry such as flex-fuel engines are increasing the competition between fuels. For this reason, consumers are becoming more prices sensitive and adjusting faster towards desired demand levels (Santos, 2009). Because of the rainfall conditions and quality of land, the sugar cane plantations are considerably concentrated in the Center South region of Brazil. For this reason, transport costs, productivity differentials and historical income concentration in the Brazilian 1 The author thanks to the comments and suggestions of Ricardo Sabbadini. The opinions and conclusions expressed in the paper are those of the author. 1

2 economy can make fuel consumption to be heterogeneous in space. The question that emerged is: how different is the consumer behavior in the largest compared to the smallest sugar cane producers regions in Brazil? To give an answer for this question, the present study is designed to make the econometric estimation of fuel demand equations for ethanol and gasoline using a dynamic panel data model. After that, binary (dummy) interaction variables will be used to evaluate differences in demand equations between the largest and smallest sugar cane producers regions. The following Section presents the main elements of the Brazilian fuel market. In the Section Three the spatial aspects of sugar cane plantations and ethanol prices differentials between the largest and smallest ethanol producers regions are presented. Section Four presents the econometric specification to estimate demand equations for ethanol and gasoline. In the Section Five the data requirements are described. The results will be presented in Section Six. Finally, in the Section Seven, some final remarks are discussed. 2. The Brazilian fuel market The dynamics of the Brazilian fuel market is under the liberalization process started in 1997, the fuel diversification, and technological advances in the automobile industry such as the introduction of the flex-fuel engines. Through the known Law of Petroleum (Law 9.478/97) and new designs for the energy policy, the competition was introduced in the fuel market through the free prices and free entry to new agents. The impact of this policy on the market with considerable diversity of substitute and complementary fuels has dramatically changed the structure of fuel demand in Brazil. According to Santos (2009), the recent dynamic of the Brazilian fuel market is centered on the ethanol demand, mainly after the introduction of the flex-fuel engine technology. The knowledge of how the location of sugar cane plantations affects the fuel demand might provide important insights about the future directions of energy policy. In order to comprehend the role of ethanol in the Brazilian market fuel it is necessary to understand the cycle from the introduction of ethanol to the spatial distribution sugar cane plantations nowadays. Ethanol was introduced in the Brazilian energy matrix in 1975 through the Pro Alcool or National Program of Alcohol. This was a large scale energy production program designed to substitute vehicular fossil fuel by ethanol produced from sugarcane1. The reasons to create the program were the first and second oil shocks of 1974 and 1979 respectively. Likewise, the decision to produce ethanol from sugarcane was driven by the low sugar prices in the international commodity market at that time. Since the beginning of Pro Alcool, the ethanol fuel was massively introduced as a complementary and substitute fuel in Brazil. Initially, ethanol was mixed into gasoline given its features of complementary good. Nowadays, this mixing still remains at the rate of 25% of ethanol in the gasoline (National Agency of Oil and Biofuel (ANP), 2007). On the other hand, the Brazilian automobile industry has started to produce vehicles with engines burning just ethanol. For this reason, it also figures as a substitute fuel to gasoline. 2

3 Jul-01 Oct-01 Jan-02 Apr-02 Jul-02 Oct-02 Jan-03 Apr-03 Jul-03 Oct-03 Jan-04 Apr-04 Jul-04 Oct-04 Jan-05 Apr-05 Jul-05 Oct-05 Jan-06 Apr-06 Jul-06 Oct-06 Jan-07 Apr-07 Jul-07 Oct-07 The program worked until the first half of the 1980 s. After that, a set of factors such as the decline of oil prices (and consequently of gasoline) along with the increases in sugar prices in the international market, led to the end of the program. In addition, the engagement of Brazil in liberal policies which enforced the elimination of subsides made the program unfeasible. Aside from that, the production of ethanol and vehicles with ethanol engines continued in small scale. Figure 01 Price relations in the Brazilian fuel market (Jul/2001-Dec/2007) Alcohol/Gasoline GNV/gasoline Source: ANP Brazilian National Agency of Oil and Biofuel. Regarding energy efficiency, because the calorific value of ethanol is smaller than gasoline, technical restriction on engines results in fewer kilometers/hour using ethanol instead of gasoline. As a consequence, the competitiveness of ethanol depends on its price about to 70% of the gasoline price in the Brazilian fuel market. In 1989, when the Pro Alcool collapsed, this percentage was larger than 75% (Roppa, 2005). Since that time, the price relation has rigorously been maintained (Figure 01), mainly due to government assistance such as smaller fuel tax 2. The end of Pro Alcool in 1989 did not eliminate the use of ethanol as fuel. A considerable fleet consuming ethanol still remained in the fuel market. The sales of vehicles with ethanol engines nonetheless declined from 80% of the total sales in the period to 0.7% in Data from the National Department of Transport (2008) indicate that in 2006 the national fleet was composed of 45.6 million vehicles. About 22.0 million were light vehicles; and from this amount 3.0 millions were pumped just with ethanol, which represented 13.6% of the light vehicles fleet. This stock of vehicles pumped with ethanol was important to continuity and new developments in the technology of large scale ethanol production. In the meantime, the GNV was introduced in the fuel market. The initial strategy was to use it as a substitute to diesel in large road vehicles. However, given the logistic problems the usage was restricted to large urban centers. In 1994, subsidies of 25% in Sao Paulo and 75% in Rio de Janeiro over Tax on Property of Automotive Vehicles for vehicles used in urban transport allowed the creation of the first demand pool of GNV in Brazil. This 2 To subsidize ethanol and GNV, the tax named Contribution of Economic Domain (CIDE) have not been charged on these two fuels. 3

4 Jul-01 Oct-01 Jan-02 Apr-02 Jul-02 Oct-02 Jan-03 Apr-03 Jul-03 Oct-03 Jan-04 Apr-04 Jul-04 Oct-04 Jan-05 Apr-05 Jul-05 Oct-05 Jan-06 Apr-06 Jul-06 Oct-06 Jan-07 Apr-07 Jul-07 Oct-07 fuel started to compete with gasoline and ethanol in other regions of the country. In 2007, according to Brazilian National Institute of Oil and Gas (2008), 1.5 million vehicles ran using GVN 3. The low price of GNV regarding the gasoline or ethanol prices (Figure 01) and environmental concerns consolidated the GNV in the fuel market (Roppa, 2005) (See also Figure 05 in the Appendix A). The GNV increased the diversification in the fuel market 4 for light vehicles. The Figure 02 shows the dynamics of this market after As can be seen, gasoline still remains as the main fuel, but the market share declined from 84.4% in 2001 to 64.2% in the end of In the same period, the consumption of ethanol increased from 13.2% to 29.3% and GNV from 2.5% to 6.5%. As a matter of the fact, the quantity of gasoline demanded in the end of 2007 was almost the same that 2001 (Figure 05 in the Appendix A). This might imply in consumers closer to desired demand levels. Figure 02 Market share of gasoline, ethanol and GNV in Brazil (Jul/2001-Dec/2007) 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% Alcohol Gasoline Vehicular Natural Gas Source: ANP National Agency of Oil and Biofuel. The recent increasing demand by ethanol in Brazil results from a revolution in the automobile industry known for flex-fuel engines 5. This technology allows pumping the same vehicle with gasoline, ethanol or GNV. It was developed in the United States in the 1980 s and has being tried in Brazil since the 1990 s. In the end of 2003, it was introduced in the market through the bi-fuel vehicles, which were pumped with ethanol or gasoline. In 2005, tri-fuel vehicles, which beyond gasoline and ethanol also might be pumped with GNV, were introduced in smaller scale in the market. In fact, this might be implying a radical structural change in the fuel market. 3 The Brazilian automobile industry does not produce vehicles with GNV engine. A conversion process is necessary. As part of this process, a second tank is introduced in the vehicle. Because the original tank is maintained, consumers may pump with a second fuel; gasoline or ethanol, depending of the original engine. 4 It is important to point out that the strategy to substitute gasoline by ethanol and GNV and the fact that Brazil produces 83.2% of electricity from hydro resources, have maintained 47,5% of energy matrix composed by renewable resources. This also has reduced the energy dependence and vulnerability to the oil peak prices. 5 This technology is embedded in the Poli-fuel engines trend of the world automobile industry. 4

5 Jan-99 May-99 Sep-99 Jan-00 May-00 Sep-00 Jan-01 May-01 Sep-01 Jan-02 May-02 Sep-02 Jan-03 May-03 Sep-03 Jan-04 May-04 Sep-04 Jan-05 May-05 Sep-05 Jan-06 May-06 Sep-06 Jan-07 May-07 Sep-07 Figure 03 Retail sales (quantity of vehicles) in Brazil by type of fuel (Jan/99 Dec/07) 100.0% 90.0% 80.0% 70.0% 60.0% 50.0% 40.0% 30.0% 20.0% 10.0% 0.0% Ethanol+flex Diesel Gasoline Source: Anfavea Brazil's National Association of Vehicle Manufacturers As a first consequence of the new technology, the production of vehicles with ethanol engine has been gradually substituted by bi-fuel and tri-fuel vehicles. As can be seem in the Figure 03, in 2005 the sales of vehicles with ethanol plus flex-fuel engines overcame those of with gasoline engines. In 2007, the sales of light vehicles amounted 2.48 million. From this amount, flex-fuel vehicles represented 73.8% (1.83 million vehicles), ethanol 8.1%, gasoline 9.8% and diesel 8.5%. The changing in the trend of production of vehicles in Brazil, and consequently in the composition of the fleet is changing the consumer s sensitivity to price. Santos (2009) estimated demand equations for the ethanol, gasoline and GNV considering all the elements above in a dynamic panel data. The results showed that the dynamics of the Brazilian fuel market is centered on the changes in the ethanol market instead of gasoline. The GNV has no significant contribution for this dynamics yet. On the other hand, the results showed that ethanol might represent a complementary good of gasoline and its consumers were more price sensitive and distant from the desire demand level. This result suggests that this fuel is considerably important to energy policy issues. Furthermore, ethanol consumers became more price sensitive after the introduction of flex-fuel engines. Finally, the results also suggested that in the recent years had occurred a shift of the dynamics from the traditional gasoline demand to the ethanol demand in the Brazilian fuel market. In order to improve that results, in this paper some features of spatial distribution of sugar cane plantations in Brazil, jointly with ethanol prices differentials, will be inserted in the ethanol and gasoline demands. These elements will be described in the nest section. Spatial distribution of sugar cane plantations The sugar cane plantations area in Brazil was of 7.09 million hectares of land in the harvest 2007/2008. In this area, it was produced million tons of sugar cane. This sugar cane resulted in the production of million barrels of ethanol and 30.7 million tons of sugar. The ethanol production might be spitted in 36.4% of ethanol anidre usually used to be mixed to gasoline, and 63.6% of ethanol hydrated that is used as substituted fuel to gasoline. A part from the considerable amount of sugar cane produced in Brazil, 5

6 Acre Amapá Rondônia Roraima Amazonas Pará Tocantins Maranhão Piauí Ceará Rio Grande do Norte Paraíba Pernambuco Alagoas Sergipe Bahia Minas Gerais Espírito Santo Rio de Janeiro São Paulo Paraná Santa Caratina Rio Grande do Sul Mato Grosso Mato Grosso do Sul Goiás Distrito Federal some factors, such as land fertility and rainfall level, makes the production extremely concentrated in the national territory. Figure 04: Sugar cane and ethanol production in the Brazilian states (% share) in % 60.0% 50.0% 40.0% 30.0% 20.0% 10.0% 0.0% North-Northeast States Center-South States Ethanol Sugar Cane Production Source: Brazilian Institute of Geography and Statistics Figure 04 shows the distribution of sugar cane production in Brazil in the harvest 2007/2008. The production is concentrated in Center-South states, which produced 87.4% (431.1 million tons) of the national production in this period. The São Paulo state is the largest producer. This state produces 60.1% (296.3 million tons) of the national production. The North-Northeast regions produce 12.6% of the national production and this production is concentrated in the sates of Pernambuco and Alagoas, which jointly produce 9.5% of the national production. Figure 05: Sugar cane production clusters in Brazil in 2007 Source: Brazilian Institute of Geography and Statistics The spatial aspects of sugar cane production were evaluated through the Moran Index and Local Indicator of Spatial Association (Lisa), using data for the 558 Brazilian microregions. The significant Moran Index of 0.61 indicated the presence of spatial autocorrelation in the sugar cane production. The Lisa indicator is plotted in the Figure 05. As can be seen, there are two important clusters of sugar cane production in the Brazilian economy. The largest is located in the Center-South region, around the State of 6

7 Jul-01 Sep-01 Nov-01 Jan-02 Mar-02 May-02 Jul-02 Sep-02 Nov-02 Jan-03 Mar-03 May-03 Jul-03 Sep-03 Nov-03 Jan-04 Mar-04 May-04 Jul-04 Sep-04 Nov-04 Jan-05 Mar-05 May-05 Jul-05 Sep-05 Nov-05 Jan-06 Mar-06 May-06 Jul-06 Sep-06 Nov-06 Jan-07 Mar-07 May-07 Jul-07 Sep-07 Nov-07 São Paulo. The second is located in the Northeast region and is composed by the states of Pernambuco and Alagoas. Because of the ethanol production is directly related to the sugar cane production, the spatial concentration might bring some implications for the ethanol prices in the Brazilian territory. The Figure 06 shows Lisa indicator for the variable productivity of sugar cane (production/hectare) in the Brazilian micro-regions. As can be seen, there are three significant productivity clusters of sugar cane in Brazil, all of them in the Center-South region. Figure 06: Sugar cane productivity clusters in Brazil in 2007 Source: Brazilian Institute of Geography and Statistics From the Figures 05 and 06 it might be extracted the fact that the most producer regions also are the most productive. When this fact is associated with transport costs, and assuming that the technology to produce ethanol from sugar cane is the same in Brazil, prices differentials might emerge in the economic space. The Figure 07 shows the real prices to final consumers of the fuel ethanol in Brazil for the eight largest producers states (São Paulo, Paraná, Minas Gerais, Mato Grosso, Mato Grosso do Sul, Goiás, Paernambuco e Alagoas) and in the smallest producers (the other 19 states). 1,6 1,4 1,2 1 0,8 0,6 0,4 0,2 0 Figure 07: Real Ethanol Prices in Brazil (jul/2001-nov/2007) Source: National Agency of Oil National Prices Largest Producers Prices Smallest Producers Prices 7

8 As can be seen, the real prices in the smallest producers states are considerably above the national prices. On the other hand, the real prices in the largest producers states are slightly below the nations prices. This evidence suggests that the consumer s behavior might differ regarding prices changes in the ethanol, gasoline or GNV. To evaluate this behavior econometric estimation of demand equations for ethanol and gasoline will be done using a dynamic panel data model. 3. Econometric specification The main papers about estimation of fuel demand equations in the energy economics literature are related to gasoline demand. In the most influential papers of Energy Economics, Dahl and Sterner (1991) summarize a set of principles, models and data requirements used for estimation of gasoline demand. Although restricted to time series data, others like Ramanathan (1999), Eltonny and Al-Mutairi (1995), and Bentzen (1994) provide good insights about the subject. The literature for Brazil shows that there are estimations of price and income elasticities (Appendix B). The general results show that fuel demand in Brazil is inelastic and the fuels are imperfect substitutes. Although interesting for international comparisons, many parameters also are not statistically significant. However, a short description of conclusions of these studies is necessary for comparisons at the end of the present study. Burnquist and Bacchi (2002) estimated the demand equations for gasoline in Brazil using year time series. The main conclusion was that consumption is more sensitive to income than price in the short and in the long-run. Alves and Bueno (2003) also estimated the demand equation for gasoline using year time series and found that ethanol is an imperfect substitute for gasoline even in the long-run. Roppa (2005) compared the competitiveness of gasoline against ethanol also using year time series year data. The results showed that ethanol is an imperfect substitute for gasoline and that the gasoline consumers are indifferent to price increases in the short and in the long-run. Iooty (2004) compared the competitiveness between gasoline and GNV using monthly time series and found that GNV is an imperfect substitute for gasoline. In spite of that, demand for GNV showed to be price inelastic. In the short-run the consumers were more price than income sensitive and in the long-run this relation showed to be inverted. The previous studies present some bottlenecks for which the present study is designed to overcome. Those studies do not consider both the substitution among the three fuels and the use different econometric tools other than time series. Due to heterogeneity of the Brazilian economy the estimations from aggregated time series at the country level might affect the results. For these reasons, those parameters in Appendix B might be biased or not reflective of the actual consumer behavior. 8

9 The first study about specification of energy demand using panel data was developed by Balestra and Nerlove (1966). The authors estimated the demand for natural gas in the United states. Baltagi and Griffin (1983) estimated the demand for gasoline using panel data for OECD countries. Since that time, panel data models have been used increasingly in demand studies. Their theoretical principles to specification might be found in Hsaio (1985 and 1986), Baltagi (2001) and Wooldridge (2004). In the present study, the specification considers that the demand is technologically dependent on the vehicles engine. It will be considered first a market without flex-fuel engines due to large previous stock of non-flexible engines. The basic assumption is that consumers have some difficulty in changing to another fuel in the short-run and in the long-run. Therefore, previews consumption determines the pattern of consumption in the present. This modeling allows the estimation of the speed of adjustment of the consumers regarding the desired demand levels. The basic model is the partial adjustment model (Pesaran, 1997). Assuming a quantity of good (y it ), its real price (p gt ), the real price of substitute good (p st ), the real per capita income I it and the quantity of the same good in the previous last time y it-1, is possible to obtain the formulation known for the lagged endogenous model: y f p, p, I, y ) (1) it ( gt st it it 1 This model is easy to estimate, to interpret and does not over-demand data requirements, see Dahl and Sterner (1991). The lagged dependent variable is a shortcut for inertia of economic behavior, deduced from assumptions about partial adjustments or adaptive expectations. There are two main theoretical interpretations about this assumption in the model. The first is that, for psychological, technological or institutional reasons, consumers do not change their habits immediately after changes in prices or income, because this may result in some level of disutility. The second interpretation considers a delay in the consumer reaction because of consumption is not only a function of the present income and price structures, but also of the past ones. Although this kind of specification seems to be from ad hoc models 6, the adequacy to theoretical foundations of consumer theory is the main stimulus to use it. See Liu (2004). Considering assumptions of partial adjustment and an autoregressive distributed lag (ADL) model in the form p q y it i i yk t i j k j x 1, 1, k, t j u t (2) In equation (2), γ i with i=1,.,p are the coefficients of (autoregressive) lagged values of the dependent variable y it, x k,t-j with j=0,1,.,q are the k-elements column vectors of the 6 For a discussion about different dynamic models of energy demand and the relationship among them, see Watkins (1991). And for an example which explicitly considers dynamic optimization over time, see Pindyck and Rotemberg (1983). 9

10 current and distributed lagged values of explanatory variables, and β k,j is a column vector with k coefficients, α is the constant term and u t the white noise error. The restrictions on lag length p and q, which refers to distribution of the coefficients γ i and β k,j, determine different ADL(p,q) models. In the present work will be used the ADL(1,0). Assume the equation * yt 0 1Pgt 2Pst 3 I t t 2 t ~ IID (0, ) (3) * yt is the unobserved desired level of demand in the period t, P gt, P st and I t and are the same of equation (1) and the α n, n=0,..,3 are the parameters of the model. The * relationship between yt and the actual demand y t is characterized by partial adjustment in consumers behavior. This adjustment is given by * y y ( y y 1) (4) t t 1 t t This specification states that the change in the consumption during one period of time ( y t y t 1) is a share of the difference between the desired demand in the current period and the actual demand in the previous period. The coefficient θ reflects the speed of adjustment toward the desired demand level and ranges as (0,1]. In other words, the larger θ is, the faster the adjustment. For instance, if θ=1, the actual demand immediately approaches to desired level. To incorporate the speed of adjustment into estimation process the y * t from (4) might be inserted into equation (3). The rearrangement considering the specification for panel data model provides y it 0 yi( t 1) 1Pg 2Ps 3I( it) u( it). (5) In equation (5), i=1,.,n is panel unit, t=1,.,t is still time period. The new parameters are: β j =θα j, j=0,.,3 and γ=(1-θ) and. The endogenous error term now is u it =θε it. The short-run price partial effects (or other interim periods until the long-run) are obtained by y t+k / p t =γ k β j, k=0,.,p. Likewise, a long-run price effect may be expressed as P k k ) /(. At last, the linearization of the variables in equation allows the elasticity of short-run and long-run related to relevant variable to be obtained. In terms of the econometric estimation of the model above, the presence of the lagged dependent variable improves considerably the statistics and fit of models, mainly when there is no variable representing to variation in vehicle stock. In the panel specification in (5), it is assumed that an unobserved time invariant 7 regionspecific fixed effect μ i exists and is not included in the regression, mainly because of free 7 The argument to be time invariant accounts mainly because of short-run period considered. 10

11 market rules with no price control. In the same way, different regional income per capita levels might influence the choice of the fuel and the sensitivity of the parameters. To include the factors above in the model, take μ i being a fixed parameter of a random variable and the error component expressed as 2 2 u, being i ~ IID(0, ) and ( it ) ~ IID (0, ). (6) i The term η it now represents the reminder error term. The new statistic assumptions are: E, ) 0; E, ) 0; E I, ) 0; (7) ( P g i ( P s i ( i E, ) 0; E, ) 0; E I, ) 0; (8) ( P g ( P s ( The terms in (7) express the correlation between the unobserved time invariant regionspecific effect and all the explanatory variables of the model. Those in (8) state the strict exogeneity of the same variables regarding the new error term. Finally, since y (it) is in function μ i and y (it-1) also is in function of μ i, an endogeneity problem must be considered in the estimation, even if η it is not serially correlated, see Nickell (1981). For the above reason, OLS estimation would be biased and inconsistent. The estimation using SUR would not account for μ i. The Fixed Effect estimator would eliminate μ i, but does not account for endogeneity problems. In this case, is suggested the GMM estimator from Arellano and Bond (1991), see Baltagi (2001) and Wooldridge (2004). Also known for dynamic panel data estimator, it is based on Anderson and Hsaio (1981 and 1982) who suggest the use of differences of lagged dependent variables as instruments. It may generate consistent parameters when the panel or time units tend to infinity. At the same time, it also eliminates the unobserved fixed effect. 4. Data requirements The two demand equations to be estimated are: lng GDP u (9) 0 lng( it 1) 1 ln PG 2 ln PE 3 ln PGNV 4 ln lne GDP u (10) 0 lne( it 1) 1lnPG 2 lnpe 3 lnpgnv 4 ln The two dependent variables will be per capita consumption of gasoline (G), ethanol (E) and GNV. The other explanatory variables are: real prices of gasoline (P G ), ethanol (P E ) and GNV (P GNV ) and per capita Gross Domestic Product (GDP). 11

12 A panel monthly dataset regarding 27 Brazilian states for the period from Jul/2001 to Dec/2007 was built. This was the period in which the ANP has surveyed data about fuel prices and consumption. After eliminating regions with no data about GNV, the sample was reduced to 15 states. The monthly price index and population data were obtained from the Brazilian Institute of Geography and Statistics (IBGE). As there is no monthly data of GDP by state, a proxy variable had to be used. The variable used was the Tax on Trade of Products and Services (ICMS) obtained from the Brazilian National Treasury. To avoid the endogeneity problem the fuel tax was excluded from ICMS. Although not introduced in the model, other variables which might be used as control variables were considered. The vehicle stock might be a good variable to be introduced, but there is no considerable monthly dataset organized states and kind of fuel. Another variable that might be used is the monthly oil prices. However, this variable showed weak correlation with other variables of the model. The main explanation for this weak correlation was that, despite new market rules, international oil price variations are still not thrown directly over the fuel distribution sectors by the Brazilian National Oil Company (Petrobras), which control almost all the refineries of the country. For all the considerations, the lagged dependent variable minimized the hypothesis of problems. 5. Estimation, results and comparisons In order to evaluate the sensitivity of consumers regarding the location of ethanol production, was carried the estimation of demand equations (9) and (10), in logarithm, to obtain adjustment coefficients, price, cross-price and income elasticities. After the first estimation of each equation, it was introduced a dummy variable of interaction between ethanol prices and the panel units representing the eight largest ethanol producers states. O objective was to evaluate if the price and cross price elasticity of ethanol demand are different in the markets which also are large ethanol producers. Table 01: One-step GMM-Arellano-Bond short-run estimations of fuel demand equations for Brazil Ethanol C ε e ε e,g ε e,gnv ε e,i θ e dummy (1.69) (-9.65) (3.50) (4.17) (10.48) (31.68) (1.45) (-8.94) (3.81) (4.22) (10.92) (3.74) (2.15) Gasoline C ε g ε g,e ε g,gnv ε g,i θ g dummy (4.20) (-3.34) (-1.90) (1.30 ) ) (8.86) (3.96) (3.96) (-3.01) (-2.86) (1.34) (8.87) (3.95) (2.16) The values in brackets refer to z statistics. 12

13 Table 01 presents the one-step GMM estimations for the elasticities ε (.), the cross-price elasticities ε (.,.), and adjustment coefficients θ (.) regarding the demand equations by ethanol and gasoline. For each equation the Sargan test suggested to reject the hypothesis of over-identification restriction for one step estimation. The Arellano-Bond test also suggested rejecting the hypothesis of no serial autocorrelation in residuals of order 1 but not to reject the hypothesis of no serial autocorrelation of order 2. This makes the dependent variables valid instrumental variables with at least two legs. Because the monthly data comprises a period of six years, the equations were estimated to obtain short-run estimations. In general, the results in Table 01 show that all price elasticities were smaller than one, which is common for energy goods, marked by inelasticity of demand. In addition, all cross-price elasticities also were smaller than one, which characterize the three fuels as imperfect substitutes, for each other. Other results are described below. Ethanol The results of the first ethanol demand equation were supported by economic theory. Due to lacks of no previous estimation in the international literature, no comparison was possible. The negative and high price elasticity of might be an evidence that ethanol consumers are from lower income groups more price sensitive and being attracted by cheaper fuel. On the other hand, this result might have reflected the introduction of flex-fuel engines in the market 8. The cross-price elasticities regarding gasoline and GNV had and expected positive sign. The high cross-price elasticity regarding gasoline price of implies that ethanol competes considerably with gasoline. It also suggests that the gasoline prices are more important for ethanol demand than gasoline demand. Similarly, the cross-price elasticity regarding GNV of also implies that ethanol already competes with GNV. The income elasticity of 0.404, considerably greater than that from gasoline demand, also is an evidence that ethanol consumers belong to lower income groups. At last, the adjustment coefficient of suggests that ethanol consumers are distant from desired demand level, i.e., the ethanol market has a strong potential to rapidly increase in Brazil. In the second ethanol demand equation the dummy interaction shows that there is an interaction statistically significant between the eight largest ethanol producer states and the ethanol prices of This positive interaction makes that the price elasticity of ethanol demand of the consumers belonging to the largest ethanol producer states decrease to , which is the sum of plus This elasticity is considerably smaller than the price elasticity of of the smallest producer regions. This shows that in the largest ethanol producer regions the consumers are less sensitive to ethanol prices. This might promote gains for the producers and consumers in these regions. On 8 For analysis of the impacts of flex-fuel engine on fuel demand equations in Brazil, see Santos (2009). 13

14 the other hand, it also might be evidence that transport cost contribute negatively to the ethanol demand in the smallest producers regions than in the largest producers and the productivity differentials contribute positively to largest producers regions. Gasoline For the results of the first gasoline demand equation, the price-elasticity of had the expected sign and is close to Burnquist and Bacchi (2002) estimation of , see Appendix B. This result also is considerably close to that of Dahl and Sterner (1991). The cross-price elasticity of ethanol of had an unexpected sign and was not close to others estimations for Brazil. However, the persistence of this negative sign also is verified in others studies such as Alves and Bueno (2003) and Roppa (2005). This suggests the hypothesis that ethanol might be considered a complementary good to gasoline. Because of the fact that gasoline is composed of 25% of ethanol and also due the high statistical significance of the parameter, this hypothesis might be confirmed. On the other hand, the cross-price elasticity regarding GNV of had expected sign but not so statistically significant; there is no estimative in the literature to be compared. The income elasticity of was close to Roppa (2005) and Iotty (2004). This small value might be explained by the fact that gasoline consumers are from higher income groups in Brazil. The adjustment coefficient of is close to one and indicates that in the short-run the demand for gasoline is close to desire demand level. This means that the main explanatory variable of gasoline demand variation might be only the large (and older) stock of vehicles with gasoline engines. In the second equation, the dummy interaction between ethanol prices the eight largest ethanol producer states also was statistically significant of This positive interaction makes that cross-price elasticity of gasoline demand regarding the ethanol prices decrease to , which is the sum of plus This cross-price elasticity is smaller than the cross-price elasticity of of the smallest producer regions. This means that the gasoline consumers located in the largest ethanol producer states are less sensitive to ethanol prices, relatively to other regions. In the same way that for ethanol consumers, this might be an evidence of possible relative gains for producers and consumers derived from differentials of transport cost e productivity of ethanol. 6. Final Remarks This paper proposed to evaluate the regional differences in the fuel consumer s behavior induced by regional production of ethanol in Brazil. Our hypothesis was that the regional differences in the productivity of ethanol and transport costs affected competitiveness of the fuels among Brazilian states. Econometric estimation of demand equations for ethanol 14

15 and gasoline were developed using a dynamic panel model and binary interaction variables to evaluate how the demand equations diverge between the largest and smallest ethanol producers states in Brazil. The results showed that ethanol and gasoline are imperfect substitutes and complementary goods. The adjustment coefficient also showed that ethanol demand is more distant from the desired demand level than gasoline. Regarding regional differences in the consumer behavior, in the largest ethanol producers states the ethanol consumers are less sensitive to ethanol price variations. Due the fact that ethanol might be considered a complementary good to gasoline, the gasoline consumers also is less sensitive to ethanol price variations in the largest ethanol producers regions. For this reason, there might be considerable gains for the ethanol and gasoline consumers in this regions explained by the production systems and location of sugar cane plantations. Finally, despite of the fact that there is some difficulty to organize a dataset for the Brazilian micro-regions, advances might be achieved handling the regional differentials in the gasoline prices and/or using spatial panel data econometrics. 8. References Anderson, T. W. and Hsaio (1982). Formulation and estimation of dynamic models using panel data. Journal of Econometrics, 18(1): (1981). Estimation of dynamic models with error components. Journal of the American Statistical Association 76(375): Arellano, M. and Bond, S. (1991). Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Review of Economic Studies, 58(2): Balestra, P. and Nerlove, M. (1996). Pooling cross-section and time-series data in the estimation of a dynamic model: The demand for natural gas. Econometrica, 34(3): Baltagi, B. H (2001). Econometric Analysis of Panel Data. Chichester: John Wiley and Sons, Ltd. Baltagi, B. H. and Griffin, J. M. (1983). Gasoline demand in the OECD: An application of pooling and testing procedures. European Economic Review, 22(2): Bentzen, J. (1994). An empirical analysis of gasoline demand in Denmark using cointegration techniques. Energy Economics, 16(2):

16 Brazilian Institute of Geography and Statistic (IBGE). Índices de preços e estimativas populacionais. Brazilian National Treasury. Arrecadação mensal dos estados, Bueno, R. and Silveira, A. Short-run, Long-run and Cross Elasticities of Gasoline Demand in Brazil. Energy Economics, 25(2): Burnquist H. L. and Bacchi, M. R. P. (2002). A demanda por gasolina no Brasil: uma análise utilizando técnicas de co-integração. CEPEA, Discussion paper, Dahl, C. and Sterner, T. (1991). Analyzing gasoline demand elasticities: a survey. Energy Economics, 13 (3): Eltony, M. N. and Al-Mutairi, N. H. (1995). Demand for gasoline in Kuwait: an empirical analysis using co-integration techniques. Energy Economics, 17(3): Hisao, C. (1985). Benefits and limitations of panel data. Econometric Reviews, 4(1): (1986). Analysis of Panel Data. Cambridge University Press: Cambridge. IOOTTY, M. et al. (2004). Uma análise da competitividade preço do GNV frente à gasolina: estimação das elasticidades da demanda por GNV no Brasil no período recente. In Rio Oil and Gas Expo and Conference, edited by IE/UFRJ. UFRJ: Rio de Janeiro. Liu, G. (2004). Estimating Energy Demand Elasticities for OECD Countries: a dynamic panel data approach. Statistics Norway, Discussion paper n National Agency of Oil and Biofuel (ANP) (2008). Levantamento mensal de preços e vendas de combustíveis no Brasil, National Department of Transit (DENATRAN). Frota nacional de veiculos automotres. Nickell, S. (1981). Biases in dynamic models with fixed effects. Econometrica, 49(6): Pesaran, M. H. and Smith, R. (1995). Estimating long-run relationships from dynamic heterogeneous panels. Journal of Econometrics, 68(1): Pindyck, R. and Rotemberg, J. 1982). Dynamic Factor Demands Under Rational Expectations. NBER Working Paper Series, 1015, pp RAMANATHAN, R. (1999). Short and long-run elasticities of gasoline demand in India: an empirical analysis using co-integration techniques. Energy Economics, 21(4):

17 Jul-01 Oct-01 Jan-02 Apr-02 Jul-02 Oct-02 Jan-03 Apr-03 Jul-03 Oct-03 Jan-04 Apr-04 Jul-04 Oct-04 Jan-05 Apr-05 Jul-05 Oct-05 Jan-06 Apr-06 Jul-06 Oct-06 Jan-07 Apr-07 Jul-07 Oct-07 Roppa, B. F. (2005). Evolução do consumo de gasolina no Brasil e suas elasticidades: 1973 a UFRJ: Rio de Janeiro. Santos, G. F. (2009). Fuel Demand in Brazil in a Dynamic Panel Data Approach (Mimeo). Watkins, G. C. (1991). Short-and long-term equilibria: relationships between first and third generation dynamic factor demand models. Energy Economics, 13(1): 2-9. Wooldridge, J. (2004). Econometric Analysis of Cross-Section and Panel Data. USA: MIT Press. Appendix A Figure 05: Consumes of gasoline, ethanol and GNV in Brazil (in barrels) 16,000, ,000, ,000, ,000, ,000, ,000, ,000, ,000, Ethanol Gasoline GNV Source: ANP Brazilian National Agency of Oil. Appendix B Table 02 Literature estimations of price, cross-price and income elasticities for Brazil Short- run Long-run Data and period Demand for Gasoline ε g ε g,e ε,gnv ε g,i ε g ε g,e ε,gnv ε g,i Burnquist and Bacchi (2002) Alves and Bueno (2003) -0,092 (NS*) Roppa (2005) -0,073 (NS) -0,230 (NS) -0,198 (NS) - 0,122-0,464-0,472-0, Y-TS ( ) 0,480-0,122 Y-TS ( ) (NS) 0,401-0,163 Y-TS ( ) Demand for GNV ε gnv ε gnv,g ε gnv,e ε gnv,i ε gnv ε gnv,g ε gnv,e ε gnv,i Iotty (2004)* ,181 (NS) (NS) Notes: Y-TS: Year Time Series; M-TS: Monthly Time Series; NS: when the level significance was smaller 90%. M-TS (Jul/2001 Dec/2003) 17

18 * In this study, the long-run estimations do not have sense due short period of the sample. 18