IBP1670_ DIFFUSION OF FLEXFUEL CARS IN BRAZIL: PRICE RESPONSE AND IMPACTS ON CO2 EMISSIONS

Luciano Losekann, Thaís Vilela and Gustavo Rabello de Castro

Copyright 2012, Brazilian Petroleum, Gas and Biofuels Institute - IBP This Technical Paper was prepared for presentation at the Rio Oi & Gas Expo and Conference 2012, held between September, 17-20, 2012, in Rio de Janeiro. This Technical Paper was selected for presentation by the Technical Committee of the event according to the information contained in the final paper submitted by the author(s). The organizers are not supposed to translate or correct the submitted papers. The material as it is presented, does not necessarily represent Brazilian Petroleum, Gas and Biofuels Institute' opinion, or that of its Members or Representatives. Authors consent to the publication of this Technical Paper in the Rio Oil & Gas Expo and Conference 2012 Proceedings.

Abstract
Introduction of flex-fuel (gasoline/ethanol) cars in Brazil in 2003 has modified the dynamics of the automotive fuel market. Nowadays, 92% of car sales in Brazil correspond to flex-fuel models and they already represent 40% of total fleet. Owners of these vehicles can choose which fuel to use according to the relative price of gasoline and ethanol that become

almost perfect substitute goods. As share of flex fuel cars in the Brazilian automotive fleet increases, ethanol and gasoline consumption become very volatile and responsive to price fluctuations in the short term. When ethanol prices are high, as in the last two years, its consumption reduces and CO2 emissions on the transport sector increase. This paper presents estimations of future fuel consumption and CO2 emissions in scenarios of high and low ethanol prices.

Keywords: Ethanol, flexfuel cars, biofuels, CO2 emissions, automotive fuel demand

UFRJ UF Keywords: Ethanol,

1.

Introduction

The Brazilian Alcohol Program (Pró-álcool) represents a unique experience on oil derivates substitution in the transport sector (Hira and Oliveira, 2009 and Coelho et al., 2006). The program was implemented in 1975 as a part of set of policies to reduce oil imports after the first oil crisis. In its first phase, ethanol (anhydrous ethanol) was used as an addictive to gasoline. After 1979, when the second oil crisis occurred, ethanol (hydrous ethanol) fuelled cars were introduced and they experimented a fast diffusion. In 1986, ethanol fuelled models represented 92% of total cars sales (figure 1). Figure 1 – Gasoline, Ethanol and Flex-fuel cars sales
3,0

2,5
milhões de unidades 2,0 1,5 1,0 0,5 0,0

Venda veículos leves a álcool

Venda veículos leves flex

Source: Elaborated from Anfavea sales data.

As oil prices reduced in the counter-crisis period and international sugar prices raised, ethanol production became less attractive. Then, sugar cane producers focued the sugar market (sugar production is substitute of ethanol) oriented a smaller part of its production to ethanol, and some shortage episodes were experimented. During the 1990s, ethanol fuelled car sales collapsed as car owners did not intended to face the uncertainty of ethanol supply. Hydrous ethanol consumption progressively decreased. However, the Pró-álcool experience gave an opportunity to Brazil to accumulate knowledge on ethanol production and on the development of ethanol fuelled cars. In 2003, when hydrous ethanol represented less than 10% of the automotive fuel market, flexible fuel (ethanol and/or gasoline) car was launched in Brazil. Flexible cars eliminated the supply uncertainty problem and allowed consumers to choose the fuel taking advantage on price differential. Flexible models soon dominated cars sales in Brazil (figure 1).

Figure 2 – Gasoline and Ethanol Consumption (thousands of oil equivalent tones)
35.000
30.000 25.000 20.000 ETHANOL 15.000 10.000 5.000 0 1970 1973 1976 1979 1982 1985 1988 1991 1994 1997 2000 2003 2006 2009 GASOLINE

Note: Consumption of anhydrous ethanol (mixed in gasoline) is included in the gasoline numbers. Ethanol data refers to hydrous ethanol consumption Source: Elaborated based on MME data

Nowadays, flex-fuel vehicles represent 92% of light vehicle sales. Brazilian car fleet is already formed by 46% flexfuel vehicles and this share keeps increasing. Therefore, a large part of car owners can chose either ethanol or gasoline in the moment they fill their car tanks. And, consumption of gasoline and ethanol became quite volatile in Brazil. Even though there are other factors influencing fuel choice, as environmental concerns, the most important is fuel prices. Due to transport costs and regional taxes, ethanol prices show significant differences in Brazil. In states close to production area, as São Paulo, Goiás and Mato Grosso, relative prices induce flexfuel car owners to choose ethanol during most part of the year. In the North, ethanol is less competitive than gasoline even during the sugarcane season. Since 2009, ethanol supply in Brazil has been tight and ethanol prices increased sharply during sugarcane off seasons. Currently, ethanol price is below 70% of the gasoline price1 only in one of the 26 Brazilian states (Goiás). As consequence, total Brazilian ethanol consumption decreased 8% in 2010 and 26% in 2011 while gasoline consumption increased 17% in 2010 and 18% in 2011. That has caused a significant increase in Brazilian CO2 emissions in the transport sector. Our objective in this study is to estimate how consumers respond to changes in fuel relative prices in Brazil. Based on different scenarios of fuel prices and economic growth we estimate Brazilian consumption of ethanol and gasoline until 2020. We will also present estimations of CO2 emissions resulting from fuel consumption. The first step to model fuel consumption is to estimate Brazilian car fleet. As official data is not accurate, we projected car fleet based on car sales evolution and an estimated scrapping function. The procedure follows the method developed by Mattos and Correia (1996). After estimating the light vehicles fleet in 2010, we project the light vehicles fleet ten years ahead using an econometric model (Losekann, Vilela and Iootty, 2010) sensitive to economic growth scenarios.
1

As energy content of ethanol is roughly 70% of the gasoline. When ethanol price is lower than that percentage, ethanol is relatively cheaper than gasoline.

We identify fleet profile by fuel type. Only the flexfuel part of the fleet is sensible to fuel prices on the short run. We analyzed how fuel consumption of flexfuel cars is related to relative prices, using monthly observations from January 2006 to December 2010 for each Brazilian state. Considering only fuel consumption by flexfuel cars (i.e. disregarding consumption of only gasoline and only ethanol cars), the ratio of effective of ethanol consumption and potential ethanol consumption (when all flexfuel fleet chose ethanol), CE/CP, is determined by relative price (PE/PG). We estimated that relationship by means of a logistic function:

Using scenarios for relative fuel price evolution, we can estimate Brazilian ethanol and gasoline consumption. Finally, we multiply ethanol and gasoline consumption by their relative CO2 emissions coefficients to estimate CO2 emissions resulting from each scenario of fuel relative price and economic growth. We projected a substantial increase of Brazilian flexfuel car fleet until 2020. In an optimistic scenario for economic growth (GDP increases by 4.8% per year), the flexfuel fleet would represent 81% of total car fleet, which would reach 51 million vehicles (it was 27 million in 2010). Fuel demand is expected to become more responsive to relative price changes in consequence of increasing of flexfuel car fleet. According our estimates, if relative price of ethanol follows its historical trend, ethanol consumption would increase significantly and account for 65% of total car fuel consumption. However, if high ethanol prices persist in the next years, consumption would be lower and CO2 emissions would be higher. Brazil has now a long experience with ethanol use as an automotive fuel. Introduction of flexfuel cars gave a new momentum to the ethanol market. It contributed to a low level of CO2 emissions in the Brazilian transport sector. However, in the last two years, ethanol consumption decreased due to short supply and high prices. Our analysis reveal that fuel demand will become more responsive to fuel prices and higher ethanol prices may undermine Brazilian environmental objectives regarding emissions of the transport sector.

2.
2.1.

Fleet Estimation
Methodology and Results

The approach used to estimate light vehicles fleet is based on Mattos and Correia (1996). Data from the National Household Research of 1988, Pesquisa Nacional de Amostras por Domicílio (PNAD), from the Brazilian Institute of Geography and Statistics (IBGE), is used to estimate the scrapping curve. Although this research is long dated, it is the only available data about light vehicles ownership by household. From this database, it is possible to identify the age profile of the fleet, which is essential to estimate a scraping rate. It is worth mentioning that PNAD is a sample from Brazil’s population. In the PNAD 1988, it was considered 83,902 households from a universe of 33 millions. Therefore, the sample correspond, approximately, to 0.2% of total population. Having determined the total fleet by age, the next step is to gather information about light vehicles sale in Brazil from Anfavea. The survival rate in the first year is given by the division of the number of cars sold in the previous year that are still running and

cars sales in the previous year. In the second year, it corresponds to the rate of cars sold two years ago that are still running. So survival rate can be defined for each year. The scraping rate is defined as being 1 minus the survival rate. The series of data for each year constitute the scraping curve. S-shaped curves are the most used to describe the scraping rate in the Brazilian literature, mainly the Logistic and Gompertz curves. On the other hand, in the international literature, the Weibull function is the most used. Aiming assess which one of these will better suit the database, these three functions are tested. In order to estimate the parameters of these functions, the optimization method applied by Mattos and Correia (1996) is used. The better function will be the one which gives the smaller sum of squared errors, being defined as the difference between the scraping rate estimated from the PNAD 1988 and the scraping rate estimated from the scraping curve. Table 1 – Scraping Curves Optimization Results Parameter a Logistic function Gompertz function Weibull function 22.073 1.7406 17.734 Parameter b 0.204 -0.144 2.0371 Sum of errors 0.1821 0.1767 0.1867 squared

From the results shown in Table 1, the Gompertz function presents the smaller error and was selected. The results obtained, a part from the Weibull function, are similar to those obtained in the literature.
Therefore, the Gompertz scraping curve is applied on cars sale database. The results are presented by type (light commercial and automobile) and fuel (gasoline, flex-fuel, diesel, ethanol and NG). To consider fuel conversion to natural gas, that is relevant in Brazil, we used conversion data provided by GasNet and assumed some hypothesis: i) diesel and ethanol vehicles do not convert for technical issues; ii) incidence of natural gas conversions in light commercials is the double of automobiles; and iii) as converted cars cover longer distances as to payback conversion costs, its scraping rate is higher (double). Converted car fleet and the profile of 2009 car fleet by vehicle type and fuel are presented in table 3 and 4. Table 3 – NGV Converted Fleet by Type and Fuel in 2009 (thousand vehicles)
Gasoline => NGV 568 141 Flex-Fuel => NGV 517 122

Cars Light Commercials

Table 4 – Fleet by Type and Fuel in 2009 (thousand vehicles)
Gasoline 11,126 1,542 Flex-Fuel 8,025 911 Diesel 1 1,144 Ethanol 1,190 170 NGV 1,085 264

Cars Light Commercials

Thereby, the total light vehicles fleet in Brazil in 2009 is equal to 25.5 million vehicles. The result is similar to the estimations of Anfavea, 25.6 million in 2007, and Sindipeças, 26.2 million in 2008, but quite different from Denatran, 36.5 million in 2009. Therefore, the Gompertz scraping curve is applied on cars sale database. The results are presented by type (light commercial and automobile) and fuel (gasoline, flex-fuel, diesel, ethanol and NG). To consider fuel conversion to natural gas, that is relevant in Brazil, we used conversion data provided by GasNet and assumed some hypothesis: i) diesel and ethanol vehicles do not convert for technical issues; ii) incidence of natural gas conversions in light commercials is the double of automobiles; and iii) as converted cars cover longer distances as to payback conversion costs, its scraping rate is higher (double). Converted car fleet and the profile of 2009 car fleet by vehicle type and fuel are presented in table 3 and 4. Table 3 – NGV Converted Fleet by Type and Fuel in 2009 (thousand vehicles)
Gasoline => NGV 568 141 Flex-Fuel => NGV 517 122

Cars Light Commercials

Table 4 – Fleet by Type and Fuel in 2009 (thousand vehicles)
Gasoline 11,126 1,542 Flex-Fuel 8,025 911 Diesel 1 1,144 Ethanol 1,190 170 NGV 1,085 264

Cars Light Commercials

Thereby, the total light vehicles fleet in Brazil in 2009 is equal to 25.5 million vehicles. The result is similar to the estimations of Anfavea, 25.6 million in 2007, and Sindipeças, 26.2 million in 2008, but quite different from Denatran, 36.5 million in 2009.

2.2 Modeling car demand This section shows the estimatimation and forecast of the light vehicles fleet ten years from now. Car sales are estimated using an vector error-correction model It is assumed that cars sales are a function of the income and the credit volume. The data ranges from 1993 to 2009. CS=f(Y,CV) Where CS is the car sales; Y is the percapita income and CV is the credit volume given to individuals. We tried to include car price in the model. However, no cointegration relation was obtained by the Johansen test. Therefore, the price was not considered. Other works include the relative price of vehicles given by the ratio between the wholesale index price of industrial products, including machine, equipments and vehicles, from IBGE

and a general price. There are two problems concerning the use of the relative prices: (a) the index composition; and (b) the fact that this index is an average of the price index of these several components. Therefore, introducing this variable would not give the right demand reaction that we would like to, besides even the interpretation of the coefficient would be difficult. So, even though we recognize the importance of price to define demand, we believe that there is still no good price estimative available. Thereby, we conduct the econometric analysis using Equation1. After testing the integration order of the variables through the correlogram and the unit root test, Augmented Dickey-Fuller, it was found that all variables were stationary in the first difference. The number of lags to be included in the model, accordingly to several selection order criteria, is one. The Johansen test for cointegration suggests one cointegration vector. Besides, the model with a restrict constant is better considering all the tests. Therefore, the vector error-correction model is estimated. The long-run relationship is given by Equation 1: LAS = -7.866+1.463 LY + 0.438 LCV (Eq.1)

The signs of the elasticities found in this paper are in accordance with the economic theory. The elasticities regarding the income and the credit volume are statistically significant at 1%. Our next step is to develop scenarios for the independent variables. It was assumed that the income growth rate will be equal to the economic growth rate minus the population growth. Besides, the volume of credit, which has been grown significantly since the end of the 90s, is assumed to follow the economic growth rate.
Having the automobile model, the next step is to project the independent variables behavior. It was assumed that the income growth rate will be equal to the economic growth rate minus the population growth. Besides, the volume of credit, which has been growing fast since the end of the 90s, will follow the economic growth rate. As a way of analyzing the fleet evolution, we utilize a scenario for the independent variables gives the necessary database to estimate the light vehicles sales until 2020 by Equation 1. It is assumed that the share of vehicles sales by fuel and type does not change in the next ten years. This hypothesis allows us to estimate the total light vehicles sale. In 2009, 82% of the total sales corresponded to cars sale, which were estimated by our econometric model, while 18% respond to light commercial. So, applying these proportions we obtain the total light vehicles sale. Then, applying the proportion by fuel and type we generate the automobile and light commercials sales by fuel. Having this information, the scrapping rate, estimated previously, can be applied. Table 6 –Light Vehicles Sales by Fuel and Type (%) in 2009 Gasoline cars Gasoline light commercials Flex-fuel cars Flex-fuel light commercials Diesel cars 3,78% 3,37% 77,77% 9,59% 0,00%

Diesel light commercials Ethanol cars Ethanol light commercials NGV cars NGV Light commercials

4,44% 0,00% 0,00% 0,80% 0,24%

The economic growth rate expected in 2020 is equal to 4.8% and the population growth rate is 0.57%, which is sustained in all scenarios. Thereby, the results suggest that in 2020 the light vehicles fleet will be equal to 46,853,067 unities. As expected, the flex-fuel vehicles will respond to 78.2% of total feet while gasoline vehicles will decrease to 14.6%. Besides, the ethanol vehicle fleet will almost disappear 0.4%. Table 7 – Fleet by Fuel (% Total Fleet) in 2009 and 2020 – Scenario 1
Gasoline 49,8% 14,6% Flex-Fuel 35,1% 78,2% Diesel 4,5% 4,5% Ethanol 5,3% 0,4% NGV 5,3% 2,4%

2009 2020

3. Modeling fuel choice
Fuel consumption modeling is separated in two models. First we estimate the aggregate of ethanol and gasoline consumption. Then we will analyze the substitution between the two fuels. In the first model we need to distinguish the tree different types of fuel technologies: the cars that run only with gasoline; those that run exclusively with ethanol; and flexfuel cars. The importance of distinguishing these different segments of consumers is based on their different response on price variation. Indeed, captive consumers (ethanol and gasoline fleet) do not respond to short-run price fluctuation. In the other hand, consumers with flexible cars can chose between fuels, depending on their relative price. 3.1 Fuel Choice Model Fuel consumption depends on the total fleet and fuel price. We use a price index that combines ethanol and gasoline prices. We propose the following model structure: FC = F (Fleet, Price) Where: FC = Total fuel consumption (ethanol e gasoline) Fleet = Total light vehicle fleet Price = price fuel index The substitution model is base on a logistic curve relating the relative price between ethanol and gasoline with the ratio of ethanol consumption and it potential

consumption. That potential is defined as the entire flexfuel fleet consuming only ethanol. So this model concerns only the flex fleet, since they are those who respond to price incentives. So we have the relative consumption as function of the relative price: EC/PC = F (EP/GP) where: EC/PC = Effective consumption on potential consumption EP/GP = Ethanol price on gasoline price - Data 3.2 Results on different price scenarios Our estimated model for fuel consumption is:

d = 2,09 Where, = dummy for the month t This estimation considers a scenario of 4.5% economic growth and a stable index price. Figure 3 shows total automotive fuel consumption estimation. Although, growth rate will decrease relating to the past five years, it is still very high 7.7% per year until 2020. Figure 3 – Observed and estimated total fuel consumption (ethanol + gasoline) – 20032020. Liters of gasoline equivalent

7.000.000 Liters of gasoline equivalent 6.000.000 5.000.000 4.000.000 3.000.000 2.000.000 1.000.000 0 01-12-03 01-10-04 01-08-05 01-06-06 01-04-07 01-02-08 01-12-08 01-10-09 01-08-10 01-06-11 01-04-12 01-02-13 01-12-13 01-10-14 01-08-15 01-06-16 01-04-17 01-02-18 01-12-18 01-10-19 01-08-20

estimation

consumption observed

Source: Elaborated by the authors The fuel substitution model focus on fuel choice by the flex fleet owners. These are able to move from one fuel to the other depending on relative fuel price. As figure (4.a) below shows, consumers are very sensible to price changes. Therefore we estimate a reaction function to price movements (figure 4.b). Figure 4.a: Relative price and relative consumption observed

Figure 4.b: Relative price and relative consumption estimation

Source: Elaborated by the authors

Consumer demand is more elastic next to the threshold of 0.7 relative price since it represents the point where both fuels are equivalents in terms of energy content. Our work is based on two different relative price scenarios for modeling the substitution between gasoline and ethanol. To analyze the impact of different relative price scenarios on ethanol and gasoline consumption, and consequently on CO2 emission, we chose the 0.7 and 0.8 relative price values that represent the national average of the last five years and the average of the year 2011, respectively. These scenarios are useful to illustrate the impact of a the price level change that has been taking place in Brazil these past few years. The following figures show this impact. In the 0.7 relative price scenario ethanol consumption represents 52% of fuel consumption in energy equivalent terms. This share reduces to 25% in the 0.8 price scenario. Figure 5: Ethanol and Gasoline consumption with 0.7 relative price (flexfuel fleet)
6.000.000 5.000.000 4.000.000

3.000.000
2.000.000 1.000.000 jan-12 out-12 jan-15 out-15 jan-18 out-18 jan-21 out-21 abr-14 abr-17 abr-20 jul-13 jul-16 jul-19 jul-22

Gasoline Ethanol

Source: Elaborated by the authors

Figure 6: Ethanol and Gasoline consumption with 0.8 relative price (flexfuel fleet)
6.000.000 5.000.000 4.000.000 3.000.000 2.000.000 1.000.000 jan-12 out-12 jan-15 out-15 jan-18 out-18 jan-21 out-21 abr-14 abr-17 abr-20 jul-13 jul-16 jul-19 jul-22 Gasoline Ethanol

Source: Elaborated by the authors

4. CO2 Emissions The different price scenarios have substantial impact on CO2 emissions. Use of ethanol traditionally implicated in a relatively low carbon automotive segment in Brazil. However, CO2 emissions depend on consumer choice between ethanol and gasoline. Considering emissions from fuel production to the use in cars, a liter of ethanol consumed generates 2.6 kg of CO2 emissions comparing to 5.6 kg generated by a liter of gasoline. Using those ratio, we estimated CO2 emissions on the two price scenarios. In the 0.7 price scenario, CO2 emission reaches 25 thousands tones. It increases to 28 thousand in the 0.8 scenario.

Figure 7: CO2 Emissions in the automotive sector 2012 -2020 toneladas CO2
30.000 28.000 26.000 24.000 22.000 20.000 18.000 16.000

14.000
12.000 10.000 abr/12 ago/12 dez/12 abr/13 ago/13 dez/13 abr/14 ago/14 dez/14 abr/15 ago/15 dez/15 abr/16 ago/16 dez/16 abr/17 ago/17 dez/17 abr/18 ago/18 dez/18 abr/19 ago/19 dez/19 abr/20 ago/20 dez/20

Source: Elaborated by the authors

References MATTOS, J A AND CORREIA, E L. (1996). Uma nova estimativa da frota de veículos automotivos no Brasil. In: Anais do VII Congresso Brasileiro de Energia, p. 1267. LOSEKANN, L., VILLELA, T. AND IOOTTY (2010) Estimation of Brazilian Car Fleet: Understanding FuelSubstitution on the Automotive Sector. Proceedings of the Rio Oil & Gas Expo and Conference 2010. HIRA, A. E L. G. OLIVEIRA (2009), No substitute for oil? How Brazil developed its ethanol industry. Energy Policy 37. Pgs. 2450–2456. COELHO, S., J. GOLDEMBERG, O. LUCON, E P. GUARDABASSI (2006), Brazilian sugarcane ethanol: lessons learned. Energy for Sustainable Development. Volume X No. 2.