- Email: [email protected]
Flex cars and competition in fuel retail markets
Accepted Manuscript
Flex Cars and Competition in Fuel Retail Markets Joao ˜ Paulo Pessoa, Leonardo Rezende, Juliano Assunc¸ao ˜ PII: DOI: Reference:
S0167-7187(17)30170-4 https://doi.org/10.1016/j.ijindorg.2018.07.005 INDOR 2462
To appear in:
International Journal of Industrial Organization
Received date: Revised date: Accepted date:
7 March 2017 16 July 2018 17 July 2018
Please cite this article as: Joao ˜ Paulo Pessoa, Leonardo Rezende, Juliano Assunc¸ao, ˜ Flex Cars and Competition in Fuel Retail Markets, International Journal of Industrial Organization (2018), doi: https://doi.org/10.1016/j.ijindorg.2018.07.005
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
Flex Cars and Competition in Fuel Retail Markets∗ Leonardo Rezende‡ October 30, 2018
Abstract
Juliano Assun¸ca˜o§
CR IP T
Jo˜ao Paulo Pessoa†
We study how the diffusion of flex (bi-fuel) cars affected competition on ethanol and gaso-
AN US
line retail markets. We propose a model of price competition in which the two fuels become closer substitutes as flex cars penetration grows. We use a large panel of weekly prices at the station level to show that fuel prices and margins have fallen in response to this change. This finding is evidence of market power in fuel retail and indicates that innovations that
M
increase consumer choice benefit even those who choose not to adopt them.
Keywords: Flex-fuel vehicles; Gasoline; Ethanol; Price competition; Spatial Competi-
ED
tion; Discrete equilibrium price dispersion.
AC
CE
PT
JEL Codes: L11, L13, L62, L71.
∗
We are grateful to Jo˜ao Manoel Pinho de Mello, Heleno Pioner, Cristian Huse and seminar participants at the IIOC 2012, EARIE 2011, LAMES 2011, Taller de Organizacion Industrial 2010, FEA USP-RP, FEA USP, IE UFRJ and IPEA for helpful comments. We also thank Gabriela Rochlin for superb research assistance and ANP for providing access to data. All remaining errors are our own. A previous version of this paper circulated under the title “Flex Cars and Competition on Ethanol and Gasoline Retail Markets”. † Sao Paulo School of Economics - FGV, S˜ao Paulo, Brazil; Centre for Economic Performance, London, UK. E-mail: [email protected]. ‡ PUC-Rio, Rio de Janeiro, Brazil. E-mail: [email protected]. § PUC-Rio, Rio de Janeiro, Brazil. E-mail: [email protected].
1
ACCEPTED MANUSCRIPT
1
Introduction
The key aspect underlying the identification of conduct in empirical studies in industrial organization is the relationship connecting changes in the price elasticity of demand and firms’ pricing decisions (Bresnahan, 1982). In this paper, we explore two features of the automotive fuel retail market in Brazil that allow us to directly document this relationship. First, Brazil is a dual fuel market: both gasoline and ethanol have been available for
CR IP T
automobiles at virtually every fuel station in the country since the 1980s. Second, flex cars have been available in Brazil since 2003 and have allowed consumers to treat these two fuels as nearly perfect substitutes at the pump. Flex, or bi-fuel, vehicles are able to run on any mix of gasoline and ethanol fuel; electronic sensors identify the mix at the fuel tank and adjust the fuel injection accordingly. They have become a commercial success:
AN US
in 2008, 94% of the new cars registered in the country were flex. This innovation provides a source of change in the cross-price elasticity between two products that allow us to directly identify its effect on pricing.
We build a model of strategic price formation to study the impact of the flex car
M
fleet on equilibrium fuel prices. In the model, stations compete by choosing the price of gasoline and ethanol, and consumers treat fuel from different stations as imperfect
ED
substitutes, due to location or other idiosyncratic preferences. The model suggests that, in equilibrium, fuel retailers respond strategically to an increase in flex car penetration
PT
(that is, an increase in substitutability between products within its own product line) by reducing markups. There is a clear difference in comparison to other differentiated-
CE
goods oligopoly models: in our setting, the law of one price has no bite. Our theory does not predict that the price of the two fuels should move closer as they become perfect
AC
substitutes to a growing share of the consumers. Because both prices are set by the same firm, it is generally optimal to keep prices apart to price discriminate for consumers who cannot freely switch. We empirically assess the model by documenting the causal effect of flex-fuel pene-
tration on fuel prices using reduced-form methods.1 We study the impact of the flex car fleet on retail prices and the spread between gasoline and ethanol. Because the speed of penetration of this technology has been unequal across localities (roughly driven by the 1
We also provide estimates of the structural parameters of the model to investigate in more detail some of our theoretical predictions.
2
ACCEPTED MANUSCRIPT pace of car fleet renewal), we have been able to employ panel data methods to control for aggregate time-varying effects and local fixed effects, using a detailed sample of weekly prices at the gas station level. Our empirical analysis exploits variation in the flex fuel penetration due to local differences in the speed of fleet renewal, which is mostly driven by cross-sectional variation in income and economic activity. (We account for this possible source of omitted variable
CR IP T
bias by adding income as an additional control.) In principle, variation in fuel prices (or rather variation in the expectation about future fuel prices) may also have an effect on fleet renewal; several studies have shown that fuel prices affect demand for automobiles (Busse, Knittel and Zettelmeyer, 2013; Goldberg, 1998; Kahn, 1986; Klier and Linn, 2010; Li, Timmins and Haefen, 2009; Pakes, Berry and Levinsohn, 1993), and even specifically
AN US
consumer choice between diesel and gasoline vehicles in Europe (Verboven, 2002). To account for the potential endogeneity due to reverse causality, we employ an IV strategy that builds on the recent empirical trade literature (Autor, Dorn and Hanson, 2013; Costa, Garred and Pessoa, 2016), using data on the local presence of car dealerships. Consistent with the predictions of our model, the results show that fuel stations have
M
significantly reduced ethanol and gasoline prices: a 10 percentage point increase in the market share of flex cars reduces ethanol and gasoline prices by approximately 8.1 and
ED
2.8 cents of BRL, respectively. The absolute effects are stronger for ethanol, which is consistent with the fact that ethanol has a smaller market share in the automotive fuel
PT
market in Brazil. Our results provide evidence of market power in fuel retail and that innovations that increase consumer choice may benefit even those consumers who choose
CE
not to adopt them. We also find (weaker) evidence that the expansion of the flex car fleet has increased the spread (that is, the absolute price difference) between gasoline and
AC
ethanol prices.
Moving beyond the model, we study how flex car penetration has affected retail mar-
gins and wholesale prices (or retail costs). We find evidence that fuel retail margins have fallen with the increase of flex cars, while the estimated effects of flex cars on wholesale prices are statistically non-significant in most of our specifications. This last result corroborates the interpretation that wholesale prices are more tightly linked to international markets and less influenced by local market effects. Our paper contributes to an increasing literature on the industrial organization of
3
ACCEPTED MANUSCRIPT ethanol as automotive fuel and its relation to the gasoline market. Anderson (2012), Corts (2010) and Shriver (2015) are examples of recent studies that investigate the ethanol market in the US.2 Anderson (2012) studies the demand for the product. Shriver (2015) studies the network effect that arises due to spatially dependent complementarities between the availability of stations supplying ethanol fuel and the local number of flex cars. Corts (2010) also analyzes the decision to supply ethanol by local stations, using as a
CR IP T
source of variation purchases of flex cars by government agencies. This emphasis on the issue of expanding the distribution network reflects the incipient nature of ethanol as automotive fuel in the US. In Brazil, by contrast, the challenge of building an extensive distribution network has been completed in the 1980s with the Pr´oa´lcool program, further discussed in section 2 below. Brazil provides a setting where it is
AN US
possible to study a mature dual-fuel market operating.
Most existing studies employing Brazilian data (Ferreira, Prado and Silveira, 2009; Salvo and Huse, 2011; Boff, 2011) use time series of average price data to look for evidence of convergence toward the law of one price between the fuels3 . In contrast to this literature, we employ much more detailed data, which allows us to document the importance of
M
price dispersion across stations (an important feature of automotive fuel markets; see, e.g., Lewis, 2008). In addition, we argue in this paper that because of the structure of
ED
the retail market for fuel in Brazil, price convergence should not necessarily occur. The paper is organized as follows: section 2 provides a brief summary of the general
PT
characteristics of the Brazilian fuel market. Section 3 presents a model of oligopolistic competition among fuel stations supplying both types of fuel, allowing us to establish some
CE
key predictions on how flex car diffusion affect equilibrium prices. Section 4 describes the data we use and provides descriptive statistics. In section 5 we employ panel data methods
AC
to empirically verify the effect of flex car diffusion on prices and on the absolute price spread between gasoline and ethanol. We provide concluding remarks in section 6. 2
More precisely, E85; in the US, retail stations supply a composition of 85% of ethanol and 15% of gasoline called E85 instead of pure ethanol (E100) supplied in Brazil. 3 Representing an exception are Salvo and Huse (2013), who employ an opinion poll among flex car owners to document the relevance of motives other than price to choose between fuels.
4
ACCEPTED MANUSCRIPT
2
Flex Cars and the Automotive Fuel Market in Brazil
2.1
Ethanol-Powered and Flex Cars
Brazil has a long history of using ethanol as a vehicular fuel. In the 1970s, in response to the first oil crisis, the military government launched the Pr´o-´alcool program to encourage the production of ethanol from sugarcane and stimulate the adoption of ethanol-fueled
CR IP T
cars. The program included the use of credit subsidies for ethanol production and set favorable fuel prices at the pump to stimulate adoption of the new technology. Consumers responded to the Pr´o-´alcool program: from 1983 to 1989, most new cars purchased were ethanol-fueled vehicles. This finding may be seen in figure 1, which presents shares of new car registrations in Brazil per year by fuel type.
AN US
In response to a sharp increase in the global price of sugar, which tripled from 1985 to 1990 (USDA, 2010, table 3a), domestic ethanol production sharply declined, and the ensuing supply crisis led to a plunge in sales of ethanol-powered vehicles. Since 1995, sales of ethanol-powered cars have represented only a small fraction of new vehicle sales in Brazil. However, ethanol-fueled cars continue to represent over 10% of the current fleet.
M
In the first quarter of 2003, flex cars (or bi-fuel cars) became commercially available in Brazil. Flex cars may run on any mixture of gasoline and ethanol. Because a liter of
ED
ethanol contains roughly as much energy as 0.7 liters of gasoline (Marjotta-Maistro and Asai, 2006), a flex car owner may save money if the price ratio drifts away from that
PT
threshold.4 As figure 1 shows, flex car penetration has been dramatic: In 2008, 94% of
2.2
CE
new cars registered in Brazil were flex cars.
The Automotive Fuel Market in Brazil
AC
In Brazil, ethanol, or ethyl alcohol, is made from sugarcane. Two types of ethanol play a role in the automotive fuel market: anhydrous and hydrated. Anhydrous ethanol is mixed with gasoline fuel in the proportion of one unit of ethanol to three units of gasoline. Hydrated ethanol, a mixture that contains 5% water, is the version of alcohol readily available in drugstores and pumps at fuel stations in Brazil. Brazil is the largest producer of sugarcane in the world, the second-largest producer 4
There are other facts that might lead consumers to choose one fuel over the other. A car running on ethanol is less hazardous to the environment, as it does not create net emissions of carbon dioxide. A car powered by gasoline demands less fuel per volume, thus allowing for less frequent refueling.
5
ACCEPTED MANUSCRIPT
1980
1985
1990
1995
2000
2005
2010
AN US
1975
CR IP T
0
20
40
%
60
80
100
Figure 1: Registration of New Cars by Fuel Type
Year
Gasoline
Ethanol
Flex
M
NOTES: Figure shows shares (in percentage) of new car registrations by fuel type. Diesel and electric cars are not included in the calculations. Sources: Anfavea.
of ethanol, and a net exporter of ethanol. Brazil was a net importer of oil until 2006.
ED
In contrast, it was a net exporter of gasoline during the period of our analysis, but this pattern reversed recently and Brazil is now a net importer of the product.
PT
Before the passage of Law 9478 in 1997 (“Lei do Petr´oleo”), the Brazilian oil industry was a monopoly in the hands of state-owned Petrobras. The law created the Agˆencia
CE
Nacional do Petr´oleo, G´as Natural e Biocombust´ veis (ANP), the sector regulatory body, and broke Petrobras’ monopoly on exploration, refining, international trade, and the sea
AC
transport of oil and its main byproducts. Since January 2002, retail prices are set freely by the market. The industry is com-
posed, on the production side, of oil companies, sugar mills and refineries, which produce the fuel to the market. On the distribution side there are wholesale distributors and retailers. According to CADE (2014) there are about 200 wholesale distributors and around 40000 retailers (gas stations) in Brazil. Approximately 60% of the gas stations have exclusivity contracts with distributors. Petrobras continues to be a major player in the domestic refining, distribution and retail of gasoline, currently holding market shares
6
ACCEPTED MANUSCRIPT in these markets of 96.6%, 28.9% and 17.8%, respectively (ANP, 2010, tables 2.34, 3.6 and 3.17).
3
Model: Price Formation in the Retail Fuel market in the Presence of Flex Cars
CR IP T
In this section, we present a model of strategic price formation in the retail fuel market and use it to investigate theoretically the effect of a larger flex car fleet on equilibrium fuel prices.
The model we propose has four features that are relevant for the automobile fuel market, particularly in areas where multiple fuels are offered through the same distribution
AN US
network: i) aggregate demand for fuel (adding up different types of fuel) is proportional to the size of the automobile fleet and thus is inelastic in the short run; ii) location differences allow fuel stations to have a degree of market power; iii) every fuel station supplies both types of fuel and selects both prices to maximize joint profits; iv) for flex car owners,
M
ethanol and gasoline are substitutes.
In the model, there is (imperfect) competition across fuel stations, but between differ-
ED
ent types of fuel within a location, there is no competition at all: each station provides both ethanol and gasoline and internalizes the effect of a price change over the sales of
PT
the other product. The effect of an increase in the flex car fleet is to change the degree of substitutability between fuels and not across stations. One may tend to suppose that
CE
because competition and the direct effect of flex car penetration operate in different dimensions of product differentiation, there would be no effect of the latter on the former. This supposition is not true: we find that in equilibrium, flex car penetration leads to
AC
more competition across fuel stations. The theoretical analysis depends on an important feature of the industry: The large
cross-sectional price dispersion even within a market. In fact, it is this price dispersion that affects the behavior of flex car owners by proving them with an option value. Price dispersion in gasoline retail is a well documented phenomenon (Lewis, 2008; Hosken, McMillan and Taylor, 2008; Chandra and Tappata, 2011a). Lack of full substi-
tutability across stations may arise because of product differentiation due to geography (Lewis, 2008; Pennerstorfer, 2009; Firgo, Pennerstorfer and Weiss, 2015) or imperfect in7
ACCEPTED MANUSCRIPT formation by consumers (Chandra and Tappata, 2011a; Pennerstorfer et al., 2015; Byrne and de Roos, 2017). While recent contributions to the literature seek to provide direct empirical evidence of the latter possibility,5 we adopt a more agnostic approach. We posit a standard demand system with differentiation across stations that is compatible with the Carlson and McAfee (1983) model of consumer search, but the analysis would carry
3.1
CR IP T
through if differentiation was due to other factors.
Baseline Model
We consider an oligopoly model where N gas stations compete by setting prices for gasoline and ethanol. The population of consumers, which we normalize to 1, is divided into three groups: gasoline-fuelled car owners, ethanol-fuelled car owners, and flex car owners. We
AN US
call θ the fraction of flex car owners, and to maintain symmetry in the model, we assume that the rest of the fleet is equally divided between gasoline and ethanol. The type of fuel generally has no effect on consumption, but for flex car owners, gasoline and ethanol within the same gas station are perfect substitutes. (We measure fuel in terms of energy
M
content, so flex car owners always buy the cheaper fuel.)
There is differentiation across fuel stations. For a given fuel price profile, let pif =
ED
min{pig , pia } (the price effectively faced by a flex type in station i). Using this notation, we assume that the demand for fuel from station i from a consumer with car type j = g
qij = α − βpij + γ p¯ij ,
CE
PT
(gasoline-powered), a (alcohol-powered), or f (flex) is
where α, β and γ are positive constants, pij is the price of fuel j in station i, and p¯ij is
AC
the average price of fuel j in all stations except i. In the baseline model, we adopt the same functional form for all fuel types. This
procedure is followed for simplicity and to isolate the effect on substitutability across fuels as the car fleet changes. (We relax this assumption in section 3.4 below.) For consumers within each car group, we assume that demand across stations exhibits a simple linear form of symmetric product differentiation. This demand system may be justified by 5
Pennerstorfer et al. (2015) and Byrne and de Roos (2017) construct altenative measures of the fraction of informed consumers (namely, fraction of commuters within a locality and web traffic in a price comparison website) and study how those relate to price dispersion.
8
ACCEPTED MANUSCRIPT Carlson and McAfee (1983), who model consumer choice by a process of costly search among identical products sold by different firms at (potentially) different prices. Carlson and McAfee show that if the distribution of search costs in the consumer population is uniform, then aggregate demand for firm i exhibits the form postulated above, with α = 1/N and β = γ = (N − 1)/N , where N is the number of firms in the market. We seek to obtain a prediction regarding Bertrand-Nash equilibrium prices for firms
CR IP T
that face demand arising from this process and have a cost function as follows: Ci (qig , qia ) = cig qig + cia qia + Fi ,
where qig and qia are the quantities sold of gasoline and ethyl alcohol in station i, cig
AN US
and cia are marginal costs, and Fi is a fixed cost component. We assume that marginal costs are constant and exogenous, but different across fuels and stations. We believe that assuming that marginal costs differ across stations is reasonable, given that, according to our data, there is substantial variation in fuel wholesale prices faced by each station.
Properties of Equilibrium Prices
M
3.2
To obtain a characterization of equilibrium prices, we must first integrate the demand
ED
over the mass of consumers with each car type. If pig 6= pia , station i will sell Qig of gasoline and Qia of ethanol, where
1−θ = (α − βpig + γ p¯ig ) 2 +1I{pig < pia }θ (α − βpig + γ p¯if )
PT AC
and
CE
Qig
Qia
1−θ = (α − βpia + γ p¯ia ) 2 +1I{pig > pia }θ (α − βpia + γ p¯if ) .
1I{A} represents the indicator function, with value one if condition A is true and zero otherwise. If pig = pia , flex car owners are indifferent between the two types of fuel, and we must specify a sharing rule τ ∈ [0, 1]. Formally, we follow the approach of Simon and Zame (1990) and adopt an endogenous sharing rule, although the specifics of the 9
ACCEPTED MANUSCRIPT tie-breaking do not affect the equilibrium determination in this model. The profit of station i is simply πi = (pig −cig )Qig +(pia −cia )Qia −Fi . Maximizing this expression with respect to pig and pia yields this firm’s best-response function. Whenever θqif is positive, πi is discontinuous at the point pig = pia (and the profit at the discontinuity point depends on the tie-breaking rule). In any pure-strategy equilibrium, we may classify stations into those that choose to
CR IP T
charge pig > pia , pig < pia or pig = pia . In the first two cases, profits are continuously differentiable around the chosen prices, and the latter may be characterized by the firstorder conditions
∂ π ∂pig i
= 0 and
∂ π ∂pia i
= 0.
In the next proposition, we show that the last alternative is never optimal: in equilibrium, no fuel station elects to charge pig = pia :
AN US
Proposition 1 If θ > 0 then any firm i will post pig 6= pia in equilibrium. Proof.
If some of the other fuel stations charge a different price for each fuel, p¯if < p¯ig or p¯ia . Without loss of generality, suppose that p¯if < p¯ig .
M
The profit of firm i, as a function of pig , is discontinuous at point pig = pia . The
ED
value of the profit at this point depends on how flex car owners break the tie between fuels, but it always lies between the left-hand and the right-hand limits limpig %pia πi and limpig &pia πi , which correspond to the extreme cases that all flex car owners buy ethanol
PT
or gasoline when prices are equal, respectively. Therefore, in order to show that pig = pia cannot be optimal, it suffices to show that
point.
CE
either the left-hand derivative is negative or the right-hand derivative is positive at that
AC
Let x and y be the right and left derivatives, respectively, at that point: x=
∂ + (1 − θ) π = [q(pig , p¯ig )) − β(pig − cig )] ∂pig i 2
y=
∂ − π = x + (θ)[q(pig , p¯if )) − β(pig − cig )] ∂pig i
and
For firm i to find it optimal to charge pig = pia , it must be the case that y ≥ 0 ≥ x. However, such a case is impossible as x ≤ 0 ⇒ y < 0. 10
ACCEPTED MANUSCRIPT If all other stations charge the same price for both fuels, p¯if = p¯ia = p¯ig = p¯. Evaluating the left and right derivatives around a point pig = pia = p, we define as before x=
∂ π+, ∂pig i
and y =
∂ π− ∂pig i
and, analogously, x0 =
∂ π+, ∂pia i
and y 0 =
∂ π−. ∂pia i
As we argued
above, x < 0 ⇒ y < 0 and x0 < 0 ⇒ y 0 < 0. For the first-order condition to be satisfied,
we need x = 0, x0 = 0 at this point. However, this condition is impossible because
CR IP T
x = [(1 − θ)/2][q(p, p¯)) − β(p − cig )] 6= x0 = [(1 − θ)/2][q(p, p¯)) − β(p − cia )], as cig 6= cia .
Considering the two possible first-order conditions that must be satisfied by pij , we obtain the following expressions:
AN US
α θ γ 1 γ (¯ pig − p¯if ) pig = cig + p¯ig + − 1I{pig < pia } 2 β β 1+θβ and
1 γ α θ γ pia = cia + p¯ia + − 1I{pig > pia } (¯ pia − p¯if ) 2 β β 1+θβ
M
Note that p¯if ≤ p¯ig and p¯if ≤ p¯ia , so the right-hand sides of the expressions above are decreasing in θ. Because prices across stations are strategic complements, we conclude
ED
that prices are decreasing in θ. This result is summarized in the proposition below.
flex cars.
PT
Proposition 2 Ethanol and gasoline prices are decreasing with respect to the fraction of
CE
A larger fleet of flex cars pulls prices down because flex cars provide an option value to their owners: if fuel prices are dispersed, flex car owners expect to find lower prices
AC
than other drivers because they can always pick the cheapest alternative. For this reason, flex car owners are willing to pay less, and fuel stations respond by lowering prices.
3.3
Effect on price spreads
Let us turn to the analysis of the dispersion between gasoline and ethanol prices. Common sense from the law of one price suggests that as the flex car fleet grows, the prices of the two fuels should move closer together. In this paper we argue that this intuition may not hold if the two fuels are sold by the same retailers, as in the case of the Brazilian market. 11
ACCEPTED MANUSCRIPT In this section we employ the model to show that in fact the price spread, defined here as the the absolute difference between fuel prices charged by the same station, actually moves in the opposite direction, and becomes larger as the market share of flex cars grow. From the price best response functions we obtain the following expression for the price spread of station i:
1 2 1 2
|cig − cia | + |cig − cia | +
γ β γ β
|¯ pig − p¯ia | + |¯ pig − p¯ia | +
θ γ (¯ pig 1+θ β θ γ (¯ pia 1+θ β
− p¯if ),
if pg < pa ;
− p¯if ),
if pg > pa .
CR IP T
|pig − pia | =
Since p¯if ≤ p¯ia , p¯ig , the spread is weakly increasin the θ, and strictly increasing whenever the cheapest fuel is not the same everywhere.
Introducing asymmetry across fuels
AN US
3.4
To simplify notation, in our baseline model markets for gasoline and ethanol are fully symmetric. In reality these markets are quite different: at the very least, the disparate sizes of the fleets make the gasoline market much larger than the ethanol market; also
M
since automobiles with different fuel types have been purchased at different types, and may be used in different ways, demand functions for each fuel may differ.
ED
In this section we investigate the effect of introducing asymmetry in our model in two steps. In the first extension, we allow for different fleet sizes, while holding the same
PT
demand function of the baseline model. This does not fundamentally change our analysis. In fact a stronger version of proposition 2 holds: any shift in the composition of the fleets
CE
that increase the share of flex cars and reduces the share of either gasoline or ethanol fleets has the effect of reducing prices of both fuels.
AC
To see that, let the share of flex, gasoline and ethanol cars be θf , θg and θa , respectively. (We no longer impose θg = θa as in the baseline model.) In that case the amount of fuel sold by station i will be Qig = θg (α − βpig + γ p¯ig ) +1I{pig < pia }θf (α − βpig + γ p¯if )
12
ACCEPTED MANUSCRIPT and Qia = θa (α − βpia + γ p¯ia ) +1I{pig > pia }θf (α − βpia + γ p¯if ) . From the first-order conditions, we obtain the following best responses:
α γ p¯ia + , 2β 2β α γ(θa p¯ia + θf p¯if ) + , 2β 2β(θa + θf )
if pg > pa ;
CR IP T
c ia + 2 pia = c ia + 2
α γ p¯ig + , 2β 2β α γ(θg p¯ig + θf p¯if ) + , 2β 2β(θg + θf )
if pg < pa ;
if pg < pa ; if pg > pa .
AN US
c ig + 2 pig = c ig + 2
Once again since p¯if ≤ p¯ia , p¯ig , any change that increases θf and reduces θg and θa has the direct effect of reducing one of the prices of each station (the lowest). In equilibrium since prices are strategic complements, equilibrium prices fall.
M
A further possible extension is to allow individual demand functions to vary according to type of fuel. Flex fuel car technology has been introduced gradually over the period of
ED
analysis; thus flex fuel cars tend to be newer than gasoline cars. It is reasonable to expect that their owners are richer (and therefore less price sensitive) and the cars are used more
PT
intensively, thus consuming more fuel. (Similarly, existing ethanol-only cars tend to be very old since they were effectively no longer manufactured in the period of analysis.)
CE
A simple way to add asymmetry in demand such that comparison with the baseline model is straightforward is to allow the coefficients of the demand function to depend on
AC
fuel type. Suppose that individual demand for a car with fuel j = g, a, f is qij = αj − βj pij + γj p¯ij .
In that case the best response functions become c αg + γg p¯ig ig + , 2 2βg pig = c θ α + θf αf + θg γg p¯ig + θf γf p¯if ig + g g , 2 2(θg βg + θf βf ) 13
if pg > pa ; if pg < pa ;
ACCEPTED MANUSCRIPT and
c αa + γa p¯ia ia + , 2 2βa pia = c θ α + θf αf + θa γa p¯ia + θf γf p¯if ia + a a , 2 2(θa βa + θf βf )
if pg < pa ; if pg > pa .
In the model, composition effects may affect our main conclusion. Taking derivatives, we find that to guarantee that an increase in the flex car fleet share lowers prices, it must be
CR IP T
true that αf βg − αg βf + βg γf p¯if − βf γg p¯ig ≤ 0
(1)
αf βa − αa βf + βa γf p¯if − βf γa p¯ia ≤ 0.
(2)
AN US
and
A simpler sufficient condition would be βf ≥ βg , βa , αf ≤ αg , αa (unlikely) and γf ≤ γf ≤ γg , γa .
In Appendix B - below, we propose and apply a method to structuraly estimate these
M
parameters (up to a normalization). Relying on our preferred specification (column 4 in table B.1), we find that the βi ’s are similar, γf is smaller than γg and γa , and αf is larger
ED
than αg and αa . These two last findings are compatible with the presumption that flex car owners use their cars more intensively (demand more fuel and are less elastic). With
PT
respect to the condition needed for an increase in the flex car fleet to bring prices down, a sufficient condition is having average ethanol prices above 0.98 and average gasoline
CE
prices above 1.69, a condition that is satisfied in the majority of locations and periods in
AC
our sample.
4
4.1
Data Data Sources
This study combines data from different sources. The first source is the Levantamento de Pre¸cos e de Margens de Comercializa¸c˜ ao de Combust´ıveis, a weekly survey conducted by ANP, the Brazilian regulatory agency covering the oil, gas and biofuel industry. ANP collects data on retail prices for ethanol and gasoline prices, as well as prices paid to fuel
14
ACCEPTED MANUSCRIPT distributors, at individual fuel stations in 10% of the municipalities in Brazil. There is also information on the brand of the station (or whether it has no brand), the date on which prices were collected and the address of the station. Our sample contains weekly prices from January 2002 to March 2008 for stations located in 38 municipalities in the Rio de Janeiro state. Not all fuel stations are surveyed every week. Coverage is 100% in small municipalities, whereas for larger markets, the survey adopts a rotating sample
CR IP T
(with random selection) that eventually covers all fuel stations in the location. Table 1 provides information on the number of stations sampled, as a proportion of the overall population of stations, in the 38 municipalities used in this study.
The second data source is a monthly data set on the number of cars with license plates from each municipality in the state of Rio de Janeiro, classified according to fuel
AN US
type (gasoline, ethanol, flex, gasoline + CNG6 , ethanol + CNG, flex + CNG). The time period is the same considered in the ANP’s survey (from January 2002 to March 2008). This data set was provided by the motor vehicles department of the state of Rio de Janeiro (Detran - RJ). Although ANP verifies the price charged by fuel stations in all Brazilian states, we are unable to expand our analysis to the entire Brazilian territory because we
M
do not have access to data on the number of cars by fuel type in other states. We included annual municipal GDP per capita as a regressor (obtained from the
ED
Brazilian institute of geography and statistics - IBGE). We also included annual data on number of hotels per square km in each municipality (provided by the Data and
PT
Information Center of Rio de Janeiro - CIDE RJ) as a proxy for markets where a large fraction of drivers are not local. The former series is available from 2002 to 2006, and the
CE
latter is available from 2002 to 2004. Because we are analyzing the period between 2002
AC
and 2008, in both cases the missing years were replaced by the most recent available ones.
4.2
Descriptive Statistics
Figure 2 presents a map of the Rio de Janeiro state with a bullet for each municipality in our sample. The size of the bullet is proportional to (the log of) the local car fleet, and the color is coded according to flex car penetration in 2007. As is shown, the flex car fleet grew considerably across time in all locations, reaching a maximum value of 14.1% 6
CNG stands for compressed natural gas. In Brazil, it is possible to convert vehicles to run on natural
gas.
15
ACCEPTED MANUSCRIPT in the city of Mangaratiba in 2007. There is variation both in the time series and the cross-section/geographic dimensions. Figure 2 strongly suggests that flex car adoption is related closely to local income, growing the most in the capital (the Rio de Janeiro city), tourist resort towns (Arma¸c˜ao de B´ uzios, Parati, Angra dos Reis), and rapidly developing areas (Maca´e, Mangaratiba). The municipalities with the lowest adoption rates are located in the north-eastern part of
CR IP T
the state, the traditional region of sugarcane production in Rio de Janeiro. Table A.1 in the Appendix shows how the percentage of flex cars changed between 2004 and 2007 in the 38 municipalities in our sample.
Table 1: Gasoline and Ethanol Average Prices by Year in the State of Rio de Janeiro (in R$)
2002 2003 2004 2005 2006 2007
1.021 1.175 1.014 1.150 1.325 1.161
1.643 1.774 1.656 1.741 1.837 1.752
Ethanol P rice Gasoline P rice
Share of Sample with Cheaper Ethanol 0.78 0.60 0.76 0.71 0.38 0.66
AN US
Ethanol Price Gasoline Price
0.62 0.66 0.61 0.66 0.72 0.66
ED
M
NOTES: Table presents retail prices for gasoline and ethanol, the ratio between the two prices and the share of stations in our sample that provide ethanol as the cheapest fuel, between 2002 and 2007. All prices are expressed in BRL real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). Sources: ANP.
PT
Table 1 presents the evolution of retail prices over time for each type of fuel. All prices are expressed in real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor
CE
Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). Ethanol average retail prices have been below the 70% threshold of gasoline average retail prices except in
AC
2006. Therefore, it is reasonable to infer that most owners of flex cars in our sample are choosing to fill their tanks with ethanol. Table A.2 in Appendix A provides other basic statistics on wholesale prices, retail prices and margins (i.e., the gap between wholesale and retail prices) in our data. Figure 3 shows the share of markets (defined here as a week-city pair) across time in which a single fuel is not the cheapest across all stations. We can see that there is no point in time when a single fuel is the cheapest across all markets. Hence, consumers looking for the cheapest fuel in a given market in our sample may end up choosing either ethanol or gasoline depending on which station they visit. This suggests that both fuels 16
17 6.9%
14.1%
6.7%
4.3%
4.6%
4.50%
PT
5.6%
6.9%
10.5%
7.4%
5.5%
4.4%
6.1%
5.3%
6.5%
ED
12.4%
5.5%
6.3%
7.3%
9.0%
4.3%
5.6%
4.4%
7.1%
5.3%
7.2%
11.1%
8.6%
10.5%
2.6%
CR IP T
AN US
6.5%
M 6.2%
4.2%
3.8%
2.3%
NOTES: Map shows the share of flex cars in the municipalities considered in our sample across the state of Rio de Janeiro in 2007. Sources: Detran RJ.
9.1%
10.0%
4.6%
> 10%
7% -10%
5% - 7%
2% - 5%
Percentage of Flex Cars 2007
CE
Figure 2: Geographic Distribution of Sample across Rio de Janeiro State.
AC
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT are being used as instruments for price competition between stations during the period we analyze.
CR IP T
.6
AN US
.4 0
.2
Share of Cities
.8
1
Figure 3: Competition between Gasoline and Ethanol over time
100
200
M
0
300
400
Week
PT
ED
NOTES: Figure plots the share of cities over time in our sample in which both gasoline and ethanol are the cheapest fuel in at least one station, i.e., the share of cities that do not have a single fuel as the cheapest across all stations. Sources: ANP.
We can also investigate evidence of price competition between gasoline and ethanol
CE
by computing a measure of within station rank-reversal, similar to the between-station rank-reversal measure presented in Chandra and Tappata (2011b).7 This measure is
AC
station specific and less prone to outliers issues when compared to our previous one. It is computed as the fraction of periods that a given station switches its cheapest fuel. For example, if a station uses gasoline as the cheapest fuel more often than ethanol, but in 10% of the periods it uses ethanol as its cheapest product, our within station rank-reversal variable would be equal to 0.1. If there is no price switching across time, this measure would be zero, and a value of 0.5 would imply that there is no fuel used more often as the cheapest in one particular station. 7
The authors use a panel data set on U.S. gasoline retailers to study how imperfect information shapes price dispersion.
18
ACCEPTED MANUSCRIPT Figure 4a presents the rank-reversal probability density function considering our full sample (kernel density with bandwidth of 0.01). Approximately 88% of the stations present a positive rank-reversal, and stations change its cheapest fuel more than 23% of the time on average (mean of 0.232 and standard deviation of 0.161). This is evidence that both fuels are instruments of price competition within a station. Figure 4a also shows a point mass at zero, meaning that 12 % of stations never switch
CR IP T
their cheapest fuel. These stations, however, do not appear as frequently as an average station in our sample and are likely to be new entrants that die fast or simply stations that are not sampled regularly. In fact, figure 4b shows that when we focus on stations that remain in our sample for more than 6 months (continuously or not), the mass at zero disappears.
AN US
Figure 4: Rank Reversal within Station (a) All Stations
.2
.3
.4
2 1.5
Probability Density Function
ED
.1
.5
0
PT
0
1
M
4 3 2 1
Probability Density Function
2.5
5
(b) Stations with more than 26 weeks
Fuel Rank Reversals
.1
.2
.3
.4
.5
Fuel Rank Reversals
AC
CE
NOTES: Figures plot the probability density function of within station rank-reversal between gasoline and ethanol (kernel density with bandwidth of 0.01). Panel A considers all stations and Panel B considers only stations with more than 26 weeks in our sample. Rank-reversal is computed as the fraction of periods that a given station switches its cheapest fuel. For example, if a station uses gasoline as the cheapest fuel more often than ethanol, but in 10% of the periods it uses ethanol as its cheapest product, our within station rank-reversal variable would be equal to 0.1 in this case. Mean of 0.232 and standard deviation of 0.161 in our sample.
To further investigate the competition between the two products, we compute a type of rank-reversal measure between stations, defined as the share of stations in a given municipality-week pair that did not choose the most likely cheapest fuel. For example, if most stations in a municipality-week pair chose ethanol as their cheapest product, our between station rank reversal measure in that market will be equal to the share of stations 19
ACCEPTED MANUSCRIPT that charge less for gasoline than for ethanol. If our measure is zero, it is unlikely that the two fuels are competing fiercely in that market. Figure A.1 in Appendix A shows the average (across markets) of this measure over time, presenting further evidence that ethanol and gasoline are indeed competitors in the fuel retail market. Table 2: Distributors’ Fuel Sales and size of Fleet in the State of Rio de Janeiro Gas sales per vehicle 0.76–0.76 0.64–0.64 0.64–0.64 0.57–0.59 0.52–0.55 0.48–0.53
Ethanol sales per vehicle 0.33–0.33 0.21–0.21 0.22–0.23 0.32–0.38 0.34–0.47 0.44–0.76
CR IP T
2002 2003 2004 2005 2006 2007
Gasoline Ethanol Gasoline Ethanol Flex Sales Sales Fleet Fleet Fleet 1,971,934 157,567 2,606,238 473,434 1,764,595 98,178 2,775,071 478,060 1,848,172 109,817 2,879,902 476,632 23,561 1,739,319 180,528 2,958,560 475,307 84,297 1,660,803 224,255 3,016,335 473,880 188,271 1,635,152 359,404 3,098,499 472,670 335,629
AN US
NOTES: Table shows distributors’ consolidated sales volume (cubic meters) in the state of Rio de Janeiro by fuel type between 2002 and 2007, as well as the total car fleet in the state by fuel type. In each of the last two columns, we provide a range of values for fuel sales per vehicle, depending on how one assigns the flex fleet in the denominator. The lower limit considers that all drivers of flex cars buy the corresponding fuel, while the upper limit assumes that flex drivers never do so.
To provide a sense of the relative size of the market of each fuel, Table 2 shows the
M
evolution of aggregate sales and the size of the fleets in the state of Rio de Janeiro. Even though we do not have access to data on gasoline and ethanol sales by station, ANP
ED
publishes the distributors’ consolidated sales volume in the state of Rio de Janeiro, as shown in columns 2 and 3. In this table, we also present the aggregate size of the fleet
PT
in our data set (columns 3-5), and compute sales per car, in cubic meters, in the last two columns. We provide a range of values for fuel sales per vehicle, depending on how one
CE
assigns the flex fleet in the denominator. The lower limit considers that all drivers of flex cars buy the corresponding fuel, while the upper limit assumes that flex drivers never do
AC
so. For example, in 2007, ethanol sales per vehicle ranged between 0.44 (if all drivers of flex cars bought only ethanol) and 0.76 (if all drivers of flex cars bought only gasoline) cubic meters.
5
Empirical Results
In this section, we document the effects of flex car penetration on the fuel market by testing two of the model predictions. First, we estimate the effect of the penetration of flex cars (θ) on retail prices for both ethanol and gasoline. Second, we investigate whether 20
ACCEPTED MANUSCRIPT θ affected the price spread between gasoline and ethanol. Finally, moving beyond the model we investigate the effect of flex car penetration on retail costs and margins. Our basic estimation equation is: t t t yim = δθm + ζ 0 Zim + µtim ,
(3)
t represents ethanol/gasoline prices8 in station i, municipality m and week where yim
CR IP T
t t t, and θm represents the share of flex cars in municipality m at time t. Zim is a vector
containing variables that are potentially important to determine prices at the local level, depending on each regression specification. It includes municipal GDP per capita, the number of stations per car, and the number of hotels per square km to control for the effect of income growth, variation in relative fuel station scarcity, and the intensity of local
AN US
tourist activity, respectively. It contains station brand fixed effects to capture pricing differences across distinct fuel suppliers (note that some stations change their brand over time, allowing us to identify the coefficients of these dummies even when we use station fixed effects in our specification). We also control for station fixed effects in order to
M
capture the within correlation between θ and our dependent variable (and not simply how levels of θ are correlated with price/margin levels). Finally, the vector includes
ED
monthly fixed effects (for example, January 2003) to account for shocks that may affect fuel prices in the Brazilian territory as a whole, and city-year linear time trends.
9
PT
The error term, µtim , represents unobserved components that affect prices in a given station. This term might be correlated with unobserved contemporaneous local market
CE
shocks that affect price decisions at the station level. Moreover, the decision of purchasing a flex car can be correlated with unobserved local price shocks even after accounting for time and/or market effects.
AC
In order to estimate the effect of flex car penetration on prices and margins controlling
for possible reverse causality bias, we adopt an instrumental variable (IV) strategy that builds on the literature on trade and labor markets (e.g. Bartik, 1991; Autor, Dorn and Hanson, 2013). Our IV is composed of two terms. The first one is the number of dealers from a particular brand that entered each municipality before 2000 and were still operating 8
We also use wholesale prices, retail margins or absolute price spreads as dependent variables - see Sub-sections 5.2 and 5.3 for details. 9 As a robustness check, we also control for fuel retail costs (or wholesale prices) at the station-time level (see Tables A.7 and A.8 in the Appendix). ANP provides information on fuel prices paid by stations to wholesalers based on a station’s most recent purchase invoice.
21
ACCEPTED MANUSCRIPT by the end of 2015: Fiat, Ford, General Motors, Honda, Mitsubishi, Nissan, PeugeotCitroen, Renault, Toyota and Volkswagen. (These were the only manufacturers that produced flex cars before early 2008.)10 This term will suffer from reverse causality if there are anticipation effects of post-2003 shocks that led to differential entry of dealers across municipalities before 2000. The second term is the share of flex car models offered by each vehicle manufacturer in the Brazilian territory.11 For example, if Ford offered 3
CR IP T
models to Brazilian consumers in January 2000 and only one of them was flex, our measure is equal to 1/3 for Ford at that point in time. This measure should be determined by the Brazilian market as a whole, and should not be influenced by shocks at the municipality level, i.e., local price shocks should not affect this measure (after we control for aggregate time effects and city year-trends).
AN US
We then compute the following IV at the municipality-month level: IVmt =
X
nkm θ˜kt ,
(4)
k
where nkm is the number of dealers of manufacturer k in city m before 2000 that were
territory in a particular month.
M
still active in 2015, and θ˜kt is the share of flex models offered by brand k in the Brazilian
ED
The idea behind the instrument is that drivers adopted flex cars faster in municipalities that were initially more exposed to brands that adopted flex engines more quickly. In fact,
t IVmt and θm .
PT
the first stage statistics presented later in this section indicate a high correlation between
CE
One potential source of concern with our instrument is that specific markets may be driving the sales of some car manufacturerers, i.e., a car brand may concentrate its sales
AC
in a few Brazilian municipalities in our sample. In that case, local (unobserved) shocks could affect our IV measure, as pointed out by Blanchard and Katz (1992) in a different framework. We do not believe that this is true in our set-up, given that Brazilian car sales are relatively spread out across the country and that the leading state in car sales 10
We obtain this information from the website of the union of automobile dealers of the state of Rio de Janeiro (SINCODIV): http://www.sincodiv-rj.com.br/. We start with a sample of 145 active local dealers on November of 2015 around the state of Rio de Janeiro. Then we are able to extract information on dealers’ date of entry from websites such as http://www.cnpjbrasil.com/. By selecting all the dealers that initiated activity before 01/01/2000, we are left with 59 dealers. 11 We obtain this information from the website of the national association of automobile vehicles manufacturers: http://www.anfavea.com.br/.
22
ACCEPTED MANUSCRIPT is not Rio de Janeiro but S˜ao Paulo.12 Hence, after controlling for local market shocks and aggregate time effects, the instrument should be uncorrelated with the error term in equation 3.
5.1
Effect of Flex Cars on Retail Prices
According to proposition 2, the penetration of flex cars tightens the competition in the
CR IP T
fuel market, decreasing gasoline and ethanol prices. Tables 3 and 4 present a reducedform analysis of the impact of flex cars on prices. Each column adopts a different set of controls. All the results presented in this and the following section adjust the price of gasoline by a factor of 0.7 so that prices per unit of energy are directly comparable across fuels.
AN US
Table 3 analyzes the effect of flex cars on the gasoline market. Column (1) suggests that, according to the prediction of the model, the estimated coefficient is negative. This specification accounts for station fixed effects and time dummies. The point estimate implies that a 10 percentage point increase in flex car penetration reduces the gasoline
M
price by 1.68/0.7 = 2.4 cents (of Brazilian Reais) per liter. We then use our IV strategy. When we compare the results in column (2) to the ones in column (1) - its OLS coun-
ED
terpart - we can see that the effect of θ is stronger, suggesting that our OLS coefficients are biased towards zero, possibly because (unobserved) local price shocks are positively correlated with the share of flex cars in this simpler specification without other controls.
PT
Our first stages are strong, as suggested by the Kleibergen-Paap statistics (significant at
CE
all reasonable levels) shown in the lower part of the panel. Considering flag/brand fixed effects in column (3) makes no significant difference (remember that brand changes across time within a station allow us to identify the coefficients of theses dummies). Considering
AC
other controls in column (4) reduces the impact to approximately zero cents per liter and the result becomes statistically insignificant. However, when we account for city-level linear time trends (year as time unit) in column (5), the coefficient becomes stronger and statistically significant. Table 4 presents the same analysis for the ethanol market. The effects on ethanol prices are similar but more pronounced. In our preferred specification in column (5), a 10 percentage point increase in flex car penetration reduces the ethanol price by 8.1 cents 12
Rio was the fourth state in sales in 2012, according to Infomoney (2012).
23
ACCEPTED MANUSCRIPT Table 3: Effect of Flex Car Penetration on Gasoline Price
-0.168*** (0.030)
GDP per Capita Stations/Cars Hotels per km2 KP F-Stat (1st Stage) Nclusters Observations Monthly Fixed Effects Station Fixed Effects Brand Fixed Effects City-Year Trends
2532 280746 Yes Yes No No
(3) 2SLS
(4) 2SLS
(5) 2SLS
Gasoline Price (pg ) -0.195*** -0.193*** -0.012 -0.195* (0.049) (0.048) (0.065) (0.108) 0.001*** -0.001*** (0.000) (0.000) -0.020*** -0.013*** (0.003) (0.003) -0.028 -0.021 (0.026) (0.021) 838 845 684 142 2532 2532 2406 2406 280746 280746 272190 272190 Yes Yes Yes Yes Yes Yes Yes Yes No Yes Yes Yes No No No Yes
CR IP T
% Flex Cars (θ)
(2) 2SLS
AN US
(1) OLS
PT
ED
M
NOTES: Table displays estimated effects of shares of flex cars, θ, on gasoline retail prices. Regressions consider a station i located in municipality m in week t as the unit of analysis, and the sample period goes from January 2002 to March 2008. All prices are expressed in BRL real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). The price of gasoline is multiplied by a factor of 0.7 so that prices per unit of energy are directly comparable across fuels. θ represents the share of flex cars in municipality m in a given month. Column 1 estimated by OLS and columns 2-5 by 2SLS. All columns include station and monthly fixed effects as controls. Columns 3-5 include station brand fixed effects. Columns 4 and 5 include yearly municipal GDP per capita, the monthly number of stations divided by the car fleet at the municipality level, and the yearly number of hotels per square kilometer in a municipality as controls. Column 5 includes city-level linear time trends (year as time unit). Instrument for share of flex cars, IVmt , is equal P to k nkm θ˜kt , where nkm is the number of dealers of car manufacturer k in city m before 2000 that were still active in 2015, and θ˜kt is the share of flex models offered by manufacturer k in the Brazilian territory in a particular month. Standard errors clustered by city-month in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
CE
(of Brazilian Reais) per liter.
We also perform some robustness analysis. In our model, competition in fuel markets
AC
increase with the share of flex cars as long as the same fuel is not the cheapest one across all stations in a given market. Hence, if this characteristic is not present across the markets we are analyzing it is possible that we are simply finding a spurious negative correlation between θ and our price measures. Figure 3 suggests that this is not the case. Nevertheless, we restrict our analysis to markets in which both gasoline and ethanol appear as the cheapest fuel in at least one station and present the results in Tables A.3 and A.4 in Appendix A. They show that the magnitude and sign of the coefficients remain basically unchanged.
24
ACCEPTED MANUSCRIPT Table 4: Effect of Flex Car penetration on Ethanol Price
-0.658*** (0.080)
GDP per Capita Stations/Cars Hotels per km2 KP F-Stat (1st Stage) Nclusters Observations Monthly Fixed Effects Station Fixed Effects Brand Fixed Effects City-Year Trends
2532 260960 Yes Yes No No
(3) 2SLS
(4) 2SLS
Ethanol Price (pa ) -0.750*** -0.753*** -0.755*** (0.145) (0.145) (0.188) -0.001 (0.001) -0.018*** (0.006) -0.034 (0.059) 852 862 722 2532 2532 2406 260960 260960 252630 Yes Yes Yes Yes Yes Yes No Yes Yes No No No
(5) 2SLS
-0.812** (0.336) -0.004*** (0.001) -0.017** (0.008) -0.061 (0.053) 144 2406 252630 Yes Yes Yes Yes
CR IP T
% Flex Cars (θ)
(2) 2SLS
AN US
(1) OLS
PT
ED
M
NOTES: Table displays estimated effects of shares of flex cars, θ, on ethanol retail prices. Regressions consider a station i located in municipality m in week t as the unit of analysis, and the sample period goes from January 2002 to March 2008. All prices are expressed in BRL real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). The price of gasoline is multiplied by a factor of 0.7 so that prices per unit of energy are directly comparable across fuels. θ represents the share of flex cars in municipality m in a given month. Column 1 estimated by OLS and columns 2-5 by 2SLS. All columns include station and monthly fixed effects as controls. Columns 3-5 include station brand fixed effects. Columns 4 and 5 include yearly municipal GDP per capita, the monthly number of stations divided by the car fleet at the municipality level, and the yearly number of hotels per square kilometer in a municipality as controls. Column 5 includes city-level linear time trends (year as time unit). Instrument for share of flex cars, IVmt , is equal P to k nkm θ˜kt , where nkm is the number of dealers of car manufacturer k in city m before 2000 that were still active in 2015, and θ˜kt is the share of flex models offered by manufacturer k in the Brazilian territory in a particular month. Standard errors clustered by city-month in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
CE
Our IV strategy assumes that manufacturers determine their supply of flex cars at the national level. However, large markets such as the city of Rio de Janeiro may have
AC
more influence than others in their decision. To verify the robustness of our results, we exclude the city of Rio from our sample and re-estimate equation 3. Tables A.5 and A.6 still show a negative effect of θ on prices. The effect of flex car penetration on gasoline
prices appears to be stronger. However, the effect on ethanol prices loses significance (and becomes positive) in our preferred specification in column (5). We have so far abstracted from the impact of retail fuel costs, which vary substantially by station and are important components of fuel retail prices. We also verify whether our findings are robust to the addition of wholesale prices as controls. Tables A.7 and A.8
25
ACCEPTED MANUSCRIPT present the results. Our results remain basically the same, with only minor shifts in magnitudes relative to Tables 3 and 4.
5.2
Effect of Flex Cars on the Spread between Gasoline and Ethanol Prices
We now investigate the empirical relationship between flex cars and the (absolute) spread
CR IP T
between gasoline and ethanol prices (pg − pa ). Our theoretical results in sub-section 3.3 suggest that price spreads are weakly increasing in the share of flex cars. We now investigate if the data supports this prediction by regressing the absolute value of the spread |pg − pa | on θ, controlling for different sets of fixed effects and other variables. The results are presented in Table 5. The specifications are similar to the ones pre-
AN US
sented in our tables in sub-section 5.1. The coefficient on the penetration of flex cars is positive and statistically significant from columns (1) to (4). However, when we include city-year linear trends, the result becomes negative and loses statistical significance.13
tration on |pg − pa |.
Additional Results
ED
5.3
M
Hence, our linear regressions show mixed evidence of a positive impact of flex car pene-
In this sub-section, we first study the robustness of our main results to a different de-
PT
pendent variable, retail margins. Panel A of Table 6 presents the results for the gasoline market. It shows a negative relationship between θ and gasoline margins in all specifi-
CE
cations. Once again, when we compare columns (1) and (2) we can see that the result becomes stronger after we implement our IV strategy. In column (5), our preferred speci-
AC
fication with the full set of controls, a 10 percentage point increase in flex car penetration reduces gasoline margins by 2.33/0.7 = 3.32 cents (of Brazilian Reais) per litre. The magnitude of the coefficient of θ in column (5), Panel A, is similar to the one
13
As in the gasoline case and differently from the ethanol one (see columns 4 and 5 of Tables 3 and 4), the switch of sign in the last column of Table 5 suggests that there are important local components affecting the price spread between the two fuels. Noting that gasoline prices are on average higher than ethanol ones during our period of analysis, the change of sign may be due to local unobserved shocks that are (positively) correlated with gasoline prices but uncorrelated with ethanol ones, and that bias the coefficients upward in the first four columns of the table (but not in column 5). Any local demand trend that affects solely gasoline markets will generate this effect. For example, gasoline car owners may be richer (as ethanol cars are considerably older) and more likely to experience positive income shocks (possibly heterogeneous across cities) when compared to ethanol car owners.
26
ACCEPTED MANUSCRIPT
Table 5: Effect of Flex Car Penetration on Spread between Gasoline and Ethanol Prices
Stations/Cars Hotels per km2
(5) 2SLS
Absolute Price Spread (|pg − pa |) 0.216*** 0.357*** 0.374*** 0.368*** -0.254 (0.065) (0.104) (0.104) (0.140) (0.305) -0.001 -0.001 (0.001) (0.001) -0.008* -0.020*** (0.005) (0.007) -0.075 -0.053 (0.047) (0.050) 854 864 723 144 2532 2532 2532 2406 2406 260383 260383 260383 252076 252076 Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes No No Yes Yes Yes No No No No Yes
ED
KP F-Stat (1st Stage) Nclusters Observations Monthly Fixed Effects Station Fixed Effects Brand Fixed Effects City-Year Trends
(4) 2SLS
AN US
GDP per Capita
(3) 2SLS
M
% Flex Cars (θ)
(2) 2SLS
CR IP T
(1) OLS
AC
CE
PT
NOTES: Table displays estimated effects of shares of flex cars, θ, on absolute retail price spreads (retail gasoline prices minus retail ethanol prices). Regressions consider a station i located in municipality m in week t as the unit of analysis, and the sample period goes from January 2002 to March 2008. All prices are expressed in BRL real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). The price of gasoline is multiplied by a factor of 0.7 so that prices per unit of energy are directly comparable across fuels. θ represents the share of flex cars in municipality m in a given month. Column 1 estimated by OLS and columns 2-5 by 2SLS. All columns include station and monthly fixed effects as controls. Columns 3-5 include station brand fixed effects. Columns 4 and 5 include yearly municipal GDP per capita, the monthly number of stations divided by the car fleet at the municipality level, and the yearly number of hotels per square kilometer in a municipality as controls. Column 5 includes P city-level linear time trends (year as time unit). Instrument for share of flex cars, IVmt , is equal to k nkm θ˜kt , where nkm is the number of dealers of car manufacturer k in city m before 2000 that were still active in 2015, and θ˜kt is the share of flex models offered by manufacturer k in the Brazilian territory in a particular month. Standard errors clustered by city-month in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
27
ACCEPTED MANUSCRIPT
Table 6: Effect of Flex Car Penetration on Gasoline Margins and Costs
-0.197*** (0.027)
GDP per Capita Stations/Cars Hotels per km2 KP F-Stat (1st Stage) Nclusters Observations Panel B % Flex Cars (θ)
2528 195457
-0.021 (0.015)
GDP per Capita
Hotels per km2
PT
ED
KP F-Stat (1st Stage) Nclusters Observations Monthly Fixed Effects Station Fixed Effects Brand Fixed Effects City-Year Trends
2528 195457 Yes Yes No No
(4) 2SLS
(5) 2SLS
Gasoline Margin (pg − cg ) -0.237*** -0.238*** -0.147** -0.233** (0.048) (0.048) (0.067) (0.108) 0.001*** -0.001*** (0.000) (0.000) -0.011*** -0.006** (0.003) (0.003) -0.022 -0.039 (0.028) (0.027) 796 804 587 150 2528 2528 2402 2402 195457 195457 189894 189894 Gasoline Cost (cg ) 0.031 0.032 0.114*** (0.028) (0.028) (0.038) 0.001*** (0.000) -0.008*** (0.002) 0.002 (0.017) 796 804 587 2528 2528 2402 195457 195457 189894 Yes Yes Yes Yes Yes Yes No Yes Yes No No No
M
Stations/Cars
(3) 2SLS
CR IP T
Panel A % Flex Cars (θ)
(2) 2SLS
AN US
(1) OLS
0.009 (0.072) -0.000 (0.000) -0.008*** (0.002) 0.013 (0.018) 150 2402 189894 Yes Yes Yes Yes
AC
CE
NOTES: Table displays estimated effects of shares of flex cars, θ, on gasoline margins (retail minus wholesale prices - Panel A) and wholesale prices (Panel B). Regressions consider a station i located in municipality m in week t as the unit of analysis, and the sample period goes from January 2002 to March 2008. All prices are expressed in BRL real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). The price of gasoline is multiplied by a factor of 0.7 so that prices per unit of energy are directly comparable across fuels. θ represents the share of flex cars in municipality m in a given month. Column 1 estimated by OLS and columns 2-5 by 2SLS. All columns include station and monthly fixed effects as controls. Columns 3-5 include station brand fixed effects. Columns 4 and 5 include yearly municipal GDP per capita, the monthly number of stations divided by the car fleet at the municipality level, and the yearly number of hotels per square kilometer in a municipality as controls. Column 5 includes Pcity-level linear time trends (year as time unit). Instrument for share of flex cars, IVmt , is equal to k nkm θ˜kt , where nkm is the number of dealers of car manufacturer k in city m before 2000 that were still active in 2015, and θ˜kt is the share of flex models offered by manufacturer k in the Brazilian territory in a particular month. Standard errors clustered by city-month in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
28
ACCEPTED MANUSCRIPT
Table 7: Effect of Flex Car Penetration on Ethanol Margins and Costs
-0.265*** (0.081)
GDP per Capita Stations/Cars Hotels per km2 KP F-Stat (1st Stage) Nclusters Observations Panel B % Flex Cars (θ)
2524 151615
-0.477*** (0.069)
GDP per Capita
Hotels per km2
PT
ED
KP F-Stat (1st Stage) Nclusters Observations Monthly Fixed Effects Station Fixed Effects Brand Fixed Effects City-Year Trends
2524 151615 Yes Yes No No
(4) 2SLS
(5) 2SLS
Ethanol Margin (pa − ca ) -0.908*** -0.912*** -1.143*** -0.529* (0.113) (0.113) (0.153) (0.276) -0.001 0.000 (0.001) (0.001) 0.028*** 0.021** (0.006) (0.009) -0.057 -0.151*** (0.045) (0.047) 756 766 568 158 2524 2524 2398 2398 151615 151615 146789 146789 Ethanol Cost (ca ) -0.008 -0.009 0.150 (0.120) (0.120) (0.164) -0.001 (0.001) -0.041*** (0.006) 0.033 (0.056) 756 766 568 2524 2524 2398 151615 151615 146789 Yes Yes Yes Yes Yes Yes No Yes Yes No No No
M
Stations/Cars
(3) 2SLS
CR IP T
Panel A % Flex Cars (θ)
(2) 2SLS
AN US
(1) OLS
-0.624** (0.262) -0.003*** (0.001) -0.042*** (0.010) 0.077 (0.061) 158 2398 146789 Yes Yes Yes Yes
AC
CE
NOTES: Table displays estimated effects of shares of flex cars, θ, on ethanol margins (retail minus wholesale prices - Panel A) and wholesale prices (Panel B). Regressions consider a station i located in municipality m in week t as the unit of analysis, and the sample period goes from January 2002 to March 2008. All prices are expressed in BRL real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). The price of gasoline is multiplied by a factor of 0.7 so that prices per unit of energy are directly comparable across fuels. θ represents the share of flex cars in municipality m in a given month. Column 1 estimated by OLS and columns 2-5 by 2SLS. All columns include station and monthly fixed effects as controls. Columns 3-5 include station brand fixed effects. Columns 4 and 5 include yearly municipal GDP per capita, the monthly number of stations divided by the car fleet at the municipality level, and the yearly number of hotels per square kilometer in a municipality as controls. Column 5 includes Pcity-level linear time trends (year as time unit). Instrument for share of flex cars, IVmt , is equal to k nkm θ˜kt , where nkm is the number of dealers of car manufacturer k in city m before 2000 that were still active in 2015, and θ˜kt is the share of flex models offered by manufacturer k in the Brazilian territory in a particular month. Standard errors clustered by city-month in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
29
ACCEPTED MANUSCRIPT found in column (5) of Table 3. This is evidence that flex cars do not affect gasoline wholesale prices substantially. We further investigate this issue by regressing gasoline wholesale prices on our variable of interest (and on other controls). The results are shown in Panel B of Table 6. In fact, flex cars penetration does not seem to impact gasoline costs, as shown by the non-statistically significant coefficients of flex car penetration in most of our specifications, including our preferred one in column (5). The absence of
CR IP T
effect is not surprising, since gasoline wholesale prices are settled by international market makers and/or Petrobras.
Table 7 presents the same exercise for the ethanol market. We still find a negative and significant effect of θ on ethanol retail margins. However, we find some evidence of the effect of flex cars on ethanol wholesale prices. In Panel B, most of the specifications show
AN US
statistically insignificant coefficients (with different signs), but our preferred specification shows a negative and significant effect. This may indicate that ethanol producer prices in Brazil are more sensitive to changes in the domestic demand for this product (when compared to gasoline).
Concluding Remarks
M
6
ED
In this work, we investigate how the penetration of flex cars has affected the fuel retail market in the state of Rio de Janeiro. Our main hypothesis was that flex car penetra-
PT
tion has increased the degree of substitution between gasoline and ethanol and that fuel stations would respond strategically to this shock, reducing retail prices in equilibrium.
CE
Our estimates suggest that the model prediction is correct. A 10 percentage point increase in the share of flex cars decrease ethanol and gasoline energy equivalent prices
AC
per liter by approximately 8 and 3 cents, respectively. Considering the volume of sales and the size of the flex car fleet in 2007, a rough estimate suggests consumer savings of almost 80 million Reais in the Rio de Janeiro state that year. Our estimates also show that the absolute retail price gap between the two fuels has increased with the adoption of flex cars, also as predicted by the model. Our results show that new technologies that increase consumer choice may benefit even those consumers who choose not to adopt them. This is an important potential welfare-improving channel to be taken into consideration by policy makers who aim to
30
ACCEPTED MANUSCRIPT incentivize the use of new technologies, especially when considering dissemination of alternative automobile fuels, such as hydrogen and electricity.
References Anderson, Soren T. 2012. “The demand for ethanol as a gasoline substitute.” Journal
CR IP T
of Environmental Economics and Management, 63(2): 151 – 168. ANP. 2010. “Anu´ario Estat´ıstico Brasileiro do Petr´oleo e do G´as Natural 2010.” http://www.anp.gov.br/?dw=33213, accessed on October 20th, 2010.
Autor, David H., David Dorn, and Gordon H. Hanson. 2013. “The China Syn-
AN US
drome: Local Labor Market Effects of Import Competition in the United States.” American Economic Review, 103(6): 2121–68.
Bartik, Timothy J. 1991. Who Benefits from State and Local Economic Development Policies? Books from Upjohn Press, W.E. Upjohn Institute for Employment Research.
M
Blanchard, Olivier Jean, and Lawrence F. Katz. 1992. “Regional Evolutions.”
ED
Brookings Papers on Economic Activity, 23(1): 1–76. Boff, H. P. 2011. “Modeling the Brazilian Ethanol Market: How Flex-Fuel Vehicles are
PT
Shaping the Long Run Equilibrium .” China-USA Business Review, 10(4): 245–264. Bresnahan, Timothy F. 1982. “The oligopoly solution concept is identified.” Economics
CE
Letters, 10(1): 87–92.
Busse, Meghan R, Christopher R Knittel, and Florian Zettelmeyer. 2013. “Are
AC
consumers myopic? Evidence from new and used car purchases.” The American Economic Review, 103(1): 220–256.
Byrne, David P, and Nicolas de Roos. 2017. “Consumer search in retail gasoline markets.” The Journal of Industrial Economics, 65(1): 183–193. CADE. 2014. “Varejo de Gasolina.” Cadernos do Cade. Carlson, J.A., and R.P. McAfee. 1983. “Discrete equilibrium price dispersion.” The Journal of Political Economy, 91(3): 480–493. 31
ACCEPTED MANUSCRIPT Chandra, Ambarish, and Mariano Tappata. 2011a. “Consumer search and dynamic price dispersion: an application to gasoline markets.” The RAND Journal of Economics, 42(4): 681–704. Chandra, Ambarish, and Mariano Tappata. 2011b. “Consumer search and dynamic price dispersion: an application to gasoline markets.” The RAND Journal of Economics,
CR IP T
42(4): 681–704. Corts, Kenneth S. 2010. “Building out alternative fuel retail infrastructure: Government fleet spillovers in {E85}.” Journal of Environmental Economics and Management, 59(3): 219 – 234.
Costa, Francisco, Jason Garred, and Jo˜ ao Paulo Pessoa. 2016. “Winners and
AN US
losers from a commodities-for-manufactures trade boom.” Journal of International Economics, 102: 50 – 69.
Ferreira, A. L., F. P. de A. Prado, and J. J. da Silveira. 2009. “Flex cars and the
M
alcohol price.” Energy Economics, 31(3): 382–394.
Firgo, Matthias, Dieter Pennerstorfer, and Christoph R Weiss. 2015. “Centrality
ED
and pricing in spatially differentiated markets: The case of gasoline.” International Journal of Industrial Organization, 40: 81–90.
PT
Goldberg, Pinelopi Koujianou. 1998. “The effects of the corporate average fuel efficiency standards in the US.” The Journal of Industrial Economics, 46(1): 1–33.
CE
Hosken, Daniel S, Robert S McMillan, and Christopher T Taylor. 2008. “Retail gasoline pricing: What do we know?” International Journal of Industrial Organization,
AC
26(6): 1425–1436.
Infomoney. a
de
2012.
Minas,
“Venda
Paran´a
e
de
carros
Rio
juntas.”
no
estado
[Accessed:
de
SP 2017
´e 11
maior 30],
que http:
//www.infomoney.com.br/minhas-financas/carros/noticia/2557745/ venda-carros-estado-maior-que-minas-parana-rio-juntas. Kahn, James A. 1986. “Gasoline prices and the used automobile market: a rational expectations asset price approach.” The Quarterly Journal of Economics, 323–339.
32
ACCEPTED MANUSCRIPT Klier, line
Thomas, and
Data.”
New
and Vehicle
American
Joshua Fuel
Economic
Linn.
Economy: Journal:
2010.
“The
Evidence Economic
Price
from
of
Gaso-
Monthly
Policy,
2:
Sales
134–153.
http://www.aeaweb.org/articles.php?doi=10.1257/pol.2.3.134. Lewis, Matthew. 2008. “Price Dispersion And Competition With Differentiated Sell-
CR IP T
ers.” The Journal of Industrial Economics, 56(3): 654–678. Li, Shanjun, Christopher Timmins, and Roger H von Haefen. 2009. “How do gasoline prices affect fleet fuel economy?” American Economic Journal: Economic Policy, 1(2): 113–137.
´ Marjotta-Maistro, M. C., and G. A. Asai. 2006. “Alcool combust´ıvel: do carro a
AN US
a´lcool ao carro flex.” mimeo.
Pakes, Ariel, Steven Berry, and James A Levinsohn. 1993. “Applications and limitations of some recent advances in empirical industrial organization: price indexes and the analysis of environmental change.” The American Economic Review, 83(2): 240–
M
246.
ED
Pennerstorfer, Dieter. 2009. “Spatial price competition in retail gasoline markets: evidence from Austria.” The Annals of Regional Science, 43(1): 133–158.
PT
Pennerstorfer, Dieter, Philipp Schmidt-Dengler, Nicolas Schutz, Christoph Weiss, and Biliana Yontcheva. 2015. “Information and price dispersion: Theory
CE
and evidence.” WIFO Working paper 502/2015. Pinkse, J., M.E. Slade, and C. Brett. 2002. “Spatial price competition: A semipara-
AC
metric approach.” Econometrica, 70(3): 1111–1153.
Salvo, Alberto, and Cristian Huse. 2011. “Is Arbitrage Tying the Price of Ethanol to that of Gasoline? Evidence from the Uptake of Flexible-Fuel Technology.” The Energy Journal, Volume 32(Number 3): 119–148. Salvo, Alberto, and Cristian Huse. 2013. “Build it, but will they come? Evidence from consumer choice between gasoline and sugarcane ethanol.” Journal of Environmental Economics and Management, 66(2): 251 – 279.
33
ACCEPTED MANUSCRIPT Shriver, Scott K. 2015. “Network Effects in Alternative Fuel Adoption: Empirical Analysis of the Market for Ethanol.” Marketing Science, 34(1): 78–97. Simon, Leo, and William Zame. 1990. “Discontinuous Games and Endogenous Sharing Rules.” Econometrica, 58: 861–872. USDA.
2010.
“Sugar
and
Sweeteners
accessed
Tables.” on
Octo-
CR IP T
http://www.ers.usda.gov/Briefing/Sugar/data.htm#yearbook,
Yearbook
ber 20th, 2010.
Verboven, Frank. 2002. “Quality-based price discrimination and tax incidence: evidence from gasoline and diesel cars.” RAND Journal of Economics, 275–297.
AN US
Weyl, Eric G., and Michal Fabinger. 2009. “Pass-Through as an Economic Tool.”
AC
CE
PT
ED
M
SSRN eLibrary.
34
ACCEPTED MANUSCRIPT
Appendix A -
Other Figures and Tables
CR IP T
.3 .2
AN US
.1 0
Rank Reversal across Stations
.4
Figure A.1: Rank Reversal between Stations
0
100
200
300
400
M
Week
AC
CE
PT
ED
NOTES: Figure plots the average (across markets) of rank-reversal between stations over time. Rankreversal between stations is computed as the share of stations in a given municipality-week pair that did not choose the most likely cheapest fuel. For example, if most stations in a municipality-week pair chose ethanol as their cheapest product, our between station rank reversal measure in that market will be equal to the share of stations that charge less for gasoline than for ethanol. Sources: ANP.
35
ACCEPTED MANUSCRIPT
Table A.1: Stock of Vehicles, Percentage of Flex Cars and Number of Fuel Stations, by City
AC
CE
% flex
CR IP T
14.1% 12.4% 11.1% 10.5% 10.5% 10.0% 9.1% 9.0% 8.9% 8.6% 7.4% 7.3% 7.2% 7.1% 6.9% 6.9% 6.7% 6.5% 6.5% 6.3% 6.2% 6.1% 5.6% 5.6% 5.5% 5.5% 5.3% 5.3% 4.6% 4.5% 4.4% 4.4% 4.3% 4.3% 4.2% 3.8% 2.6% 2.3% 8.8%
Number of fuel stations In the Total % sampled sample (2010) weekly 10 11 91% 34 85 40% 14 23 61% 190 805 24% 7 7 100% 13 23 57% 7 11 64% 11 22 50% 14 26 54% 10 23 43% 9 10 90% 10 15 67% 16 23 70% 9 13 69% 21 31 68% 6 10 60% 21 30 70% 7 7 100% 35 88 40% 18 54 33% 10 13 77% 11 19 58% 23 31 74% 7 7 100% 25 65 38% 10 13 77% 30 88 34% 21 38 55% 8 13 62% 10 15 67% 20 33 61% 11 17 65% 20 34 59% 10 16 63% 10 16 63% 16 20 80% 38 109 35% 8 10 80% 750 1874 40%
AN US
M
ED
PT
Mangaratiba Niter´oi Maca´e Rio de Janeiro A. de B´ uzios Angra dos Reis Parati Maric´a Resende Cabo Frio Nil´opolis Trˆes Rios Araruama Saquarema Barra Mansa Vassouras Volta Redonda Para´ıba do Sul S˜ao Gon¸calo Petr´opolis Sapucaia Belford Roxo Teres´opolis Queimados Nova Igua¸cu Mag´e Duque de Caxias Nova Friburgo Itagua´ı Valen¸ca S. J. de Meriti Rio Bonito Itabora´ı Barra do Pira´ı S. Ant de Padua Itaperuna Campos dos Goit. S. Fr. de Itabapoana Total
2004 2007 Number of % flex Number of vehicles vehicles 3 090 1.2% 4 812 176 647 1.0% 194 511 41 048 1.1% 57 261 1 800 614 0.8% 1 969 128 4 340 1.4% 6 703 21 503 0.9% 27 030 2 937 0.6% 3 675 14 193 0.7% 24 120 26 085 0.8% 33 368 32 315 1.0% 46 371 24 305 0.5% 29 579 14 792 0.9% 18 246 20 290 0.6% 28 037 9 955 0.6% 13 771 31 788 0.8% 36 420 7 297 1.0% 8 319 68 441 0.7% 81 203 6 101 0.8% 7 427 107 977 0.4% 134 860 86 912 0.6% 97 177 1 381 0.9% 1 508 26 574 0.3% 37 744 42 920 0.5% 50 787 9 278 0.3% 13 612 115 523 0.4% 137 337 21 034 0.4% 27 495 128 426 0.3% 151 257 60 539 0.4% 68 588 28 016 0.2% 31 609 10 297 0.3% 11 540 64 607 0.2% 76 891 22 251 0.5% 35 453 27 800 0.3% 37 075 18 436 0.5% 21 239 9 234 0.7% 10 540 18 810 0.6% 22 123 91 651 0.3% 109 388 3 395 0.3% 4 204 3 200 802 0.7% 3 670 408
NOTES: Table shows stocks and percentages of flex cars by municipality in the state of Rio de Janeiro, in 2004 and in 2007. It also shows the total number of stations by city in 2010, the average number of stations by municipality in a week in our sample and the ratio between the two measures, i.e., the average percentage of stations sampled weekly by municipality. Sources: Detran RJ and ANP.
36
ACCEPTED MANUSCRIPT
Table A.2: Fuel Wholesale and Retail Prices and Margins (in R$) Ethanol Price Ethanol Wholesale Price Ethanol Margin Gasoline Price Gasoline Wholesale Price Gasoline Margin
Observations 265988 154158 154158 286077 198829 198829
Mean 1.144 0.952 0.193 1.734 1.537 0.198
St. Deviation 0.189 0.202 0.105 0.125 0.095 0.078
Min 0.529 0.223 -0.775 1.287 0.855 -0.557
Max 2.035 2.126 0.817 2.287 2.277 1.073
CR IP T
NOTES: Table shows usual descriptive statistics for fuel wholesale and retail prices and margins in our sample. All prices are expressed in BRL real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). Sources: ANP.
Table A.3: Effect of Flex Car Penetration on Gasoline Price - Competition
% Flex Cars (θ)
-0.147*** (0.031)
GDP per Capita
ED
Hotels per km2
CE
PT
KP F-Stat (1st Stage) Nclusters Observations Monthly Fixed Effects Station Fixed Effects Brand Fixed Effects City-Year Trends
2055 238858 Yes Yes No No
(3) 2SLS
(4) 2SLS
Gasoline Price (pg ) -0.223*** -0.223*** -0.014 (0.048) (0.048) (0.065) 0.002*** (0.000) -0.020*** (0.003) -0.033 (0.026) 809 815 614 2055 2055 1974 238858 238858 232940 Yes Yes Yes Yes Yes Yes No Yes Yes No No No
M
Stations/Cars
(2) 2SLS
AN US
(1) OLS
(5) 2SLS
-0.231** (0.113) -0.001** (0.000) -0.015*** (0.003) -0.002 (0.022) 146 1974 232940 Yes Yes Yes Yes
AC
NOTES: Table displays estimated effects of shares of flex cars, θ, on gasoline retail prices excluding from the sample city-week pairs where either gasoline or ethanol is the cheapest fuel in all stations. Regressions consider a station i located in municipality m in week t as the unit of analysis, and the sample period goes from January 2002 to March 2008. All prices are expressed in BRL real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). The price of gasoline is multiplied by a factor of 0.7 so that prices per unit of energy are directly comparable across fuels. θ represents the share of flex cars in municipality m in a given month. Column 1 estimated by OLS and columns 2-5 by 2SLS. All columns include station and monthly fixed effects as controls. Columns 3-5 include station brand fixed effects. Columns 4 and 5 include yearly municipal GDP per capita, the monthly number of stations divided by the car fleet at the municipality level, and the yearly number of hotels per square kilometer in a municipality as controls. Column 5 includes city-level linear time trends (year as time unit). Instrument for share of flex cars, IVmt , is equal P to k nkm θ˜kt , where nkm is the number of dealers of car manufacturer k in city m before 2000 that were still active in 2015, and θ˜kt is the share of flex models offered by manufacturer k in the Brazilian territory in a particular month. Standard errors clustered by city-month in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
37
ACCEPTED MANUSCRIPT
Table A.4: Effect of Flex Car Penetration on Ethanol Price - Competition
-0.601*** (0.085)
GDP per Capita Stations/Cars Hotels per km2
M
2055 219741 Yes Yes No No
(4) 2SLS
Ethanol Price (pa ) -0.676*** -0.687*** -0.645*** (0.147) (0.147) (0.200) -0.000 (0.001) -0.012* (0.006) -0.038 (0.064) 836 844 650 2055 2055 1974 219741 219741 214048 Yes Yes Yes Yes Yes Yes No Yes Yes No No No
ED
KP F-Stat (1st Stage) Nclusters Observations Monthly Fixed Effects Station Fixed Effects Brand Fixed Effects City-Year Trends
(3) 2SLS
(5) 2SLS
CR IP T
% Flex Cars (θ)
(2) 2SLS
AN US
(1) OLS
-0.906*** (0.344) -0.003*** (0.001) -0.014 (0.009) -0.019 (0.057) 148 1974 214048 Yes Yes Yes Yes
AC
CE
PT
NOTES: Table displays estimated effects of shares of flex cars, θ, on ethanol retail prices excluding from the sample city-week pairs where either gasoline or ethanol is the cheapest fuel in all stations. Regressions consider a station i located in municipality m in week t as the unit of analysis, and the sample period goes from January 2002 to March 2008. All prices are expressed in BRL real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). The price of gasoline is multiplied by a factor of 0.7 so that prices per unit of energy are directly comparable across fuels. θ represents the share of flex cars in municipality m in a given month. Column 1 estimated by OLS and columns 2-5 by 2SLS. All columns include station and monthly fixed effects as controls. Columns 3-5 include station brand fixed effects. Columns 4 and 5 include yearly municipal GDP per capita, the monthly number of stations divided by the car fleet at the municipality level, and the yearly number of hotels per square kilometer in a municipality as controls. Column 5 includes city-level linear time trends (year as time unit). Instrument for share of flex cars, IVmt , is equal P to k nkm θ˜kt , where nkm is the number of dealers of car manufacturer k in city m before 2000 that were still active in 2015, and θ˜kt is the share of flex models offered by manufacturer k in the Brazilian territory in a particular month. Standard errors clustered by city-month in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
38
ACCEPTED MANUSCRIPT
Table A.5: Effect of Flex Car Penetration on Gasoline Price - State Capital
-0.162*** (0.039)
GDP per Capita Stations/Cars Hotels per km2
(5) 2SLS
M
2458 191123 Yes Yes No No
(4) 2SLS
Gasoline Price (pg ) -0.437*** -0.442*** -0.365*** -0.602*** (0.119) (0.118) (0.130) (0.233) 0.001*** -0.002*** (0.000) (0.000) -0.021*** -0.014*** (0.003) (0.003) -0.005 -0.047** (0.024) (0.023) 138 142 116 75 2458 2458 2335 2335 191123 191123 184455 184455 Yes Yes Yes Yes Yes Yes Yes Yes No Yes Yes Yes No No No Yes
ED
KP F-Stat (1st Stage) Nclusters Observations Monthly Fixed Effects Station Fixed Effects Brand Fixed Effects City-Year Trends
(3) 2SLS
CR IP T
% Flex Cars (θ)
(2) 2SLS
AN US
(1) OLS
AC
CE
PT
NOTES: Table displays estimated effects of shares of flex cars, θ, on gasoline retail prices excluding from the sample the city of Rio de Janeiro (the capital of the state of RJ). Regressions consider a station i located in municipality m in week t as the unit of analysis, and the sample period goes from January 2002 to March 2008. All prices are expressed in BRL real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). The price of gasoline is multiplied by a factor of 0.7 so that prices per unit of energy are directly comparable across fuels. θ represents the share of flex cars in municipality m in a given month. Column 1 estimated by OLS and columns 2-5 by 2SLS. All columns include station and monthly fixed effects as controls. Columns 3-5 include station brand fixed effects. Columns 4 and 5 include yearly municipal GDP per capita, the monthly number of stations divided by the car fleet at the municipality level, and the yearly number of hotels per square kilometer in a municipality as controls. Column 5 includes Pcity-level linear time trends (year as time unit). Instrument for share of flex cars, IVmt , is equal to k nkm θ˜kt , where nkm is the number of dealers of car manufacturer k in city m before 2000 that were still active in 2015, and θ˜kt is the share of flex models offered by manufacturer k in the Brazilian territory in a particular month. Standard errors clustered by city-month in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
39
ACCEPTED MANUSCRIPT
Table A.6: Effect of Flex Car Penetration on Ethanol Price - State Capital
-0.479*** (0.093)
GDP per Capita Stations/Cars Hotels per km2
M
2458 180090 Yes Yes No No
(4) 2SLS
Ethanol Price (pa ) -0.717*** -0.730*** -0.650*** (0.224) (0.222) (0.249) -0.001 (0.001) -0.023*** (0.007) 0.066 (0.061) 156 160 132 2458 2458 2335 180090 180090 173589 Yes Yes Yes Yes Yes Yes No Yes Yes No No No
ED
KP F-Stat (1st Stage) Nclusters Observations Monthly Fixed Effects Station Fixed Effects Brand Fixed Effects City-Year Trends
(3) 2SLS
(5) 2SLS
CR IP T
% Flex Cars (θ)
(2) 2SLS
AN US
(1) OLS
0.243 (0.573) -0.004*** (0.001) -0.021** (0.009) 0.023 (0.062) 78 2335 173589 Yes Yes Yes Yes
AC
CE
PT
NOTES: Table displays estimated effects of shares of flex cars, θ, on ethanol retail prices excluding from the sample the city of Rio de Janeiro (the capital of the state of RJ). Regressions consider a station i located in municipality m in week t as the unit of analysis, and the sample period goes from January 2002 to March 2008. All prices are expressed in BRL real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). The price of gasoline is multiplied by a factor of 0.7 so that prices per unit of energy are directly comparable across fuels. θ represents the share of flex cars in municipality m in a given month. Column 1 estimated by OLS and columns 2-5 by 2SLS. All columns include station and monthly fixed effects as controls. Columns 3-5 include station brand fixed effects. Columns 4 and 5 include yearly municipal GDP per capita, the monthly number of stations divided by the car fleet at the municipality level, and the yearly number of hotels per square kilometer in a municipality as controls. Column 5 includes Pcity-level linear time trends (year as time unit). Instrument for share of flex cars, IVmt , is equal to k nkm θ˜kt , where nkm is the number of dealers of car manufacturer k in city m before 2000 that were still active in 2015, and θ˜kt is the share of flex models offered by manufacturer k in the Brazilian territory in a particular month. Standard errors clustered by city-month in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
40
ACCEPTED MANUSCRIPT
Table A.7: Effect of Flex Car Penetration on Gasoline Price - Controlling for Costs
Gasoline Cost (cg )
-0.211*** (0.028) 0.345*** (0.011)
GDP per Capita Stations/Cars Hotels per km2
(4) 2SLS
(5) 2SLS
Gasoline Price (pg ) -0.217*** -0.217*** -0.072 -0.227** (0.046) (0.046) (0.065) (0.103) 0.345*** 0.343*** 0.342*** 0.323*** (0.011) (0.011) (0.011) (0.011) 0.001*** -0.001*** (0.000) (0.000) -0.016*** -0.012*** (0.003) (0.003) -0.020 -0.031 (0.025) (0.021) 797 805 588 150 2528 2528 2402 2402 195457 195457 189894 189894 Yes Yes Yes Yes Yes Yes Yes Yes No Yes Yes Yes No No No Yes
M
2528 195457 Yes Yes No No
ED
KP F-Stat (1st Stage) Nclusters Observations Monthly Fixed Effects Station Fixed Effects Brand Fixed Effects City-Year Trends
(3) 2SLS
CR IP T
% Flex Cars (θ)
(2) 2SLS
AN US
(1) OLS
AC
CE
PT
NOTES: Table displays estimated effects of shares of flex cars, θ, on gasoline retail prices. Regressions consider a station i located in municipality m in week t as the unit of analysis, and the sample period goes from January 2002 to March 2008. All prices are expressed in BRL real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). The price of gasoline is multiplied by a factor of 0.7 so that prices per unit of energy are directly comparable across fuels. θ represents the share of flex cars in municipality m in a given month. Column 1 estimated by OLS and columns 2-5 by 2SLS. All columns include gasoline wholesale prices, station and monthly fixed effects as controls. Columns 3-5 include station brand fixed effects. Columns 4 and 5 include yearly municipal GDP per capita, the monthly number of stations divided by the car fleet at the municipality level, and the yearly number of hotels per square kilometer in a municipality as controls. Column 5 includes P city-level linear time trends (year as time unit). Instrument for share of flex cars, IVmt , is equal to k nkm θ˜kt , where nkm is the number of dealers of car manufacturer k in city m before 2000 that were still active in 2015, and θ˜kt is the share of flex models offered by manufacturer k in the Brazilian territory in a particular month. Standard errors clustered by city-month in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
41
ACCEPTED MANUSCRIPT
Table A.8: Effect of Flex Car Penetration on Ethanol Price - Controlling for Costs
-0.576*** (0.079) 0.348*** (0.007)
Ethanol Cost (ca ) GDP per Capita Stations/Cars Hotels per km2
(4) 2SLS
Ethanol Price (pa ) -0.913*** -0.917*** -1.045*** (0.134) (0.134) (0.178) 0.347*** 0.346*** 0.345*** (0.007) (0.007) (0.007) -0.001* (0.001) 0.001 (0.006) -0.035 (0.047) 759 769 572 2524 2524 2398 151615 151615 146789 Yes Yes Yes Yes Yes Yes No Yes Yes No No No
M
2524 151615 Yes Yes No No
ED
KP F-Stat (1st Stage) Nclusters Observations Monthly Fixed Effects Station Fixed Effects Brand Fixed Effects City-Year Trends
(3) 2SLS
(5) 2SLS
CR IP T
% Flex Cars (θ)
(2) 2SLS
AN US
(1) OLS
-0.941*** (0.293) 0.340*** (0.007) -0.002** (0.001) -0.007 (0.008) -0.100** (0.043) 158 2398 146789 Yes Yes Yes Yes
AC
CE
PT
NOTES: Table displays estimated effects of shares of flex cars, θ, on ethanol retail prices. Regressions consider a station i located in municipality m in week t as the unit of analysis, and the sample period goes from January 2002 to March 2008. All prices are expressed in BRL real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). The price of gasoline is multiplied by a factor of 0.7 so that prices per unit of energy are directly comparable across fuels. θ represents the share of flex cars in municipality m in a given month. Column 1 estimated by OLS and columns 2-5 by 2SLS. All columns include ethanol wholesale prices, station and monthly fixed effects as controls. Columns 3-5 include station brand fixed effects. Columns 4 and 5 include yearly municipal GDP per capita, the monthly number of stations divided by the car fleet at the municipality level, and the yearly number of hotels per square kilometer in a municipality as controls. Column 5 includes P city-level linear time trends (year as time unit). Instrument for share of flex cars, IVmt , is equal to k nkm θ˜kt , where nkm is the number of dealers of car manufacturer k in city m before 2000 that were still active in 2015, and θ˜kt is the share of flex models offered by manufacturer k in the Brazilian territory in a particular month. Standard errors clustered by city-month in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
42
ACCEPTED MANUSCRIPT
Appendix B -
Estimating the Price Formation Model
In this section, we estimate price best response functions using our datase and use them to estimate the parameters of the price formation model we proposed in Section 3. To bring the theory to the data, we employ the richer version of the model presented in section 3.4, recognizing that demand functions may differ across fuel types. Further, we allow for heterogeneity across markets.
qijt = αmj − βj ptij + γj p¯tij ,
CR IP T
Let the demand for fuel from a vehicle of type j from station i in market m be
(5)
where αmj = αm + αj represents the composition of a market-specific fixed effect and
AN US
a fuel-specific intercept.
Returning to the same analysis performed in section 3.4, we obtain the following firstorder conditions:
(6)
1 t αma γa t t t = cia + 1I{pia > pig } + p¯ 2 2βa 2βa ia # " t t t t t t θ α + θ α + θ γ p ¯ + θ γ p ¯ ma mf a f a a ia f f if + 1I{ptia < ptig } + tia . 2(θat βa + θft βf )
(7)
M
1 t γg t αmg t t = cig + 1I{pig > pia } + p¯ 2 2βg 2βg ig " # θgt αmg + θft αmf + θgt γg p¯tig + θft γf p¯tif t t + tig , + 1I{pig < pia } t t 2(θg βg + θf βf )
CE
ptia
PT
ED
ptig
AC
tij may be interpreted as an unobserved (to the econometrician) i.i.d. cost shock that
is fuel-station-time specific.14 Our objective in this section is to identify the demand parameters by estimating price
best response functions (Pinkse, Slade and Brett, 2002). Because we do not observe quantities directly, we cannot identify the absolute scale of the demand coefficients only from pricing responses, and a normalization must be made. 14
More precisely, we assume that total marginal cost for a given fuel j in station i is equal to ctij + 2tij , where the number 2 multiplying the error is simply a normalization that does not affect the results. The two estimating equations follow by plugging this in our profit equation and deriving the optimal solution.
43
ACCEPTED MANUSCRIPT For convenience, we normalize βf = 1. If we further assume that βg = βa = βf = 1, we obtain best responses that are linear in the remaining parameters, and hence, can be estimated using simpler methods (OLS, for example):
(8)
hα γa i 1 ma + p¯tia ptia = ctia + 1I{ptia > ptig } 2 2 2 # " t t t t t t γ p ¯ + θ γ p ¯ α + θ α + θ θ f a mf ma ia a a if f f + 1I{ptia < ptig } + tia . 2(θat + θft )
(9)
AN US
CR IP T
hα 1 γg i mg ptig = ctig + 1I{ptig > ptia } + p¯tig 2 2 2 # " t t t t t t γ p ¯ + θ γ p ¯ α + θ α + θ θ f g mf mg ig g g if f f + 1I{ptig < ptia } + tig , 2(θgt + θft )
We report estimates both for the case where β coefficients are assumed to be identical (our linear case) and for the more general non-linear case.
M
To estimate these models, we must address two additional issues. First, we must define the relevant market for each fuel station. One possible approach is to define a
ED
market as a municipality; however, in the case of large cities, this definition appears to be inappropriate. Rio de Janeiro, the state capital, has 805 fuel stations spread over 1,250
PT
square kilometers; it is unreasonable to assume that they are all competing in the same market. To account for local competition in a simple manner, we define a market to consist
CE
of all stations that share the same four-digit postal code (CEP). We also experiment with narrowing the market definition to five-digit CEP areas.15
AC
Representing a second challenge is endogeneity. If there are stochastic unobserved components in the demand function or in the marginal cost, because all prices are determined in equilibrium, all terms on the right-hand side of the form p¯ij or 1I{pij < pik } are
endogenous.
To address this problem, we follow Pinkse, Slade and Brett (2002) and instrument 15
In Brazil, the postal code has 8 digits, with the first five dividing the country into increasingly fine partitions. The four-digit level corresponds to neighborhoods in large cities and to small municipalities; at the five-digit level, neighborhoods of large cities are divided into several areas. Three-digit areas are too coarse for the purposes of our model, as some of these areas cover different municipalities in our sample; conversely, eight-digit areas are too narrow, as most gas stations in the sample would be considered monopolies.
44
ACCEPTED MANUSCRIPT each endogenous regressor by the analogous term involving costs: that is, we substitute 1I{ctij < ctik } for 1I{ptij < ptik }, etc. We also instrument for the share of flex cars using the variable IVmt described in the previous section.
In addition, we estimate the model with market fixed effects to account for unobserved variation across markets. We use municipality-specific fixed effects to avoid the computational burden of estimating the non-linear model with a large number of 4-digit CEP
CR IP T
fixed effects. Finally, we may estimate price reaction functions separately for each fuel type (which will yield two different sets of estimates for the demand from flex car owners) or stack the data to impose the restriction that αf and γf must be the same in both regressions. Table B.1 presents the results of estimating the linear model separately and jointly,
AN US
respectively. Panel A show the results of the linear model, while Panel B presents the results of the non-linear estimation. In columns (1), (2), and (3) we do not impose the theoretical restriction that the coefficients on cost should be 0.5. A remarkable finding is that our estimates of this effect are near this figure; they fluctuate between 0.35 and 0.54 among all specifications. (However, because these effects are precisely estimated, they
M
are statistically different from 0.5.) This fact suggests that pricing in this market does indeed comply with the logic of a price-setting oligopoly game. This coefficient is also of
ED
independent interest: the fact that is less than unity means that demand is cost-absorbing and, according to Weyl and Fabinger (2009), a number of comparative statics predictions
PT
may be derived from that fact. For example, the entry of a new station will necessarily reduce prices, and a merger (without synergies) will raise prices of all firms (Weyl and
CE
Fabinger, 2009, theorem 4).
In column (4), we present estimates of the model obtained when we impose the the-
AC
oretical restriction that the coefficients on costs should be 0.5. This is our preferred specification.
In our model, the elasticity of demand is proportional to the difference between βj
and γj . In our linear model in panel A, we estimate that γF is 37% smaller than βF , γA is 12% smaller than βA , and γG statistically equal to βG (these numbers are similar to the ones in the non-linear model); our estimated demands are inelastic. This finding suggests that gasoline and ethanol car owners are less responsive to fluctuations in prices across fuel stations within the market.
45
ACCEPTED MANUSCRIPT Table B.1: Structural Estimation (1) Panel A cG
αG
αA γA αF
AN US
γF
Panel B cG cA
γG
PT
ED
αA
CE
βA
βG
AC
Non-Linear Model - βF 0.351*** 0.543*** (0.029) (0.014) 0.496*** 0.473*** (0.008) (0.012) 0.354*** 0.170*** (0.041) (0.012) 1.112*** 1.049*** (0.021) (0.012) 0.487*** 0.277*** (0.014) (0.019) 1.022*** 1.046*** (0.013) (0.015) -3.171*** 0.595*** 0.431*** (0.024) (0.011) (0.018) 4.726*** 0.601*** 0.911*** (0.030) (0.012) (0.021) 1.174*** 1.025*** (0.016) (0.034) 1.008*** 1.104*** (0.003) (0.019) No Yes No Yes No No No No No
M
αG
γF
(4)
CR IP T
γG
αF
(3)
Linear Model - βA = βG = βF = 1 0.455*** 0.475*** 0.5 (0.016) (0.012) 0.532*** 0.508*** 0.5 (0.010) (0.008) 0.198*** 0.151*** 0.162*** (0.026) (0.017) (0.021) 1.050*** 1.058*** 1.008*** (0.041) (0.029) (0.016) 0.378*** 0.374*** (0.014) (0.015) 0.857*** 0.878*** (0.019) (0.012) -3.594** 0.548* 0.320 0.564** (1.799) (0.305) (0.254) (0.261) 4.322*** 0.593** 0.829*** 0.625*** (1.448) (0.272) (0.226) (0.238)
cA
γA
(2)
Single Equation: Ethanol Single Equation: Gasoline Restricted Model
=1 0.5 0.5 0.229*** (0.040) 1.073*** (0.019) 0.375*** (0.049) 0.988*** (0.035) 0.608*** (0.021) 0.644*** (0.017) 1.085*** (0.022) 1.095*** (0.027) No No Yes
NOTES: Table displays estimated parameters from the structural model. Columns 1 and 2 in Panel A (Panel B) display separate estimations of equations 8 and 9 (equations 6 and 7), respectively, by 2SLS (by GMM). Columns 3 and 4 in Panel A (Panel B) display joint estimations of equations 8 and 9 (equations 6 and 7) by 2SLS (by GMM), and in column 4 the coefficients of wholesale prices are restricted to be equal to 0.5. Estimations consider a station i located in municipality m in week t as the unit of analysis, from January 2002 to March 2008 [Observations = 117,941]. The relevant market for each fuel station is the 4-digit postal code (CEP). All prices are expressed in BRL real terms, deflated by the monthly IPCA, and the price of gasoline is multiplied by a factor of 0.7. Instruments used in the estimations consider only cig , cia , θat (share of ethanol cars in a municipality), θgt (share of gasoline cars in a municipality), IVmt (defined in Equation 4) and municipality fixed effects, as well as different interactions of these terms. For example, we use 1I{ctia < ctig }IVmt as an instrument for 1I{ptia < ptig }θft . Bootstrapped standard errors clustered by city-month in parentheses [Nclusters = 2,443]. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
46
ACCEPTED MANUSCRIPT Table B.2 presents the results we obtain if we use the narrow market definition. It shows that the coefficients in our preferred specification (column 4) have similar magnitudes to the ones presented in Table B.1. Our findings with respect to pass-through are
AC
CE
PT
ED
M
AN US
CR IP T
also similar in this case.
47
ACCEPTED MANUSCRIPT Table B.2: Structural Estimation - 5-digit CEP Market Definition (1) Panel A cG
αG
αA γA αF
AN US
γF
Panel B cG cA
γG
PT
ED
αA
CE
βA
βG
AC
Non-Linear Model - βF 0.386*** 0.552*** (0.030) (0.013) 0.513*** 0.496*** (0.008) (0.013) 0.389*** 0.195*** (0.045) (0.012) 1.035*** 0.987*** (0.022) (0.011) 0.521*** 0.350*** (0.015) (0.025) 0.973*** 1.003*** (0.015) (0.016) -3.356*** 0.843*** 0.542*** (0.026) (0.013) (0.019) 4.859*** 0.356*** 0.744*** (0.032) (0.014) (0.022) 1.180*** 1.069*** (0.017) (0.040) 1.011*** 1.075*** (0.004) (0.022) No Yes No Yes No No No No No
M
αG
γF
(4)
CR IP T
γG
αF
(3)
Linear Model - βA = βG = βF = 1 0.484*** 0.504*** 0.500 (0.015) (0.014) . 0.549*** 0.524*** 0.500 (0.008) (0.010) . 0.237*** 0.188*** 0.187*** (0.028) (0.018) (0.017) 0.975*** 0.985*** 0.992*** (0.041) (0.032) (0.012) 0.477*** 0.415*** 0.395*** (0.019) (0.015) (0.013) 0.726*** 0.807*** 0.864*** (0.024) (0.025) (0.011) -3.783** 0.798** 0.452* 0.655*** (1.806) (0.365) (0.272) (0.234) 4.450*** 0.353 0.691*** 0.549*** (1.427) (0.323) (0.238) (0.211)
cA
γA
(2)
Single Equation: Ethanol Single Equation: Gasoline Restricted Model
=1
0.265*** (0.041) 1.042*** (0.022) 0.312*** (0.046) 1.028*** (0.037) 0.726*** (0.021) 0.589*** (0.017) 1.069*** (0.021) 1.085*** (0.028) No No Yes
NOTES: Table displays estimated parameters from the structural model. Columns 1 and 2 in Panel A (Panel B) display separate estimations of equations 8 and 9 (equations 6 and 7), respectively, by 2SLS (by GMM). Columns 3 and 4 in Panel A (Panel B) display joint estimations of equations 8 and 9 (equations 6 and 7) by 2SLS (by GMM), and in column 4 the coefficients of wholesale prices are restricted to be equal to 0.5. Estimations consider a station i located in municipality m in week t as the unit of analysis, from January 2002 to March 2008 [Observations = 93,360]. The relevant market for each fuel station is the 5-digit postal code (CEP). All prices are expressed in BRL real terms, deflated by the monthly IPCA, and the price of gasoline is multiplied by a factor of 0.7. Instruments used in the estimations consider only cig , cia , θat (share of ethanol cars in a municipality), θgt (share of gasoline cars in a municipality), IVmt (defined in Equation 4) and municipality fixed effects, as well as different interactions of these terms. For example, we use 1I{ctia < ctig }IVmt as an instrument for 1I{ptia < ptig }θft . Bootstrapped standard errors clustered by city-month in parentheses [Nclusters = 2,407]. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
48
Flex Cars and Competition in Fuel Retail Markets Joao ˜ Paulo Pessoa, Leonardo Rezende, Juliano Assunc¸ao ˜ PII: DOI: Reference:
S0167-7187(17)30170-4 https://doi.org/10.1016/j.ijindorg.2018.07.005 INDOR 2462
To appear in:
International Journal of Industrial Organization
Received date: Revised date: Accepted date:
7 March 2017 16 July 2018 17 July 2018
Please cite this article as: Joao ˜ Paulo Pessoa, Leonardo Rezende, Juliano Assunc¸ao, ˜ Flex Cars and Competition in Fuel Retail Markets, International Journal of Industrial Organization (2018), doi: https://doi.org/10.1016/j.ijindorg.2018.07.005
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
Flex Cars and Competition in Fuel Retail Markets∗ Leonardo Rezende‡ October 30, 2018
Abstract
Juliano Assun¸ca˜o§
CR IP T
Jo˜ao Paulo Pessoa†
We study how the diffusion of flex (bi-fuel) cars affected competition on ethanol and gaso-
AN US
line retail markets. We propose a model of price competition in which the two fuels become closer substitutes as flex cars penetration grows. We use a large panel of weekly prices at the station level to show that fuel prices and margins have fallen in response to this change. This finding is evidence of market power in fuel retail and indicates that innovations that
M
increase consumer choice benefit even those who choose not to adopt them.
Keywords: Flex-fuel vehicles; Gasoline; Ethanol; Price competition; Spatial Competi-
ED
tion; Discrete equilibrium price dispersion.
AC
CE
PT
JEL Codes: L11, L13, L62, L71.
∗
We are grateful to Jo˜ao Manoel Pinho de Mello, Heleno Pioner, Cristian Huse and seminar participants at the IIOC 2012, EARIE 2011, LAMES 2011, Taller de Organizacion Industrial 2010, FEA USP-RP, FEA USP, IE UFRJ and IPEA for helpful comments. We also thank Gabriela Rochlin for superb research assistance and ANP for providing access to data. All remaining errors are our own. A previous version of this paper circulated under the title “Flex Cars and Competition on Ethanol and Gasoline Retail Markets”. † Sao Paulo School of Economics - FGV, S˜ao Paulo, Brazil; Centre for Economic Performance, London, UK. E-mail: [email protected]. ‡ PUC-Rio, Rio de Janeiro, Brazil. E-mail: [email protected]. § PUC-Rio, Rio de Janeiro, Brazil. E-mail: [email protected].
1
ACCEPTED MANUSCRIPT
1
Introduction
The key aspect underlying the identification of conduct in empirical studies in industrial organization is the relationship connecting changes in the price elasticity of demand and firms’ pricing decisions (Bresnahan, 1982). In this paper, we explore two features of the automotive fuel retail market in Brazil that allow us to directly document this relationship. First, Brazil is a dual fuel market: both gasoline and ethanol have been available for
CR IP T
automobiles at virtually every fuel station in the country since the 1980s. Second, flex cars have been available in Brazil since 2003 and have allowed consumers to treat these two fuels as nearly perfect substitutes at the pump. Flex, or bi-fuel, vehicles are able to run on any mix of gasoline and ethanol fuel; electronic sensors identify the mix at the fuel tank and adjust the fuel injection accordingly. They have become a commercial success:
AN US
in 2008, 94% of the new cars registered in the country were flex. This innovation provides a source of change in the cross-price elasticity between two products that allow us to directly identify its effect on pricing.
We build a model of strategic price formation to study the impact of the flex car
M
fleet on equilibrium fuel prices. In the model, stations compete by choosing the price of gasoline and ethanol, and consumers treat fuel from different stations as imperfect
ED
substitutes, due to location or other idiosyncratic preferences. The model suggests that, in equilibrium, fuel retailers respond strategically to an increase in flex car penetration
PT
(that is, an increase in substitutability between products within its own product line) by reducing markups. There is a clear difference in comparison to other differentiated-
CE
goods oligopoly models: in our setting, the law of one price has no bite. Our theory does not predict that the price of the two fuels should move closer as they become perfect
AC
substitutes to a growing share of the consumers. Because both prices are set by the same firm, it is generally optimal to keep prices apart to price discriminate for consumers who cannot freely switch. We empirically assess the model by documenting the causal effect of flex-fuel pene-
tration on fuel prices using reduced-form methods.1 We study the impact of the flex car fleet on retail prices and the spread between gasoline and ethanol. Because the speed of penetration of this technology has been unequal across localities (roughly driven by the 1
We also provide estimates of the structural parameters of the model to investigate in more detail some of our theoretical predictions.
2
ACCEPTED MANUSCRIPT pace of car fleet renewal), we have been able to employ panel data methods to control for aggregate time-varying effects and local fixed effects, using a detailed sample of weekly prices at the gas station level. Our empirical analysis exploits variation in the flex fuel penetration due to local differences in the speed of fleet renewal, which is mostly driven by cross-sectional variation in income and economic activity. (We account for this possible source of omitted variable
CR IP T
bias by adding income as an additional control.) In principle, variation in fuel prices (or rather variation in the expectation about future fuel prices) may also have an effect on fleet renewal; several studies have shown that fuel prices affect demand for automobiles (Busse, Knittel and Zettelmeyer, 2013; Goldberg, 1998; Kahn, 1986; Klier and Linn, 2010; Li, Timmins and Haefen, 2009; Pakes, Berry and Levinsohn, 1993), and even specifically
AN US
consumer choice between diesel and gasoline vehicles in Europe (Verboven, 2002). To account for the potential endogeneity due to reverse causality, we employ an IV strategy that builds on the recent empirical trade literature (Autor, Dorn and Hanson, 2013; Costa, Garred and Pessoa, 2016), using data on the local presence of car dealerships. Consistent with the predictions of our model, the results show that fuel stations have
M
significantly reduced ethanol and gasoline prices: a 10 percentage point increase in the market share of flex cars reduces ethanol and gasoline prices by approximately 8.1 and
ED
2.8 cents of BRL, respectively. The absolute effects are stronger for ethanol, which is consistent with the fact that ethanol has a smaller market share in the automotive fuel
PT
market in Brazil. Our results provide evidence of market power in fuel retail and that innovations that increase consumer choice may benefit even those consumers who choose
CE
not to adopt them. We also find (weaker) evidence that the expansion of the flex car fleet has increased the spread (that is, the absolute price difference) between gasoline and
AC
ethanol prices.
Moving beyond the model, we study how flex car penetration has affected retail mar-
gins and wholesale prices (or retail costs). We find evidence that fuel retail margins have fallen with the increase of flex cars, while the estimated effects of flex cars on wholesale prices are statistically non-significant in most of our specifications. This last result corroborates the interpretation that wholesale prices are more tightly linked to international markets and less influenced by local market effects. Our paper contributes to an increasing literature on the industrial organization of
3
ACCEPTED MANUSCRIPT ethanol as automotive fuel and its relation to the gasoline market. Anderson (2012), Corts (2010) and Shriver (2015) are examples of recent studies that investigate the ethanol market in the US.2 Anderson (2012) studies the demand for the product. Shriver (2015) studies the network effect that arises due to spatially dependent complementarities between the availability of stations supplying ethanol fuel and the local number of flex cars. Corts (2010) also analyzes the decision to supply ethanol by local stations, using as a
CR IP T
source of variation purchases of flex cars by government agencies. This emphasis on the issue of expanding the distribution network reflects the incipient nature of ethanol as automotive fuel in the US. In Brazil, by contrast, the challenge of building an extensive distribution network has been completed in the 1980s with the Pr´oa´lcool program, further discussed in section 2 below. Brazil provides a setting where it is
AN US
possible to study a mature dual-fuel market operating.
Most existing studies employing Brazilian data (Ferreira, Prado and Silveira, 2009; Salvo and Huse, 2011; Boff, 2011) use time series of average price data to look for evidence of convergence toward the law of one price between the fuels3 . In contrast to this literature, we employ much more detailed data, which allows us to document the importance of
M
price dispersion across stations (an important feature of automotive fuel markets; see, e.g., Lewis, 2008). In addition, we argue in this paper that because of the structure of
ED
the retail market for fuel in Brazil, price convergence should not necessarily occur. The paper is organized as follows: section 2 provides a brief summary of the general
PT
characteristics of the Brazilian fuel market. Section 3 presents a model of oligopolistic competition among fuel stations supplying both types of fuel, allowing us to establish some
CE
key predictions on how flex car diffusion affect equilibrium prices. Section 4 describes the data we use and provides descriptive statistics. In section 5 we employ panel data methods
AC
to empirically verify the effect of flex car diffusion on prices and on the absolute price spread between gasoline and ethanol. We provide concluding remarks in section 6. 2
More precisely, E85; in the US, retail stations supply a composition of 85% of ethanol and 15% of gasoline called E85 instead of pure ethanol (E100) supplied in Brazil. 3 Representing an exception are Salvo and Huse (2013), who employ an opinion poll among flex car owners to document the relevance of motives other than price to choose between fuels.
4
ACCEPTED MANUSCRIPT
2
Flex Cars and the Automotive Fuel Market in Brazil
2.1
Ethanol-Powered and Flex Cars
Brazil has a long history of using ethanol as a vehicular fuel. In the 1970s, in response to the first oil crisis, the military government launched the Pr´o-´alcool program to encourage the production of ethanol from sugarcane and stimulate the adoption of ethanol-fueled
CR IP T
cars. The program included the use of credit subsidies for ethanol production and set favorable fuel prices at the pump to stimulate adoption of the new technology. Consumers responded to the Pr´o-´alcool program: from 1983 to 1989, most new cars purchased were ethanol-fueled vehicles. This finding may be seen in figure 1, which presents shares of new car registrations in Brazil per year by fuel type.
AN US
In response to a sharp increase in the global price of sugar, which tripled from 1985 to 1990 (USDA, 2010, table 3a), domestic ethanol production sharply declined, and the ensuing supply crisis led to a plunge in sales of ethanol-powered vehicles. Since 1995, sales of ethanol-powered cars have represented only a small fraction of new vehicle sales in Brazil. However, ethanol-fueled cars continue to represent over 10% of the current fleet.
M
In the first quarter of 2003, flex cars (or bi-fuel cars) became commercially available in Brazil. Flex cars may run on any mixture of gasoline and ethanol. Because a liter of
ED
ethanol contains roughly as much energy as 0.7 liters of gasoline (Marjotta-Maistro and Asai, 2006), a flex car owner may save money if the price ratio drifts away from that
PT
threshold.4 As figure 1 shows, flex car penetration has been dramatic: In 2008, 94% of
2.2
CE
new cars registered in Brazil were flex cars.
The Automotive Fuel Market in Brazil
AC
In Brazil, ethanol, or ethyl alcohol, is made from sugarcane. Two types of ethanol play a role in the automotive fuel market: anhydrous and hydrated. Anhydrous ethanol is mixed with gasoline fuel in the proportion of one unit of ethanol to three units of gasoline. Hydrated ethanol, a mixture that contains 5% water, is the version of alcohol readily available in drugstores and pumps at fuel stations in Brazil. Brazil is the largest producer of sugarcane in the world, the second-largest producer 4
There are other facts that might lead consumers to choose one fuel over the other. A car running on ethanol is less hazardous to the environment, as it does not create net emissions of carbon dioxide. A car powered by gasoline demands less fuel per volume, thus allowing for less frequent refueling.
5
ACCEPTED MANUSCRIPT
1980
1985
1990
1995
2000
2005
2010
AN US
1975
CR IP T
0
20
40
%
60
80
100
Figure 1: Registration of New Cars by Fuel Type
Year
Gasoline
Ethanol
Flex
M
NOTES: Figure shows shares (in percentage) of new car registrations by fuel type. Diesel and electric cars are not included in the calculations. Sources: Anfavea.
of ethanol, and a net exporter of ethanol. Brazil was a net importer of oil until 2006.
ED
In contrast, it was a net exporter of gasoline during the period of our analysis, but this pattern reversed recently and Brazil is now a net importer of the product.
PT
Before the passage of Law 9478 in 1997 (“Lei do Petr´oleo”), the Brazilian oil industry was a monopoly in the hands of state-owned Petrobras. The law created the Agˆencia
CE
Nacional do Petr´oleo, G´as Natural e Biocombust´ veis (ANP), the sector regulatory body, and broke Petrobras’ monopoly on exploration, refining, international trade, and the sea
AC
transport of oil and its main byproducts. Since January 2002, retail prices are set freely by the market. The industry is com-
posed, on the production side, of oil companies, sugar mills and refineries, which produce the fuel to the market. On the distribution side there are wholesale distributors and retailers. According to CADE (2014) there are about 200 wholesale distributors and around 40000 retailers (gas stations) in Brazil. Approximately 60% of the gas stations have exclusivity contracts with distributors. Petrobras continues to be a major player in the domestic refining, distribution and retail of gasoline, currently holding market shares
6
ACCEPTED MANUSCRIPT in these markets of 96.6%, 28.9% and 17.8%, respectively (ANP, 2010, tables 2.34, 3.6 and 3.17).
3
Model: Price Formation in the Retail Fuel market in the Presence of Flex Cars
CR IP T
In this section, we present a model of strategic price formation in the retail fuel market and use it to investigate theoretically the effect of a larger flex car fleet on equilibrium fuel prices.
The model we propose has four features that are relevant for the automobile fuel market, particularly in areas where multiple fuels are offered through the same distribution
AN US
network: i) aggregate demand for fuel (adding up different types of fuel) is proportional to the size of the automobile fleet and thus is inelastic in the short run; ii) location differences allow fuel stations to have a degree of market power; iii) every fuel station supplies both types of fuel and selects both prices to maximize joint profits; iv) for flex car owners,
M
ethanol and gasoline are substitutes.
In the model, there is (imperfect) competition across fuel stations, but between differ-
ED
ent types of fuel within a location, there is no competition at all: each station provides both ethanol and gasoline and internalizes the effect of a price change over the sales of
PT
the other product. The effect of an increase in the flex car fleet is to change the degree of substitutability between fuels and not across stations. One may tend to suppose that
CE
because competition and the direct effect of flex car penetration operate in different dimensions of product differentiation, there would be no effect of the latter on the former. This supposition is not true: we find that in equilibrium, flex car penetration leads to
AC
more competition across fuel stations. The theoretical analysis depends on an important feature of the industry: The large
cross-sectional price dispersion even within a market. In fact, it is this price dispersion that affects the behavior of flex car owners by proving them with an option value. Price dispersion in gasoline retail is a well documented phenomenon (Lewis, 2008; Hosken, McMillan and Taylor, 2008; Chandra and Tappata, 2011a). Lack of full substi-
tutability across stations may arise because of product differentiation due to geography (Lewis, 2008; Pennerstorfer, 2009; Firgo, Pennerstorfer and Weiss, 2015) or imperfect in7
ACCEPTED MANUSCRIPT formation by consumers (Chandra and Tappata, 2011a; Pennerstorfer et al., 2015; Byrne and de Roos, 2017). While recent contributions to the literature seek to provide direct empirical evidence of the latter possibility,5 we adopt a more agnostic approach. We posit a standard demand system with differentiation across stations that is compatible with the Carlson and McAfee (1983) model of consumer search, but the analysis would carry
3.1
CR IP T
through if differentiation was due to other factors.
Baseline Model
We consider an oligopoly model where N gas stations compete by setting prices for gasoline and ethanol. The population of consumers, which we normalize to 1, is divided into three groups: gasoline-fuelled car owners, ethanol-fuelled car owners, and flex car owners. We
AN US
call θ the fraction of flex car owners, and to maintain symmetry in the model, we assume that the rest of the fleet is equally divided between gasoline and ethanol. The type of fuel generally has no effect on consumption, but for flex car owners, gasoline and ethanol within the same gas station are perfect substitutes. (We measure fuel in terms of energy
M
content, so flex car owners always buy the cheaper fuel.)
There is differentiation across fuel stations. For a given fuel price profile, let pif =
ED
min{pig , pia } (the price effectively faced by a flex type in station i). Using this notation, we assume that the demand for fuel from station i from a consumer with car type j = g
qij = α − βpij + γ p¯ij ,
CE
PT
(gasoline-powered), a (alcohol-powered), or f (flex) is
where α, β and γ are positive constants, pij is the price of fuel j in station i, and p¯ij is
AC
the average price of fuel j in all stations except i. In the baseline model, we adopt the same functional form for all fuel types. This
procedure is followed for simplicity and to isolate the effect on substitutability across fuels as the car fleet changes. (We relax this assumption in section 3.4 below.) For consumers within each car group, we assume that demand across stations exhibits a simple linear form of symmetric product differentiation. This demand system may be justified by 5
Pennerstorfer et al. (2015) and Byrne and de Roos (2017) construct altenative measures of the fraction of informed consumers (namely, fraction of commuters within a locality and web traffic in a price comparison website) and study how those relate to price dispersion.
8
ACCEPTED MANUSCRIPT Carlson and McAfee (1983), who model consumer choice by a process of costly search among identical products sold by different firms at (potentially) different prices. Carlson and McAfee show that if the distribution of search costs in the consumer population is uniform, then aggregate demand for firm i exhibits the form postulated above, with α = 1/N and β = γ = (N − 1)/N , where N is the number of firms in the market. We seek to obtain a prediction regarding Bertrand-Nash equilibrium prices for firms
CR IP T
that face demand arising from this process and have a cost function as follows: Ci (qig , qia ) = cig qig + cia qia + Fi ,
where qig and qia are the quantities sold of gasoline and ethyl alcohol in station i, cig
AN US
and cia are marginal costs, and Fi is a fixed cost component. We assume that marginal costs are constant and exogenous, but different across fuels and stations. We believe that assuming that marginal costs differ across stations is reasonable, given that, according to our data, there is substantial variation in fuel wholesale prices faced by each station.
Properties of Equilibrium Prices
M
3.2
To obtain a characterization of equilibrium prices, we must first integrate the demand
ED
over the mass of consumers with each car type. If pig 6= pia , station i will sell Qig of gasoline and Qia of ethanol, where
1−θ = (α − βpig + γ p¯ig ) 2 +1I{pig < pia }θ (α − βpig + γ p¯if )
PT AC
and
CE
Qig
Qia
1−θ = (α − βpia + γ p¯ia ) 2 +1I{pig > pia }θ (α − βpia + γ p¯if ) .
1I{A} represents the indicator function, with value one if condition A is true and zero otherwise. If pig = pia , flex car owners are indifferent between the two types of fuel, and we must specify a sharing rule τ ∈ [0, 1]. Formally, we follow the approach of Simon and Zame (1990) and adopt an endogenous sharing rule, although the specifics of the 9
ACCEPTED MANUSCRIPT tie-breaking do not affect the equilibrium determination in this model. The profit of station i is simply πi = (pig −cig )Qig +(pia −cia )Qia −Fi . Maximizing this expression with respect to pig and pia yields this firm’s best-response function. Whenever θqif is positive, πi is discontinuous at the point pig = pia (and the profit at the discontinuity point depends on the tie-breaking rule). In any pure-strategy equilibrium, we may classify stations into those that choose to
CR IP T
charge pig > pia , pig < pia or pig = pia . In the first two cases, profits are continuously differentiable around the chosen prices, and the latter may be characterized by the firstorder conditions
∂ π ∂pig i
= 0 and
∂ π ∂pia i
= 0.
In the next proposition, we show that the last alternative is never optimal: in equilibrium, no fuel station elects to charge pig = pia :
AN US
Proposition 1 If θ > 0 then any firm i will post pig 6= pia in equilibrium. Proof.
If some of the other fuel stations charge a different price for each fuel, p¯if < p¯ig or p¯ia . Without loss of generality, suppose that p¯if < p¯ig .
M
The profit of firm i, as a function of pig , is discontinuous at point pig = pia . The
ED
value of the profit at this point depends on how flex car owners break the tie between fuels, but it always lies between the left-hand and the right-hand limits limpig %pia πi and limpig &pia πi , which correspond to the extreme cases that all flex car owners buy ethanol
PT
or gasoline when prices are equal, respectively. Therefore, in order to show that pig = pia cannot be optimal, it suffices to show that
point.
CE
either the left-hand derivative is negative or the right-hand derivative is positive at that
AC
Let x and y be the right and left derivatives, respectively, at that point: x=
∂ + (1 − θ) π = [q(pig , p¯ig )) − β(pig − cig )] ∂pig i 2
y=
∂ − π = x + (θ)[q(pig , p¯if )) − β(pig − cig )] ∂pig i
and
For firm i to find it optimal to charge pig = pia , it must be the case that y ≥ 0 ≥ x. However, such a case is impossible as x ≤ 0 ⇒ y < 0. 10
ACCEPTED MANUSCRIPT If all other stations charge the same price for both fuels, p¯if = p¯ia = p¯ig = p¯. Evaluating the left and right derivatives around a point pig = pia = p, we define as before x=
∂ π+, ∂pig i
and y =
∂ π− ∂pig i
and, analogously, x0 =
∂ π+, ∂pia i
and y 0 =
∂ π−. ∂pia i
As we argued
above, x < 0 ⇒ y < 0 and x0 < 0 ⇒ y 0 < 0. For the first-order condition to be satisfied,
we need x = 0, x0 = 0 at this point. However, this condition is impossible because
CR IP T
x = [(1 − θ)/2][q(p, p¯)) − β(p − cig )] 6= x0 = [(1 − θ)/2][q(p, p¯)) − β(p − cia )], as cig 6= cia .
Considering the two possible first-order conditions that must be satisfied by pij , we obtain the following expressions:
AN US
α θ γ 1 γ (¯ pig − p¯if ) pig = cig + p¯ig + − 1I{pig < pia } 2 β β 1+θβ and
1 γ α θ γ pia = cia + p¯ia + − 1I{pig > pia } (¯ pia − p¯if ) 2 β β 1+θβ
M
Note that p¯if ≤ p¯ig and p¯if ≤ p¯ia , so the right-hand sides of the expressions above are decreasing in θ. Because prices across stations are strategic complements, we conclude
ED
that prices are decreasing in θ. This result is summarized in the proposition below.
flex cars.
PT
Proposition 2 Ethanol and gasoline prices are decreasing with respect to the fraction of
CE
A larger fleet of flex cars pulls prices down because flex cars provide an option value to their owners: if fuel prices are dispersed, flex car owners expect to find lower prices
AC
than other drivers because they can always pick the cheapest alternative. For this reason, flex car owners are willing to pay less, and fuel stations respond by lowering prices.
3.3
Effect on price spreads
Let us turn to the analysis of the dispersion between gasoline and ethanol prices. Common sense from the law of one price suggests that as the flex car fleet grows, the prices of the two fuels should move closer together. In this paper we argue that this intuition may not hold if the two fuels are sold by the same retailers, as in the case of the Brazilian market. 11
ACCEPTED MANUSCRIPT In this section we employ the model to show that in fact the price spread, defined here as the the absolute difference between fuel prices charged by the same station, actually moves in the opposite direction, and becomes larger as the market share of flex cars grow. From the price best response functions we obtain the following expression for the price spread of station i:
1 2 1 2
|cig − cia | + |cig − cia | +
γ β γ β
|¯ pig − p¯ia | + |¯ pig − p¯ia | +
θ γ (¯ pig 1+θ β θ γ (¯ pia 1+θ β
− p¯if ),
if pg < pa ;
− p¯if ),
if pg > pa .
CR IP T
|pig − pia | =
Since p¯if ≤ p¯ia , p¯ig , the spread is weakly increasin the θ, and strictly increasing whenever the cheapest fuel is not the same everywhere.
Introducing asymmetry across fuels
AN US
3.4
To simplify notation, in our baseline model markets for gasoline and ethanol are fully symmetric. In reality these markets are quite different: at the very least, the disparate sizes of the fleets make the gasoline market much larger than the ethanol market; also
M
since automobiles with different fuel types have been purchased at different types, and may be used in different ways, demand functions for each fuel may differ.
ED
In this section we investigate the effect of introducing asymmetry in our model in two steps. In the first extension, we allow for different fleet sizes, while holding the same
PT
demand function of the baseline model. This does not fundamentally change our analysis. In fact a stronger version of proposition 2 holds: any shift in the composition of the fleets
CE
that increase the share of flex cars and reduces the share of either gasoline or ethanol fleets has the effect of reducing prices of both fuels.
AC
To see that, let the share of flex, gasoline and ethanol cars be θf , θg and θa , respectively. (We no longer impose θg = θa as in the baseline model.) In that case the amount of fuel sold by station i will be Qig = θg (α − βpig + γ p¯ig ) +1I{pig < pia }θf (α − βpig + γ p¯if )
12
ACCEPTED MANUSCRIPT and Qia = θa (α − βpia + γ p¯ia ) +1I{pig > pia }θf (α − βpia + γ p¯if ) . From the first-order conditions, we obtain the following best responses:
α γ p¯ia + , 2β 2β α γ(θa p¯ia + θf p¯if ) + , 2β 2β(θa + θf )
if pg > pa ;
CR IP T
c ia + 2 pia = c ia + 2
α γ p¯ig + , 2β 2β α γ(θg p¯ig + θf p¯if ) + , 2β 2β(θg + θf )
if pg < pa ;
if pg < pa ; if pg > pa .
AN US
c ig + 2 pig = c ig + 2
Once again since p¯if ≤ p¯ia , p¯ig , any change that increases θf and reduces θg and θa has the direct effect of reducing one of the prices of each station (the lowest). In equilibrium since prices are strategic complements, equilibrium prices fall.
M
A further possible extension is to allow individual demand functions to vary according to type of fuel. Flex fuel car technology has been introduced gradually over the period of
ED
analysis; thus flex fuel cars tend to be newer than gasoline cars. It is reasonable to expect that their owners are richer (and therefore less price sensitive) and the cars are used more
PT
intensively, thus consuming more fuel. (Similarly, existing ethanol-only cars tend to be very old since they were effectively no longer manufactured in the period of analysis.)
CE
A simple way to add asymmetry in demand such that comparison with the baseline model is straightforward is to allow the coefficients of the demand function to depend on
AC
fuel type. Suppose that individual demand for a car with fuel j = g, a, f is qij = αj − βj pij + γj p¯ij .
In that case the best response functions become c αg + γg p¯ig ig + , 2 2βg pig = c θ α + θf αf + θg γg p¯ig + θf γf p¯if ig + g g , 2 2(θg βg + θf βf ) 13
if pg > pa ; if pg < pa ;
ACCEPTED MANUSCRIPT and
c αa + γa p¯ia ia + , 2 2βa pia = c θ α + θf αf + θa γa p¯ia + θf γf p¯if ia + a a , 2 2(θa βa + θf βf )
if pg < pa ; if pg > pa .
In the model, composition effects may affect our main conclusion. Taking derivatives, we find that to guarantee that an increase in the flex car fleet share lowers prices, it must be
CR IP T
true that αf βg − αg βf + βg γf p¯if − βf γg p¯ig ≤ 0
(1)
αf βa − αa βf + βa γf p¯if − βf γa p¯ia ≤ 0.
(2)
AN US
and
A simpler sufficient condition would be βf ≥ βg , βa , αf ≤ αg , αa (unlikely) and γf ≤ γf ≤ γg , γa .
In Appendix B - below, we propose and apply a method to structuraly estimate these
M
parameters (up to a normalization). Relying on our preferred specification (column 4 in table B.1), we find that the βi ’s are similar, γf is smaller than γg and γa , and αf is larger
ED
than αg and αa . These two last findings are compatible with the presumption that flex car owners use their cars more intensively (demand more fuel and are less elastic). With
PT
respect to the condition needed for an increase in the flex car fleet to bring prices down, a sufficient condition is having average ethanol prices above 0.98 and average gasoline
CE
prices above 1.69, a condition that is satisfied in the majority of locations and periods in
AC
our sample.
4
4.1
Data Data Sources
This study combines data from different sources. The first source is the Levantamento de Pre¸cos e de Margens de Comercializa¸c˜ ao de Combust´ıveis, a weekly survey conducted by ANP, the Brazilian regulatory agency covering the oil, gas and biofuel industry. ANP collects data on retail prices for ethanol and gasoline prices, as well as prices paid to fuel
14
ACCEPTED MANUSCRIPT distributors, at individual fuel stations in 10% of the municipalities in Brazil. There is also information on the brand of the station (or whether it has no brand), the date on which prices were collected and the address of the station. Our sample contains weekly prices from January 2002 to March 2008 for stations located in 38 municipalities in the Rio de Janeiro state. Not all fuel stations are surveyed every week. Coverage is 100% in small municipalities, whereas for larger markets, the survey adopts a rotating sample
CR IP T
(with random selection) that eventually covers all fuel stations in the location. Table 1 provides information on the number of stations sampled, as a proportion of the overall population of stations, in the 38 municipalities used in this study.
The second data source is a monthly data set on the number of cars with license plates from each municipality in the state of Rio de Janeiro, classified according to fuel
AN US
type (gasoline, ethanol, flex, gasoline + CNG6 , ethanol + CNG, flex + CNG). The time period is the same considered in the ANP’s survey (from January 2002 to March 2008). This data set was provided by the motor vehicles department of the state of Rio de Janeiro (Detran - RJ). Although ANP verifies the price charged by fuel stations in all Brazilian states, we are unable to expand our analysis to the entire Brazilian territory because we
M
do not have access to data on the number of cars by fuel type in other states. We included annual municipal GDP per capita as a regressor (obtained from the
ED
Brazilian institute of geography and statistics - IBGE). We also included annual data on number of hotels per square km in each municipality (provided by the Data and
PT
Information Center of Rio de Janeiro - CIDE RJ) as a proxy for markets where a large fraction of drivers are not local. The former series is available from 2002 to 2006, and the
CE
latter is available from 2002 to 2004. Because we are analyzing the period between 2002
AC
and 2008, in both cases the missing years were replaced by the most recent available ones.
4.2
Descriptive Statistics
Figure 2 presents a map of the Rio de Janeiro state with a bullet for each municipality in our sample. The size of the bullet is proportional to (the log of) the local car fleet, and the color is coded according to flex car penetration in 2007. As is shown, the flex car fleet grew considerably across time in all locations, reaching a maximum value of 14.1% 6
CNG stands for compressed natural gas. In Brazil, it is possible to convert vehicles to run on natural
gas.
15
ACCEPTED MANUSCRIPT in the city of Mangaratiba in 2007. There is variation both in the time series and the cross-section/geographic dimensions. Figure 2 strongly suggests that flex car adoption is related closely to local income, growing the most in the capital (the Rio de Janeiro city), tourist resort towns (Arma¸c˜ao de B´ uzios, Parati, Angra dos Reis), and rapidly developing areas (Maca´e, Mangaratiba). The municipalities with the lowest adoption rates are located in the north-eastern part of
CR IP T
the state, the traditional region of sugarcane production in Rio de Janeiro. Table A.1 in the Appendix shows how the percentage of flex cars changed between 2004 and 2007 in the 38 municipalities in our sample.
Table 1: Gasoline and Ethanol Average Prices by Year in the State of Rio de Janeiro (in R$)
2002 2003 2004 2005 2006 2007
1.021 1.175 1.014 1.150 1.325 1.161
1.643 1.774 1.656 1.741 1.837 1.752
Ethanol P rice Gasoline P rice
Share of Sample with Cheaper Ethanol 0.78 0.60 0.76 0.71 0.38 0.66
AN US
Ethanol Price Gasoline Price
0.62 0.66 0.61 0.66 0.72 0.66
ED
M
NOTES: Table presents retail prices for gasoline and ethanol, the ratio between the two prices and the share of stations in our sample that provide ethanol as the cheapest fuel, between 2002 and 2007. All prices are expressed in BRL real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). Sources: ANP.
PT
Table 1 presents the evolution of retail prices over time for each type of fuel. All prices are expressed in real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor
CE
Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). Ethanol average retail prices have been below the 70% threshold of gasoline average retail prices except in
AC
2006. Therefore, it is reasonable to infer that most owners of flex cars in our sample are choosing to fill their tanks with ethanol. Table A.2 in Appendix A provides other basic statistics on wholesale prices, retail prices and margins (i.e., the gap between wholesale and retail prices) in our data. Figure 3 shows the share of markets (defined here as a week-city pair) across time in which a single fuel is not the cheapest across all stations. We can see that there is no point in time when a single fuel is the cheapest across all markets. Hence, consumers looking for the cheapest fuel in a given market in our sample may end up choosing either ethanol or gasoline depending on which station they visit. This suggests that both fuels 16
17 6.9%
14.1%
6.7%
4.3%
4.6%
4.50%
PT
5.6%
6.9%
10.5%
7.4%
5.5%
4.4%
6.1%
5.3%
6.5%
ED
12.4%
5.5%
6.3%
7.3%
9.0%
4.3%
5.6%
4.4%
7.1%
5.3%
7.2%
11.1%
8.6%
10.5%
2.6%
CR IP T
AN US
6.5%
M 6.2%
4.2%
3.8%
2.3%
NOTES: Map shows the share of flex cars in the municipalities considered in our sample across the state of Rio de Janeiro in 2007. Sources: Detran RJ.
9.1%
10.0%
4.6%
> 10%
7% -10%
5% - 7%
2% - 5%
Percentage of Flex Cars 2007
CE
Figure 2: Geographic Distribution of Sample across Rio de Janeiro State.
AC
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT are being used as instruments for price competition between stations during the period we analyze.
CR IP T
.6
AN US
.4 0
.2
Share of Cities
.8
1
Figure 3: Competition between Gasoline and Ethanol over time
100
200
M
0
300
400
Week
PT
ED
NOTES: Figure plots the share of cities over time in our sample in which both gasoline and ethanol are the cheapest fuel in at least one station, i.e., the share of cities that do not have a single fuel as the cheapest across all stations. Sources: ANP.
We can also investigate evidence of price competition between gasoline and ethanol
CE
by computing a measure of within station rank-reversal, similar to the between-station rank-reversal measure presented in Chandra and Tappata (2011b).7 This measure is
AC
station specific and less prone to outliers issues when compared to our previous one. It is computed as the fraction of periods that a given station switches its cheapest fuel. For example, if a station uses gasoline as the cheapest fuel more often than ethanol, but in 10% of the periods it uses ethanol as its cheapest product, our within station rank-reversal variable would be equal to 0.1. If there is no price switching across time, this measure would be zero, and a value of 0.5 would imply that there is no fuel used more often as the cheapest in one particular station. 7
The authors use a panel data set on U.S. gasoline retailers to study how imperfect information shapes price dispersion.
18
ACCEPTED MANUSCRIPT Figure 4a presents the rank-reversal probability density function considering our full sample (kernel density with bandwidth of 0.01). Approximately 88% of the stations present a positive rank-reversal, and stations change its cheapest fuel more than 23% of the time on average (mean of 0.232 and standard deviation of 0.161). This is evidence that both fuels are instruments of price competition within a station. Figure 4a also shows a point mass at zero, meaning that 12 % of stations never switch
CR IP T
their cheapest fuel. These stations, however, do not appear as frequently as an average station in our sample and are likely to be new entrants that die fast or simply stations that are not sampled regularly. In fact, figure 4b shows that when we focus on stations that remain in our sample for more than 6 months (continuously or not), the mass at zero disappears.
AN US
Figure 4: Rank Reversal within Station (a) All Stations
.2
.3
.4
2 1.5
Probability Density Function
ED
.1
.5
0
PT
0
1
M
4 3 2 1
Probability Density Function
2.5
5
(b) Stations with more than 26 weeks
Fuel Rank Reversals
.1
.2
.3
.4
.5
Fuel Rank Reversals
AC
CE
NOTES: Figures plot the probability density function of within station rank-reversal between gasoline and ethanol (kernel density with bandwidth of 0.01). Panel A considers all stations and Panel B considers only stations with more than 26 weeks in our sample. Rank-reversal is computed as the fraction of periods that a given station switches its cheapest fuel. For example, if a station uses gasoline as the cheapest fuel more often than ethanol, but in 10% of the periods it uses ethanol as its cheapest product, our within station rank-reversal variable would be equal to 0.1 in this case. Mean of 0.232 and standard deviation of 0.161 in our sample.
To further investigate the competition between the two products, we compute a type of rank-reversal measure between stations, defined as the share of stations in a given municipality-week pair that did not choose the most likely cheapest fuel. For example, if most stations in a municipality-week pair chose ethanol as their cheapest product, our between station rank reversal measure in that market will be equal to the share of stations 19
ACCEPTED MANUSCRIPT that charge less for gasoline than for ethanol. If our measure is zero, it is unlikely that the two fuels are competing fiercely in that market. Figure A.1 in Appendix A shows the average (across markets) of this measure over time, presenting further evidence that ethanol and gasoline are indeed competitors in the fuel retail market. Table 2: Distributors’ Fuel Sales and size of Fleet in the State of Rio de Janeiro Gas sales per vehicle 0.76–0.76 0.64–0.64 0.64–0.64 0.57–0.59 0.52–0.55 0.48–0.53
Ethanol sales per vehicle 0.33–0.33 0.21–0.21 0.22–0.23 0.32–0.38 0.34–0.47 0.44–0.76
CR IP T
2002 2003 2004 2005 2006 2007
Gasoline Ethanol Gasoline Ethanol Flex Sales Sales Fleet Fleet Fleet 1,971,934 157,567 2,606,238 473,434 1,764,595 98,178 2,775,071 478,060 1,848,172 109,817 2,879,902 476,632 23,561 1,739,319 180,528 2,958,560 475,307 84,297 1,660,803 224,255 3,016,335 473,880 188,271 1,635,152 359,404 3,098,499 472,670 335,629
AN US
NOTES: Table shows distributors’ consolidated sales volume (cubic meters) in the state of Rio de Janeiro by fuel type between 2002 and 2007, as well as the total car fleet in the state by fuel type. In each of the last two columns, we provide a range of values for fuel sales per vehicle, depending on how one assigns the flex fleet in the denominator. The lower limit considers that all drivers of flex cars buy the corresponding fuel, while the upper limit assumes that flex drivers never do so.
To provide a sense of the relative size of the market of each fuel, Table 2 shows the
M
evolution of aggregate sales and the size of the fleets in the state of Rio de Janeiro. Even though we do not have access to data on gasoline and ethanol sales by station, ANP
ED
publishes the distributors’ consolidated sales volume in the state of Rio de Janeiro, as shown in columns 2 and 3. In this table, we also present the aggregate size of the fleet
PT
in our data set (columns 3-5), and compute sales per car, in cubic meters, in the last two columns. We provide a range of values for fuel sales per vehicle, depending on how one
CE
assigns the flex fleet in the denominator. The lower limit considers that all drivers of flex cars buy the corresponding fuel, while the upper limit assumes that flex drivers never do
AC
so. For example, in 2007, ethanol sales per vehicle ranged between 0.44 (if all drivers of flex cars bought only ethanol) and 0.76 (if all drivers of flex cars bought only gasoline) cubic meters.
5
Empirical Results
In this section, we document the effects of flex car penetration on the fuel market by testing two of the model predictions. First, we estimate the effect of the penetration of flex cars (θ) on retail prices for both ethanol and gasoline. Second, we investigate whether 20
ACCEPTED MANUSCRIPT θ affected the price spread between gasoline and ethanol. Finally, moving beyond the model we investigate the effect of flex car penetration on retail costs and margins. Our basic estimation equation is: t t t yim = δθm + ζ 0 Zim + µtim ,
(3)
t represents ethanol/gasoline prices8 in station i, municipality m and week where yim
CR IP T
t t t, and θm represents the share of flex cars in municipality m at time t. Zim is a vector
containing variables that are potentially important to determine prices at the local level, depending on each regression specification. It includes municipal GDP per capita, the number of stations per car, and the number of hotels per square km to control for the effect of income growth, variation in relative fuel station scarcity, and the intensity of local
AN US
tourist activity, respectively. It contains station brand fixed effects to capture pricing differences across distinct fuel suppliers (note that some stations change their brand over time, allowing us to identify the coefficients of these dummies even when we use station fixed effects in our specification). We also control for station fixed effects in order to
M
capture the within correlation between θ and our dependent variable (and not simply how levels of θ are correlated with price/margin levels). Finally, the vector includes
ED
monthly fixed effects (for example, January 2003) to account for shocks that may affect fuel prices in the Brazilian territory as a whole, and city-year linear time trends.
9
PT
The error term, µtim , represents unobserved components that affect prices in a given station. This term might be correlated with unobserved contemporaneous local market
CE
shocks that affect price decisions at the station level. Moreover, the decision of purchasing a flex car can be correlated with unobserved local price shocks even after accounting for time and/or market effects.
AC
In order to estimate the effect of flex car penetration on prices and margins controlling
for possible reverse causality bias, we adopt an instrumental variable (IV) strategy that builds on the literature on trade and labor markets (e.g. Bartik, 1991; Autor, Dorn and Hanson, 2013). Our IV is composed of two terms. The first one is the number of dealers from a particular brand that entered each municipality before 2000 and were still operating 8
We also use wholesale prices, retail margins or absolute price spreads as dependent variables - see Sub-sections 5.2 and 5.3 for details. 9 As a robustness check, we also control for fuel retail costs (or wholesale prices) at the station-time level (see Tables A.7 and A.8 in the Appendix). ANP provides information on fuel prices paid by stations to wholesalers based on a station’s most recent purchase invoice.
21
ACCEPTED MANUSCRIPT by the end of 2015: Fiat, Ford, General Motors, Honda, Mitsubishi, Nissan, PeugeotCitroen, Renault, Toyota and Volkswagen. (These were the only manufacturers that produced flex cars before early 2008.)10 This term will suffer from reverse causality if there are anticipation effects of post-2003 shocks that led to differential entry of dealers across municipalities before 2000. The second term is the share of flex car models offered by each vehicle manufacturer in the Brazilian territory.11 For example, if Ford offered 3
CR IP T
models to Brazilian consumers in January 2000 and only one of them was flex, our measure is equal to 1/3 for Ford at that point in time. This measure should be determined by the Brazilian market as a whole, and should not be influenced by shocks at the municipality level, i.e., local price shocks should not affect this measure (after we control for aggregate time effects and city year-trends).
AN US
We then compute the following IV at the municipality-month level: IVmt =
X
nkm θ˜kt ,
(4)
k
where nkm is the number of dealers of manufacturer k in city m before 2000 that were
territory in a particular month.
M
still active in 2015, and θ˜kt is the share of flex models offered by brand k in the Brazilian
ED
The idea behind the instrument is that drivers adopted flex cars faster in municipalities that were initially more exposed to brands that adopted flex engines more quickly. In fact,
t IVmt and θm .
PT
the first stage statistics presented later in this section indicate a high correlation between
CE
One potential source of concern with our instrument is that specific markets may be driving the sales of some car manufacturerers, i.e., a car brand may concentrate its sales
AC
in a few Brazilian municipalities in our sample. In that case, local (unobserved) shocks could affect our IV measure, as pointed out by Blanchard and Katz (1992) in a different framework. We do not believe that this is true in our set-up, given that Brazilian car sales are relatively spread out across the country and that the leading state in car sales 10
We obtain this information from the website of the union of automobile dealers of the state of Rio de Janeiro (SINCODIV): http://www.sincodiv-rj.com.br/. We start with a sample of 145 active local dealers on November of 2015 around the state of Rio de Janeiro. Then we are able to extract information on dealers’ date of entry from websites such as http://www.cnpjbrasil.com/. By selecting all the dealers that initiated activity before 01/01/2000, we are left with 59 dealers. 11 We obtain this information from the website of the national association of automobile vehicles manufacturers: http://www.anfavea.com.br/.
22
ACCEPTED MANUSCRIPT is not Rio de Janeiro but S˜ao Paulo.12 Hence, after controlling for local market shocks and aggregate time effects, the instrument should be uncorrelated with the error term in equation 3.
5.1
Effect of Flex Cars on Retail Prices
According to proposition 2, the penetration of flex cars tightens the competition in the
CR IP T
fuel market, decreasing gasoline and ethanol prices. Tables 3 and 4 present a reducedform analysis of the impact of flex cars on prices. Each column adopts a different set of controls. All the results presented in this and the following section adjust the price of gasoline by a factor of 0.7 so that prices per unit of energy are directly comparable across fuels.
AN US
Table 3 analyzes the effect of flex cars on the gasoline market. Column (1) suggests that, according to the prediction of the model, the estimated coefficient is negative. This specification accounts for station fixed effects and time dummies. The point estimate implies that a 10 percentage point increase in flex car penetration reduces the gasoline
M
price by 1.68/0.7 = 2.4 cents (of Brazilian Reais) per liter. We then use our IV strategy. When we compare the results in column (2) to the ones in column (1) - its OLS coun-
ED
terpart - we can see that the effect of θ is stronger, suggesting that our OLS coefficients are biased towards zero, possibly because (unobserved) local price shocks are positively correlated with the share of flex cars in this simpler specification without other controls.
PT
Our first stages are strong, as suggested by the Kleibergen-Paap statistics (significant at
CE
all reasonable levels) shown in the lower part of the panel. Considering flag/brand fixed effects in column (3) makes no significant difference (remember that brand changes across time within a station allow us to identify the coefficients of theses dummies). Considering
AC
other controls in column (4) reduces the impact to approximately zero cents per liter and the result becomes statistically insignificant. However, when we account for city-level linear time trends (year as time unit) in column (5), the coefficient becomes stronger and statistically significant. Table 4 presents the same analysis for the ethanol market. The effects on ethanol prices are similar but more pronounced. In our preferred specification in column (5), a 10 percentage point increase in flex car penetration reduces the ethanol price by 8.1 cents 12
Rio was the fourth state in sales in 2012, according to Infomoney (2012).
23
ACCEPTED MANUSCRIPT Table 3: Effect of Flex Car Penetration on Gasoline Price
-0.168*** (0.030)
GDP per Capita Stations/Cars Hotels per km2 KP F-Stat (1st Stage) Nclusters Observations Monthly Fixed Effects Station Fixed Effects Brand Fixed Effects City-Year Trends
2532 280746 Yes Yes No No
(3) 2SLS
(4) 2SLS
(5) 2SLS
Gasoline Price (pg ) -0.195*** -0.193*** -0.012 -0.195* (0.049) (0.048) (0.065) (0.108) 0.001*** -0.001*** (0.000) (0.000) -0.020*** -0.013*** (0.003) (0.003) -0.028 -0.021 (0.026) (0.021) 838 845 684 142 2532 2532 2406 2406 280746 280746 272190 272190 Yes Yes Yes Yes Yes Yes Yes Yes No Yes Yes Yes No No No Yes
CR IP T
% Flex Cars (θ)
(2) 2SLS
AN US
(1) OLS
PT
ED
M
NOTES: Table displays estimated effects of shares of flex cars, θ, on gasoline retail prices. Regressions consider a station i located in municipality m in week t as the unit of analysis, and the sample period goes from January 2002 to March 2008. All prices are expressed in BRL real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). The price of gasoline is multiplied by a factor of 0.7 so that prices per unit of energy are directly comparable across fuels. θ represents the share of flex cars in municipality m in a given month. Column 1 estimated by OLS and columns 2-5 by 2SLS. All columns include station and monthly fixed effects as controls. Columns 3-5 include station brand fixed effects. Columns 4 and 5 include yearly municipal GDP per capita, the monthly number of stations divided by the car fleet at the municipality level, and the yearly number of hotels per square kilometer in a municipality as controls. Column 5 includes city-level linear time trends (year as time unit). Instrument for share of flex cars, IVmt , is equal P to k nkm θ˜kt , where nkm is the number of dealers of car manufacturer k in city m before 2000 that were still active in 2015, and θ˜kt is the share of flex models offered by manufacturer k in the Brazilian territory in a particular month. Standard errors clustered by city-month in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
CE
(of Brazilian Reais) per liter.
We also perform some robustness analysis. In our model, competition in fuel markets
AC
increase with the share of flex cars as long as the same fuel is not the cheapest one across all stations in a given market. Hence, if this characteristic is not present across the markets we are analyzing it is possible that we are simply finding a spurious negative correlation between θ and our price measures. Figure 3 suggests that this is not the case. Nevertheless, we restrict our analysis to markets in which both gasoline and ethanol appear as the cheapest fuel in at least one station and present the results in Tables A.3 and A.4 in Appendix A. They show that the magnitude and sign of the coefficients remain basically unchanged.
24
ACCEPTED MANUSCRIPT Table 4: Effect of Flex Car penetration on Ethanol Price
-0.658*** (0.080)
GDP per Capita Stations/Cars Hotels per km2 KP F-Stat (1st Stage) Nclusters Observations Monthly Fixed Effects Station Fixed Effects Brand Fixed Effects City-Year Trends
2532 260960 Yes Yes No No
(3) 2SLS
(4) 2SLS
Ethanol Price (pa ) -0.750*** -0.753*** -0.755*** (0.145) (0.145) (0.188) -0.001 (0.001) -0.018*** (0.006) -0.034 (0.059) 852 862 722 2532 2532 2406 260960 260960 252630 Yes Yes Yes Yes Yes Yes No Yes Yes No No No
(5) 2SLS
-0.812** (0.336) -0.004*** (0.001) -0.017** (0.008) -0.061 (0.053) 144 2406 252630 Yes Yes Yes Yes
CR IP T
% Flex Cars (θ)
(2) 2SLS
AN US
(1) OLS
PT
ED
M
NOTES: Table displays estimated effects of shares of flex cars, θ, on ethanol retail prices. Regressions consider a station i located in municipality m in week t as the unit of analysis, and the sample period goes from January 2002 to March 2008. All prices are expressed in BRL real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). The price of gasoline is multiplied by a factor of 0.7 so that prices per unit of energy are directly comparable across fuels. θ represents the share of flex cars in municipality m in a given month. Column 1 estimated by OLS and columns 2-5 by 2SLS. All columns include station and monthly fixed effects as controls. Columns 3-5 include station brand fixed effects. Columns 4 and 5 include yearly municipal GDP per capita, the monthly number of stations divided by the car fleet at the municipality level, and the yearly number of hotels per square kilometer in a municipality as controls. Column 5 includes city-level linear time trends (year as time unit). Instrument for share of flex cars, IVmt , is equal P to k nkm θ˜kt , where nkm is the number of dealers of car manufacturer k in city m before 2000 that were still active in 2015, and θ˜kt is the share of flex models offered by manufacturer k in the Brazilian territory in a particular month. Standard errors clustered by city-month in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
CE
Our IV strategy assumes that manufacturers determine their supply of flex cars at the national level. However, large markets such as the city of Rio de Janeiro may have
AC
more influence than others in their decision. To verify the robustness of our results, we exclude the city of Rio from our sample and re-estimate equation 3. Tables A.5 and A.6 still show a negative effect of θ on prices. The effect of flex car penetration on gasoline
prices appears to be stronger. However, the effect on ethanol prices loses significance (and becomes positive) in our preferred specification in column (5). We have so far abstracted from the impact of retail fuel costs, which vary substantially by station and are important components of fuel retail prices. We also verify whether our findings are robust to the addition of wholesale prices as controls. Tables A.7 and A.8
25
ACCEPTED MANUSCRIPT present the results. Our results remain basically the same, with only minor shifts in magnitudes relative to Tables 3 and 4.
5.2
Effect of Flex Cars on the Spread between Gasoline and Ethanol Prices
We now investigate the empirical relationship between flex cars and the (absolute) spread
CR IP T
between gasoline and ethanol prices (pg − pa ). Our theoretical results in sub-section 3.3 suggest that price spreads are weakly increasing in the share of flex cars. We now investigate if the data supports this prediction by regressing the absolute value of the spread |pg − pa | on θ, controlling for different sets of fixed effects and other variables. The results are presented in Table 5. The specifications are similar to the ones pre-
AN US
sented in our tables in sub-section 5.1. The coefficient on the penetration of flex cars is positive and statistically significant from columns (1) to (4). However, when we include city-year linear trends, the result becomes negative and loses statistical significance.13
tration on |pg − pa |.
Additional Results
ED
5.3
M
Hence, our linear regressions show mixed evidence of a positive impact of flex car pene-
In this sub-section, we first study the robustness of our main results to a different de-
PT
pendent variable, retail margins. Panel A of Table 6 presents the results for the gasoline market. It shows a negative relationship between θ and gasoline margins in all specifi-
CE
cations. Once again, when we compare columns (1) and (2) we can see that the result becomes stronger after we implement our IV strategy. In column (5), our preferred speci-
AC
fication with the full set of controls, a 10 percentage point increase in flex car penetration reduces gasoline margins by 2.33/0.7 = 3.32 cents (of Brazilian Reais) per litre. The magnitude of the coefficient of θ in column (5), Panel A, is similar to the one
13
As in the gasoline case and differently from the ethanol one (see columns 4 and 5 of Tables 3 and 4), the switch of sign in the last column of Table 5 suggests that there are important local components affecting the price spread between the two fuels. Noting that gasoline prices are on average higher than ethanol ones during our period of analysis, the change of sign may be due to local unobserved shocks that are (positively) correlated with gasoline prices but uncorrelated with ethanol ones, and that bias the coefficients upward in the first four columns of the table (but not in column 5). Any local demand trend that affects solely gasoline markets will generate this effect. For example, gasoline car owners may be richer (as ethanol cars are considerably older) and more likely to experience positive income shocks (possibly heterogeneous across cities) when compared to ethanol car owners.
26
ACCEPTED MANUSCRIPT
Table 5: Effect of Flex Car Penetration on Spread between Gasoline and Ethanol Prices
Stations/Cars Hotels per km2
(5) 2SLS
Absolute Price Spread (|pg − pa |) 0.216*** 0.357*** 0.374*** 0.368*** -0.254 (0.065) (0.104) (0.104) (0.140) (0.305) -0.001 -0.001 (0.001) (0.001) -0.008* -0.020*** (0.005) (0.007) -0.075 -0.053 (0.047) (0.050) 854 864 723 144 2532 2532 2532 2406 2406 260383 260383 260383 252076 252076 Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes No No Yes Yes Yes No No No No Yes
ED
KP F-Stat (1st Stage) Nclusters Observations Monthly Fixed Effects Station Fixed Effects Brand Fixed Effects City-Year Trends
(4) 2SLS
AN US
GDP per Capita
(3) 2SLS
M
% Flex Cars (θ)
(2) 2SLS
CR IP T
(1) OLS
AC
CE
PT
NOTES: Table displays estimated effects of shares of flex cars, θ, on absolute retail price spreads (retail gasoline prices minus retail ethanol prices). Regressions consider a station i located in municipality m in week t as the unit of analysis, and the sample period goes from January 2002 to March 2008. All prices are expressed in BRL real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). The price of gasoline is multiplied by a factor of 0.7 so that prices per unit of energy are directly comparable across fuels. θ represents the share of flex cars in municipality m in a given month. Column 1 estimated by OLS and columns 2-5 by 2SLS. All columns include station and monthly fixed effects as controls. Columns 3-5 include station brand fixed effects. Columns 4 and 5 include yearly municipal GDP per capita, the monthly number of stations divided by the car fleet at the municipality level, and the yearly number of hotels per square kilometer in a municipality as controls. Column 5 includes P city-level linear time trends (year as time unit). Instrument for share of flex cars, IVmt , is equal to k nkm θ˜kt , where nkm is the number of dealers of car manufacturer k in city m before 2000 that were still active in 2015, and θ˜kt is the share of flex models offered by manufacturer k in the Brazilian territory in a particular month. Standard errors clustered by city-month in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
27
ACCEPTED MANUSCRIPT
Table 6: Effect of Flex Car Penetration on Gasoline Margins and Costs
-0.197*** (0.027)
GDP per Capita Stations/Cars Hotels per km2 KP F-Stat (1st Stage) Nclusters Observations Panel B % Flex Cars (θ)
2528 195457
-0.021 (0.015)
GDP per Capita
Hotels per km2
PT
ED
KP F-Stat (1st Stage) Nclusters Observations Monthly Fixed Effects Station Fixed Effects Brand Fixed Effects City-Year Trends
2528 195457 Yes Yes No No
(4) 2SLS
(5) 2SLS
Gasoline Margin (pg − cg ) -0.237*** -0.238*** -0.147** -0.233** (0.048) (0.048) (0.067) (0.108) 0.001*** -0.001*** (0.000) (0.000) -0.011*** -0.006** (0.003) (0.003) -0.022 -0.039 (0.028) (0.027) 796 804 587 150 2528 2528 2402 2402 195457 195457 189894 189894 Gasoline Cost (cg ) 0.031 0.032 0.114*** (0.028) (0.028) (0.038) 0.001*** (0.000) -0.008*** (0.002) 0.002 (0.017) 796 804 587 2528 2528 2402 195457 195457 189894 Yes Yes Yes Yes Yes Yes No Yes Yes No No No
M
Stations/Cars
(3) 2SLS
CR IP T
Panel A % Flex Cars (θ)
(2) 2SLS
AN US
(1) OLS
0.009 (0.072) -0.000 (0.000) -0.008*** (0.002) 0.013 (0.018) 150 2402 189894 Yes Yes Yes Yes
AC
CE
NOTES: Table displays estimated effects of shares of flex cars, θ, on gasoline margins (retail minus wholesale prices - Panel A) and wholesale prices (Panel B). Regressions consider a station i located in municipality m in week t as the unit of analysis, and the sample period goes from January 2002 to March 2008. All prices are expressed in BRL real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). The price of gasoline is multiplied by a factor of 0.7 so that prices per unit of energy are directly comparable across fuels. θ represents the share of flex cars in municipality m in a given month. Column 1 estimated by OLS and columns 2-5 by 2SLS. All columns include station and monthly fixed effects as controls. Columns 3-5 include station brand fixed effects. Columns 4 and 5 include yearly municipal GDP per capita, the monthly number of stations divided by the car fleet at the municipality level, and the yearly number of hotels per square kilometer in a municipality as controls. Column 5 includes Pcity-level linear time trends (year as time unit). Instrument for share of flex cars, IVmt , is equal to k nkm θ˜kt , where nkm is the number of dealers of car manufacturer k in city m before 2000 that were still active in 2015, and θ˜kt is the share of flex models offered by manufacturer k in the Brazilian territory in a particular month. Standard errors clustered by city-month in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
28
ACCEPTED MANUSCRIPT
Table 7: Effect of Flex Car Penetration on Ethanol Margins and Costs
-0.265*** (0.081)
GDP per Capita Stations/Cars Hotels per km2 KP F-Stat (1st Stage) Nclusters Observations Panel B % Flex Cars (θ)
2524 151615
-0.477*** (0.069)
GDP per Capita
Hotels per km2
PT
ED
KP F-Stat (1st Stage) Nclusters Observations Monthly Fixed Effects Station Fixed Effects Brand Fixed Effects City-Year Trends
2524 151615 Yes Yes No No
(4) 2SLS
(5) 2SLS
Ethanol Margin (pa − ca ) -0.908*** -0.912*** -1.143*** -0.529* (0.113) (0.113) (0.153) (0.276) -0.001 0.000 (0.001) (0.001) 0.028*** 0.021** (0.006) (0.009) -0.057 -0.151*** (0.045) (0.047) 756 766 568 158 2524 2524 2398 2398 151615 151615 146789 146789 Ethanol Cost (ca ) -0.008 -0.009 0.150 (0.120) (0.120) (0.164) -0.001 (0.001) -0.041*** (0.006) 0.033 (0.056) 756 766 568 2524 2524 2398 151615 151615 146789 Yes Yes Yes Yes Yes Yes No Yes Yes No No No
M
Stations/Cars
(3) 2SLS
CR IP T
Panel A % Flex Cars (θ)
(2) 2SLS
AN US
(1) OLS
-0.624** (0.262) -0.003*** (0.001) -0.042*** (0.010) 0.077 (0.061) 158 2398 146789 Yes Yes Yes Yes
AC
CE
NOTES: Table displays estimated effects of shares of flex cars, θ, on ethanol margins (retail minus wholesale prices - Panel A) and wholesale prices (Panel B). Regressions consider a station i located in municipality m in week t as the unit of analysis, and the sample period goes from January 2002 to March 2008. All prices are expressed in BRL real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). The price of gasoline is multiplied by a factor of 0.7 so that prices per unit of energy are directly comparable across fuels. θ represents the share of flex cars in municipality m in a given month. Column 1 estimated by OLS and columns 2-5 by 2SLS. All columns include station and monthly fixed effects as controls. Columns 3-5 include station brand fixed effects. Columns 4 and 5 include yearly municipal GDP per capita, the monthly number of stations divided by the car fleet at the municipality level, and the yearly number of hotels per square kilometer in a municipality as controls. Column 5 includes Pcity-level linear time trends (year as time unit). Instrument for share of flex cars, IVmt , is equal to k nkm θ˜kt , where nkm is the number of dealers of car manufacturer k in city m before 2000 that were still active in 2015, and θ˜kt is the share of flex models offered by manufacturer k in the Brazilian territory in a particular month. Standard errors clustered by city-month in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
29
ACCEPTED MANUSCRIPT found in column (5) of Table 3. This is evidence that flex cars do not affect gasoline wholesale prices substantially. We further investigate this issue by regressing gasoline wholesale prices on our variable of interest (and on other controls). The results are shown in Panel B of Table 6. In fact, flex cars penetration does not seem to impact gasoline costs, as shown by the non-statistically significant coefficients of flex car penetration in most of our specifications, including our preferred one in column (5). The absence of
CR IP T
effect is not surprising, since gasoline wholesale prices are settled by international market makers and/or Petrobras.
Table 7 presents the same exercise for the ethanol market. We still find a negative and significant effect of θ on ethanol retail margins. However, we find some evidence of the effect of flex cars on ethanol wholesale prices. In Panel B, most of the specifications show
AN US
statistically insignificant coefficients (with different signs), but our preferred specification shows a negative and significant effect. This may indicate that ethanol producer prices in Brazil are more sensitive to changes in the domestic demand for this product (when compared to gasoline).
Concluding Remarks
M
6
ED
In this work, we investigate how the penetration of flex cars has affected the fuel retail market in the state of Rio de Janeiro. Our main hypothesis was that flex car penetra-
PT
tion has increased the degree of substitution between gasoline and ethanol and that fuel stations would respond strategically to this shock, reducing retail prices in equilibrium.
CE
Our estimates suggest that the model prediction is correct. A 10 percentage point increase in the share of flex cars decrease ethanol and gasoline energy equivalent prices
AC
per liter by approximately 8 and 3 cents, respectively. Considering the volume of sales and the size of the flex car fleet in 2007, a rough estimate suggests consumer savings of almost 80 million Reais in the Rio de Janeiro state that year. Our estimates also show that the absolute retail price gap between the two fuels has increased with the adoption of flex cars, also as predicted by the model. Our results show that new technologies that increase consumer choice may benefit even those consumers who choose not to adopt them. This is an important potential welfare-improving channel to be taken into consideration by policy makers who aim to
30
ACCEPTED MANUSCRIPT incentivize the use of new technologies, especially when considering dissemination of alternative automobile fuels, such as hydrogen and electricity.
References Anderson, Soren T. 2012. “The demand for ethanol as a gasoline substitute.” Journal
CR IP T
of Environmental Economics and Management, 63(2): 151 – 168. ANP. 2010. “Anu´ario Estat´ıstico Brasileiro do Petr´oleo e do G´as Natural 2010.” http://www.anp.gov.br/?dw=33213, accessed on October 20th, 2010.
Autor, David H., David Dorn, and Gordon H. Hanson. 2013. “The China Syn-
AN US
drome: Local Labor Market Effects of Import Competition in the United States.” American Economic Review, 103(6): 2121–68.
Bartik, Timothy J. 1991. Who Benefits from State and Local Economic Development Policies? Books from Upjohn Press, W.E. Upjohn Institute for Employment Research.
M
Blanchard, Olivier Jean, and Lawrence F. Katz. 1992. “Regional Evolutions.”
ED
Brookings Papers on Economic Activity, 23(1): 1–76. Boff, H. P. 2011. “Modeling the Brazilian Ethanol Market: How Flex-Fuel Vehicles are
PT
Shaping the Long Run Equilibrium .” China-USA Business Review, 10(4): 245–264. Bresnahan, Timothy F. 1982. “The oligopoly solution concept is identified.” Economics
CE
Letters, 10(1): 87–92.
Busse, Meghan R, Christopher R Knittel, and Florian Zettelmeyer. 2013. “Are
AC
consumers myopic? Evidence from new and used car purchases.” The American Economic Review, 103(1): 220–256.
Byrne, David P, and Nicolas de Roos. 2017. “Consumer search in retail gasoline markets.” The Journal of Industrial Economics, 65(1): 183–193. CADE. 2014. “Varejo de Gasolina.” Cadernos do Cade. Carlson, J.A., and R.P. McAfee. 1983. “Discrete equilibrium price dispersion.” The Journal of Political Economy, 91(3): 480–493. 31
ACCEPTED MANUSCRIPT Chandra, Ambarish, and Mariano Tappata. 2011a. “Consumer search and dynamic price dispersion: an application to gasoline markets.” The RAND Journal of Economics, 42(4): 681–704. Chandra, Ambarish, and Mariano Tappata. 2011b. “Consumer search and dynamic price dispersion: an application to gasoline markets.” The RAND Journal of Economics,
CR IP T
42(4): 681–704. Corts, Kenneth S. 2010. “Building out alternative fuel retail infrastructure: Government fleet spillovers in {E85}.” Journal of Environmental Economics and Management, 59(3): 219 – 234.
Costa, Francisco, Jason Garred, and Jo˜ ao Paulo Pessoa. 2016. “Winners and
AN US
losers from a commodities-for-manufactures trade boom.” Journal of International Economics, 102: 50 – 69.
Ferreira, A. L., F. P. de A. Prado, and J. J. da Silveira. 2009. “Flex cars and the
M
alcohol price.” Energy Economics, 31(3): 382–394.
Firgo, Matthias, Dieter Pennerstorfer, and Christoph R Weiss. 2015. “Centrality
ED
and pricing in spatially differentiated markets: The case of gasoline.” International Journal of Industrial Organization, 40: 81–90.
PT
Goldberg, Pinelopi Koujianou. 1998. “The effects of the corporate average fuel efficiency standards in the US.” The Journal of Industrial Economics, 46(1): 1–33.
CE
Hosken, Daniel S, Robert S McMillan, and Christopher T Taylor. 2008. “Retail gasoline pricing: What do we know?” International Journal of Industrial Organization,
AC
26(6): 1425–1436.
Infomoney. a
de
2012.
Minas,
“Venda
Paran´a
e
de
carros
Rio
juntas.”
no
estado
[Accessed:
de
SP 2017
´e 11
maior 30],
que http:
//www.infomoney.com.br/minhas-financas/carros/noticia/2557745/ venda-carros-estado-maior-que-minas-parana-rio-juntas. Kahn, James A. 1986. “Gasoline prices and the used automobile market: a rational expectations asset price approach.” The Quarterly Journal of Economics, 323–339.
32
ACCEPTED MANUSCRIPT Klier, line
Thomas, and
Data.”
New
and Vehicle
American
Joshua Fuel
Economic
Linn.
Economy: Journal:
2010.
“The
Evidence Economic
Price
from
of
Gaso-
Monthly
Policy,
2:
Sales
134–153.
http://www.aeaweb.org/articles.php?doi=10.1257/pol.2.3.134. Lewis, Matthew. 2008. “Price Dispersion And Competition With Differentiated Sell-
CR IP T
ers.” The Journal of Industrial Economics, 56(3): 654–678. Li, Shanjun, Christopher Timmins, and Roger H von Haefen. 2009. “How do gasoline prices affect fleet fuel economy?” American Economic Journal: Economic Policy, 1(2): 113–137.
´ Marjotta-Maistro, M. C., and G. A. Asai. 2006. “Alcool combust´ıvel: do carro a
AN US
a´lcool ao carro flex.” mimeo.
Pakes, Ariel, Steven Berry, and James A Levinsohn. 1993. “Applications and limitations of some recent advances in empirical industrial organization: price indexes and the analysis of environmental change.” The American Economic Review, 83(2): 240–
M
246.
ED
Pennerstorfer, Dieter. 2009. “Spatial price competition in retail gasoline markets: evidence from Austria.” The Annals of Regional Science, 43(1): 133–158.
PT
Pennerstorfer, Dieter, Philipp Schmidt-Dengler, Nicolas Schutz, Christoph Weiss, and Biliana Yontcheva. 2015. “Information and price dispersion: Theory
CE
and evidence.” WIFO Working paper 502/2015. Pinkse, J., M.E. Slade, and C. Brett. 2002. “Spatial price competition: A semipara-
AC
metric approach.” Econometrica, 70(3): 1111–1153.
Salvo, Alberto, and Cristian Huse. 2011. “Is Arbitrage Tying the Price of Ethanol to that of Gasoline? Evidence from the Uptake of Flexible-Fuel Technology.” The Energy Journal, Volume 32(Number 3): 119–148. Salvo, Alberto, and Cristian Huse. 2013. “Build it, but will they come? Evidence from consumer choice between gasoline and sugarcane ethanol.” Journal of Environmental Economics and Management, 66(2): 251 – 279.
33
ACCEPTED MANUSCRIPT Shriver, Scott K. 2015. “Network Effects in Alternative Fuel Adoption: Empirical Analysis of the Market for Ethanol.” Marketing Science, 34(1): 78–97. Simon, Leo, and William Zame. 1990. “Discontinuous Games and Endogenous Sharing Rules.” Econometrica, 58: 861–872. USDA.
2010.
“Sugar
and
Sweeteners
accessed
Tables.” on
Octo-
CR IP T
http://www.ers.usda.gov/Briefing/Sugar/data.htm#yearbook,
Yearbook
ber 20th, 2010.
Verboven, Frank. 2002. “Quality-based price discrimination and tax incidence: evidence from gasoline and diesel cars.” RAND Journal of Economics, 275–297.
AN US
Weyl, Eric G., and Michal Fabinger. 2009. “Pass-Through as an Economic Tool.”
AC
CE
PT
ED
M
SSRN eLibrary.
34
ACCEPTED MANUSCRIPT
Appendix A -
Other Figures and Tables
CR IP T
.3 .2
AN US
.1 0
Rank Reversal across Stations
.4
Figure A.1: Rank Reversal between Stations
0
100
200
300
400
M
Week
AC
CE
PT
ED
NOTES: Figure plots the average (across markets) of rank-reversal between stations over time. Rankreversal between stations is computed as the share of stations in a given municipality-week pair that did not choose the most likely cheapest fuel. For example, if most stations in a municipality-week pair chose ethanol as their cheapest product, our between station rank reversal measure in that market will be equal to the share of stations that charge less for gasoline than for ethanol. Sources: ANP.
35
ACCEPTED MANUSCRIPT
Table A.1: Stock of Vehicles, Percentage of Flex Cars and Number of Fuel Stations, by City
AC
CE
% flex
CR IP T
14.1% 12.4% 11.1% 10.5% 10.5% 10.0% 9.1% 9.0% 8.9% 8.6% 7.4% 7.3% 7.2% 7.1% 6.9% 6.9% 6.7% 6.5% 6.5% 6.3% 6.2% 6.1% 5.6% 5.6% 5.5% 5.5% 5.3% 5.3% 4.6% 4.5% 4.4% 4.4% 4.3% 4.3% 4.2% 3.8% 2.6% 2.3% 8.8%
Number of fuel stations In the Total % sampled sample (2010) weekly 10 11 91% 34 85 40% 14 23 61% 190 805 24% 7 7 100% 13 23 57% 7 11 64% 11 22 50% 14 26 54% 10 23 43% 9 10 90% 10 15 67% 16 23 70% 9 13 69% 21 31 68% 6 10 60% 21 30 70% 7 7 100% 35 88 40% 18 54 33% 10 13 77% 11 19 58% 23 31 74% 7 7 100% 25 65 38% 10 13 77% 30 88 34% 21 38 55% 8 13 62% 10 15 67% 20 33 61% 11 17 65% 20 34 59% 10 16 63% 10 16 63% 16 20 80% 38 109 35% 8 10 80% 750 1874 40%
AN US
M
ED
PT
Mangaratiba Niter´oi Maca´e Rio de Janeiro A. de B´ uzios Angra dos Reis Parati Maric´a Resende Cabo Frio Nil´opolis Trˆes Rios Araruama Saquarema Barra Mansa Vassouras Volta Redonda Para´ıba do Sul S˜ao Gon¸calo Petr´opolis Sapucaia Belford Roxo Teres´opolis Queimados Nova Igua¸cu Mag´e Duque de Caxias Nova Friburgo Itagua´ı Valen¸ca S. J. de Meriti Rio Bonito Itabora´ı Barra do Pira´ı S. Ant de Padua Itaperuna Campos dos Goit. S. Fr. de Itabapoana Total
2004 2007 Number of % flex Number of vehicles vehicles 3 090 1.2% 4 812 176 647 1.0% 194 511 41 048 1.1% 57 261 1 800 614 0.8% 1 969 128 4 340 1.4% 6 703 21 503 0.9% 27 030 2 937 0.6% 3 675 14 193 0.7% 24 120 26 085 0.8% 33 368 32 315 1.0% 46 371 24 305 0.5% 29 579 14 792 0.9% 18 246 20 290 0.6% 28 037 9 955 0.6% 13 771 31 788 0.8% 36 420 7 297 1.0% 8 319 68 441 0.7% 81 203 6 101 0.8% 7 427 107 977 0.4% 134 860 86 912 0.6% 97 177 1 381 0.9% 1 508 26 574 0.3% 37 744 42 920 0.5% 50 787 9 278 0.3% 13 612 115 523 0.4% 137 337 21 034 0.4% 27 495 128 426 0.3% 151 257 60 539 0.4% 68 588 28 016 0.2% 31 609 10 297 0.3% 11 540 64 607 0.2% 76 891 22 251 0.5% 35 453 27 800 0.3% 37 075 18 436 0.5% 21 239 9 234 0.7% 10 540 18 810 0.6% 22 123 91 651 0.3% 109 388 3 395 0.3% 4 204 3 200 802 0.7% 3 670 408
NOTES: Table shows stocks and percentages of flex cars by municipality in the state of Rio de Janeiro, in 2004 and in 2007. It also shows the total number of stations by city in 2010, the average number of stations by municipality in a week in our sample and the ratio between the two measures, i.e., the average percentage of stations sampled weekly by municipality. Sources: Detran RJ and ANP.
36
ACCEPTED MANUSCRIPT
Table A.2: Fuel Wholesale and Retail Prices and Margins (in R$) Ethanol Price Ethanol Wholesale Price Ethanol Margin Gasoline Price Gasoline Wholesale Price Gasoline Margin
Observations 265988 154158 154158 286077 198829 198829
Mean 1.144 0.952 0.193 1.734 1.537 0.198
St. Deviation 0.189 0.202 0.105 0.125 0.095 0.078
Min 0.529 0.223 -0.775 1.287 0.855 -0.557
Max 2.035 2.126 0.817 2.287 2.277 1.073
CR IP T
NOTES: Table shows usual descriptive statistics for fuel wholesale and retail prices and margins in our sample. All prices are expressed in BRL real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). Sources: ANP.
Table A.3: Effect of Flex Car Penetration on Gasoline Price - Competition
% Flex Cars (θ)
-0.147*** (0.031)
GDP per Capita
ED
Hotels per km2
CE
PT
KP F-Stat (1st Stage) Nclusters Observations Monthly Fixed Effects Station Fixed Effects Brand Fixed Effects City-Year Trends
2055 238858 Yes Yes No No
(3) 2SLS
(4) 2SLS
Gasoline Price (pg ) -0.223*** -0.223*** -0.014 (0.048) (0.048) (0.065) 0.002*** (0.000) -0.020*** (0.003) -0.033 (0.026) 809 815 614 2055 2055 1974 238858 238858 232940 Yes Yes Yes Yes Yes Yes No Yes Yes No No No
M
Stations/Cars
(2) 2SLS
AN US
(1) OLS
(5) 2SLS
-0.231** (0.113) -0.001** (0.000) -0.015*** (0.003) -0.002 (0.022) 146 1974 232940 Yes Yes Yes Yes
AC
NOTES: Table displays estimated effects of shares of flex cars, θ, on gasoline retail prices excluding from the sample city-week pairs where either gasoline or ethanol is the cheapest fuel in all stations. Regressions consider a station i located in municipality m in week t as the unit of analysis, and the sample period goes from January 2002 to March 2008. All prices are expressed in BRL real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). The price of gasoline is multiplied by a factor of 0.7 so that prices per unit of energy are directly comparable across fuels. θ represents the share of flex cars in municipality m in a given month. Column 1 estimated by OLS and columns 2-5 by 2SLS. All columns include station and monthly fixed effects as controls. Columns 3-5 include station brand fixed effects. Columns 4 and 5 include yearly municipal GDP per capita, the monthly number of stations divided by the car fleet at the municipality level, and the yearly number of hotels per square kilometer in a municipality as controls. Column 5 includes city-level linear time trends (year as time unit). Instrument for share of flex cars, IVmt , is equal P to k nkm θ˜kt , where nkm is the number of dealers of car manufacturer k in city m before 2000 that were still active in 2015, and θ˜kt is the share of flex models offered by manufacturer k in the Brazilian territory in a particular month. Standard errors clustered by city-month in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
37
ACCEPTED MANUSCRIPT
Table A.4: Effect of Flex Car Penetration on Ethanol Price - Competition
-0.601*** (0.085)
GDP per Capita Stations/Cars Hotels per km2
M
2055 219741 Yes Yes No No
(4) 2SLS
Ethanol Price (pa ) -0.676*** -0.687*** -0.645*** (0.147) (0.147) (0.200) -0.000 (0.001) -0.012* (0.006) -0.038 (0.064) 836 844 650 2055 2055 1974 219741 219741 214048 Yes Yes Yes Yes Yes Yes No Yes Yes No No No
ED
KP F-Stat (1st Stage) Nclusters Observations Monthly Fixed Effects Station Fixed Effects Brand Fixed Effects City-Year Trends
(3) 2SLS
(5) 2SLS
CR IP T
% Flex Cars (θ)
(2) 2SLS
AN US
(1) OLS
-0.906*** (0.344) -0.003*** (0.001) -0.014 (0.009) -0.019 (0.057) 148 1974 214048 Yes Yes Yes Yes
AC
CE
PT
NOTES: Table displays estimated effects of shares of flex cars, θ, on ethanol retail prices excluding from the sample city-week pairs where either gasoline or ethanol is the cheapest fuel in all stations. Regressions consider a station i located in municipality m in week t as the unit of analysis, and the sample period goes from January 2002 to March 2008. All prices are expressed in BRL real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). The price of gasoline is multiplied by a factor of 0.7 so that prices per unit of energy are directly comparable across fuels. θ represents the share of flex cars in municipality m in a given month. Column 1 estimated by OLS and columns 2-5 by 2SLS. All columns include station and monthly fixed effects as controls. Columns 3-5 include station brand fixed effects. Columns 4 and 5 include yearly municipal GDP per capita, the monthly number of stations divided by the car fleet at the municipality level, and the yearly number of hotels per square kilometer in a municipality as controls. Column 5 includes city-level linear time trends (year as time unit). Instrument for share of flex cars, IVmt , is equal P to k nkm θ˜kt , where nkm is the number of dealers of car manufacturer k in city m before 2000 that were still active in 2015, and θ˜kt is the share of flex models offered by manufacturer k in the Brazilian territory in a particular month. Standard errors clustered by city-month in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
38
ACCEPTED MANUSCRIPT
Table A.5: Effect of Flex Car Penetration on Gasoline Price - State Capital
-0.162*** (0.039)
GDP per Capita Stations/Cars Hotels per km2
(5) 2SLS
M
2458 191123 Yes Yes No No
(4) 2SLS
Gasoline Price (pg ) -0.437*** -0.442*** -0.365*** -0.602*** (0.119) (0.118) (0.130) (0.233) 0.001*** -0.002*** (0.000) (0.000) -0.021*** -0.014*** (0.003) (0.003) -0.005 -0.047** (0.024) (0.023) 138 142 116 75 2458 2458 2335 2335 191123 191123 184455 184455 Yes Yes Yes Yes Yes Yes Yes Yes No Yes Yes Yes No No No Yes
ED
KP F-Stat (1st Stage) Nclusters Observations Monthly Fixed Effects Station Fixed Effects Brand Fixed Effects City-Year Trends
(3) 2SLS
CR IP T
% Flex Cars (θ)
(2) 2SLS
AN US
(1) OLS
AC
CE
PT
NOTES: Table displays estimated effects of shares of flex cars, θ, on gasoline retail prices excluding from the sample the city of Rio de Janeiro (the capital of the state of RJ). Regressions consider a station i located in municipality m in week t as the unit of analysis, and the sample period goes from January 2002 to March 2008. All prices are expressed in BRL real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). The price of gasoline is multiplied by a factor of 0.7 so that prices per unit of energy are directly comparable across fuels. θ represents the share of flex cars in municipality m in a given month. Column 1 estimated by OLS and columns 2-5 by 2SLS. All columns include station and monthly fixed effects as controls. Columns 3-5 include station brand fixed effects. Columns 4 and 5 include yearly municipal GDP per capita, the monthly number of stations divided by the car fleet at the municipality level, and the yearly number of hotels per square kilometer in a municipality as controls. Column 5 includes Pcity-level linear time trends (year as time unit). Instrument for share of flex cars, IVmt , is equal to k nkm θ˜kt , where nkm is the number of dealers of car manufacturer k in city m before 2000 that were still active in 2015, and θ˜kt is the share of flex models offered by manufacturer k in the Brazilian territory in a particular month. Standard errors clustered by city-month in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
39
ACCEPTED MANUSCRIPT
Table A.6: Effect of Flex Car Penetration on Ethanol Price - State Capital
-0.479*** (0.093)
GDP per Capita Stations/Cars Hotels per km2
M
2458 180090 Yes Yes No No
(4) 2SLS
Ethanol Price (pa ) -0.717*** -0.730*** -0.650*** (0.224) (0.222) (0.249) -0.001 (0.001) -0.023*** (0.007) 0.066 (0.061) 156 160 132 2458 2458 2335 180090 180090 173589 Yes Yes Yes Yes Yes Yes No Yes Yes No No No
ED
KP F-Stat (1st Stage) Nclusters Observations Monthly Fixed Effects Station Fixed Effects Brand Fixed Effects City-Year Trends
(3) 2SLS
(5) 2SLS
CR IP T
% Flex Cars (θ)
(2) 2SLS
AN US
(1) OLS
0.243 (0.573) -0.004*** (0.001) -0.021** (0.009) 0.023 (0.062) 78 2335 173589 Yes Yes Yes Yes
AC
CE
PT
NOTES: Table displays estimated effects of shares of flex cars, θ, on ethanol retail prices excluding from the sample the city of Rio de Janeiro (the capital of the state of RJ). Regressions consider a station i located in municipality m in week t as the unit of analysis, and the sample period goes from January 2002 to March 2008. All prices are expressed in BRL real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). The price of gasoline is multiplied by a factor of 0.7 so that prices per unit of energy are directly comparable across fuels. θ represents the share of flex cars in municipality m in a given month. Column 1 estimated by OLS and columns 2-5 by 2SLS. All columns include station and monthly fixed effects as controls. Columns 3-5 include station brand fixed effects. Columns 4 and 5 include yearly municipal GDP per capita, the monthly number of stations divided by the car fleet at the municipality level, and the yearly number of hotels per square kilometer in a municipality as controls. Column 5 includes Pcity-level linear time trends (year as time unit). Instrument for share of flex cars, IVmt , is equal to k nkm θ˜kt , where nkm is the number of dealers of car manufacturer k in city m before 2000 that were still active in 2015, and θ˜kt is the share of flex models offered by manufacturer k in the Brazilian territory in a particular month. Standard errors clustered by city-month in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
40
ACCEPTED MANUSCRIPT
Table A.7: Effect of Flex Car Penetration on Gasoline Price - Controlling for Costs
Gasoline Cost (cg )
-0.211*** (0.028) 0.345*** (0.011)
GDP per Capita Stations/Cars Hotels per km2
(4) 2SLS
(5) 2SLS
Gasoline Price (pg ) -0.217*** -0.217*** -0.072 -0.227** (0.046) (0.046) (0.065) (0.103) 0.345*** 0.343*** 0.342*** 0.323*** (0.011) (0.011) (0.011) (0.011) 0.001*** -0.001*** (0.000) (0.000) -0.016*** -0.012*** (0.003) (0.003) -0.020 -0.031 (0.025) (0.021) 797 805 588 150 2528 2528 2402 2402 195457 195457 189894 189894 Yes Yes Yes Yes Yes Yes Yes Yes No Yes Yes Yes No No No Yes
M
2528 195457 Yes Yes No No
ED
KP F-Stat (1st Stage) Nclusters Observations Monthly Fixed Effects Station Fixed Effects Brand Fixed Effects City-Year Trends
(3) 2SLS
CR IP T
% Flex Cars (θ)
(2) 2SLS
AN US
(1) OLS
AC
CE
PT
NOTES: Table displays estimated effects of shares of flex cars, θ, on gasoline retail prices. Regressions consider a station i located in municipality m in week t as the unit of analysis, and the sample period goes from January 2002 to March 2008. All prices are expressed in BRL real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). The price of gasoline is multiplied by a factor of 0.7 so that prices per unit of energy are directly comparable across fuels. θ represents the share of flex cars in municipality m in a given month. Column 1 estimated by OLS and columns 2-5 by 2SLS. All columns include gasoline wholesale prices, station and monthly fixed effects as controls. Columns 3-5 include station brand fixed effects. Columns 4 and 5 include yearly municipal GDP per capita, the monthly number of stations divided by the car fleet at the municipality level, and the yearly number of hotels per square kilometer in a municipality as controls. Column 5 includes P city-level linear time trends (year as time unit). Instrument for share of flex cars, IVmt , is equal to k nkm θ˜kt , where nkm is the number of dealers of car manufacturer k in city m before 2000 that were still active in 2015, and θ˜kt is the share of flex models offered by manufacturer k in the Brazilian territory in a particular month. Standard errors clustered by city-month in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
41
ACCEPTED MANUSCRIPT
Table A.8: Effect of Flex Car Penetration on Ethanol Price - Controlling for Costs
-0.576*** (0.079) 0.348*** (0.007)
Ethanol Cost (ca ) GDP per Capita Stations/Cars Hotels per km2
(4) 2SLS
Ethanol Price (pa ) -0.913*** -0.917*** -1.045*** (0.134) (0.134) (0.178) 0.347*** 0.346*** 0.345*** (0.007) (0.007) (0.007) -0.001* (0.001) 0.001 (0.006) -0.035 (0.047) 759 769 572 2524 2524 2398 151615 151615 146789 Yes Yes Yes Yes Yes Yes No Yes Yes No No No
M
2524 151615 Yes Yes No No
ED
KP F-Stat (1st Stage) Nclusters Observations Monthly Fixed Effects Station Fixed Effects Brand Fixed Effects City-Year Trends
(3) 2SLS
(5) 2SLS
CR IP T
% Flex Cars (θ)
(2) 2SLS
AN US
(1) OLS
-0.941*** (0.293) 0.340*** (0.007) -0.002** (0.001) -0.007 (0.008) -0.100** (0.043) 158 2398 146789 Yes Yes Yes Yes
AC
CE
PT
NOTES: Table displays estimated effects of shares of flex cars, θ, on ethanol retail prices. Regressions consider a station i located in municipality m in week t as the unit of analysis, and the sample period goes from January 2002 to March 2008. All prices are expressed in BRL real terms, deflated by the monthly ´Indice de Pre¸cos ao Consumidor Amplo - IPCA (the Brazilian version of the Consumer Price Index - CPI). The price of gasoline is multiplied by a factor of 0.7 so that prices per unit of energy are directly comparable across fuels. θ represents the share of flex cars in municipality m in a given month. Column 1 estimated by OLS and columns 2-5 by 2SLS. All columns include ethanol wholesale prices, station and monthly fixed effects as controls. Columns 3-5 include station brand fixed effects. Columns 4 and 5 include yearly municipal GDP per capita, the monthly number of stations divided by the car fleet at the municipality level, and the yearly number of hotels per square kilometer in a municipality as controls. Column 5 includes P city-level linear time trends (year as time unit). Instrument for share of flex cars, IVmt , is equal to k nkm θ˜kt , where nkm is the number of dealers of car manufacturer k in city m before 2000 that were still active in 2015, and θ˜kt is the share of flex models offered by manufacturer k in the Brazilian territory in a particular month. Standard errors clustered by city-month in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
42
ACCEPTED MANUSCRIPT
Appendix B -
Estimating the Price Formation Model
In this section, we estimate price best response functions using our datase and use them to estimate the parameters of the price formation model we proposed in Section 3. To bring the theory to the data, we employ the richer version of the model presented in section 3.4, recognizing that demand functions may differ across fuel types. Further, we allow for heterogeneity across markets.
qijt = αmj − βj ptij + γj p¯tij ,
CR IP T
Let the demand for fuel from a vehicle of type j from station i in market m be
(5)
where αmj = αm + αj represents the composition of a market-specific fixed effect and
AN US
a fuel-specific intercept.
Returning to the same analysis performed in section 3.4, we obtain the following firstorder conditions:
(6)
1 t αma γa t t t = cia + 1I{pia > pig } + p¯ 2 2βa 2βa ia # " t t t t t t θ α + θ α + θ γ p ¯ + θ γ p ¯ ma mf a f a a ia f f if + 1I{ptia < ptig } + tia . 2(θat βa + θft βf )
(7)
M
1 t γg t αmg t t = cig + 1I{pig > pia } + p¯ 2 2βg 2βg ig " # θgt αmg + θft αmf + θgt γg p¯tig + θft γf p¯tif t t + tig , + 1I{pig < pia } t t 2(θg βg + θf βf )
CE
ptia
PT
ED
ptig
AC
tij may be interpreted as an unobserved (to the econometrician) i.i.d. cost shock that
is fuel-station-time specific.14 Our objective in this section is to identify the demand parameters by estimating price
best response functions (Pinkse, Slade and Brett, 2002). Because we do not observe quantities directly, we cannot identify the absolute scale of the demand coefficients only from pricing responses, and a normalization must be made. 14
More precisely, we assume that total marginal cost for a given fuel j in station i is equal to ctij + 2tij , where the number 2 multiplying the error is simply a normalization that does not affect the results. The two estimating equations follow by plugging this in our profit equation and deriving the optimal solution.
43
ACCEPTED MANUSCRIPT For convenience, we normalize βf = 1. If we further assume that βg = βa = βf = 1, we obtain best responses that are linear in the remaining parameters, and hence, can be estimated using simpler methods (OLS, for example):
(8)
hα γa i 1 ma + p¯tia ptia = ctia + 1I{ptia > ptig } 2 2 2 # " t t t t t t γ p ¯ + θ γ p ¯ α + θ α + θ θ f a mf ma ia a a if f f + 1I{ptia < ptig } + tia . 2(θat + θft )
(9)
AN US
CR IP T
hα 1 γg i mg ptig = ctig + 1I{ptig > ptia } + p¯tig 2 2 2 # " t t t t t t γ p ¯ + θ γ p ¯ α + θ α + θ θ f g mf mg ig g g if f f + 1I{ptig < ptia } + tig , 2(θgt + θft )
We report estimates both for the case where β coefficients are assumed to be identical (our linear case) and for the more general non-linear case.
M
To estimate these models, we must address two additional issues. First, we must define the relevant market for each fuel station. One possible approach is to define a
ED
market as a municipality; however, in the case of large cities, this definition appears to be inappropriate. Rio de Janeiro, the state capital, has 805 fuel stations spread over 1,250
PT
square kilometers; it is unreasonable to assume that they are all competing in the same market. To account for local competition in a simple manner, we define a market to consist
CE
of all stations that share the same four-digit postal code (CEP). We also experiment with narrowing the market definition to five-digit CEP areas.15
AC
Representing a second challenge is endogeneity. If there are stochastic unobserved components in the demand function or in the marginal cost, because all prices are determined in equilibrium, all terms on the right-hand side of the form p¯ij or 1I{pij < pik } are
endogenous.
To address this problem, we follow Pinkse, Slade and Brett (2002) and instrument 15
In Brazil, the postal code has 8 digits, with the first five dividing the country into increasingly fine partitions. The four-digit level corresponds to neighborhoods in large cities and to small municipalities; at the five-digit level, neighborhoods of large cities are divided into several areas. Three-digit areas are too coarse for the purposes of our model, as some of these areas cover different municipalities in our sample; conversely, eight-digit areas are too narrow, as most gas stations in the sample would be considered monopolies.
44
ACCEPTED MANUSCRIPT each endogenous regressor by the analogous term involving costs: that is, we substitute 1I{ctij < ctik } for 1I{ptij < ptik }, etc. We also instrument for the share of flex cars using the variable IVmt described in the previous section.
In addition, we estimate the model with market fixed effects to account for unobserved variation across markets. We use municipality-specific fixed effects to avoid the computational burden of estimating the non-linear model with a large number of 4-digit CEP
CR IP T
fixed effects. Finally, we may estimate price reaction functions separately for each fuel type (which will yield two different sets of estimates for the demand from flex car owners) or stack the data to impose the restriction that αf and γf must be the same in both regressions. Table B.1 presents the results of estimating the linear model separately and jointly,
AN US
respectively. Panel A show the results of the linear model, while Panel B presents the results of the non-linear estimation. In columns (1), (2), and (3) we do not impose the theoretical restriction that the coefficients on cost should be 0.5. A remarkable finding is that our estimates of this effect are near this figure; they fluctuate between 0.35 and 0.54 among all specifications. (However, because these effects are precisely estimated, they
M
are statistically different from 0.5.) This fact suggests that pricing in this market does indeed comply with the logic of a price-setting oligopoly game. This coefficient is also of
ED
independent interest: the fact that is less than unity means that demand is cost-absorbing and, according to Weyl and Fabinger (2009), a number of comparative statics predictions
PT
may be derived from that fact. For example, the entry of a new station will necessarily reduce prices, and a merger (without synergies) will raise prices of all firms (Weyl and
CE
Fabinger, 2009, theorem 4).
In column (4), we present estimates of the model obtained when we impose the the-
AC
oretical restriction that the coefficients on costs should be 0.5. This is our preferred specification.
In our model, the elasticity of demand is proportional to the difference between βj
and γj . In our linear model in panel A, we estimate that γF is 37% smaller than βF , γA is 12% smaller than βA , and γG statistically equal to βG (these numbers are similar to the ones in the non-linear model); our estimated demands are inelastic. This finding suggests that gasoline and ethanol car owners are less responsive to fluctuations in prices across fuel stations within the market.
45
ACCEPTED MANUSCRIPT Table B.1: Structural Estimation (1) Panel A cG
αG
αA γA αF
AN US
γF
Panel B cG cA
γG
PT
ED
αA
CE
βA
βG
AC
Non-Linear Model - βF 0.351*** 0.543*** (0.029) (0.014) 0.496*** 0.473*** (0.008) (0.012) 0.354*** 0.170*** (0.041) (0.012) 1.112*** 1.049*** (0.021) (0.012) 0.487*** 0.277*** (0.014) (0.019) 1.022*** 1.046*** (0.013) (0.015) -3.171*** 0.595*** 0.431*** (0.024) (0.011) (0.018) 4.726*** 0.601*** 0.911*** (0.030) (0.012) (0.021) 1.174*** 1.025*** (0.016) (0.034) 1.008*** 1.104*** (0.003) (0.019) No Yes No Yes No No No No No
M
αG
γF
(4)
CR IP T
γG
αF
(3)
Linear Model - βA = βG = βF = 1 0.455*** 0.475*** 0.5 (0.016) (0.012) 0.532*** 0.508*** 0.5 (0.010) (0.008) 0.198*** 0.151*** 0.162*** (0.026) (0.017) (0.021) 1.050*** 1.058*** 1.008*** (0.041) (0.029) (0.016) 0.378*** 0.374*** (0.014) (0.015) 0.857*** 0.878*** (0.019) (0.012) -3.594** 0.548* 0.320 0.564** (1.799) (0.305) (0.254) (0.261) 4.322*** 0.593** 0.829*** 0.625*** (1.448) (0.272) (0.226) (0.238)
cA
γA
(2)
Single Equation: Ethanol Single Equation: Gasoline Restricted Model
=1 0.5 0.5 0.229*** (0.040) 1.073*** (0.019) 0.375*** (0.049) 0.988*** (0.035) 0.608*** (0.021) 0.644*** (0.017) 1.085*** (0.022) 1.095*** (0.027) No No Yes
NOTES: Table displays estimated parameters from the structural model. Columns 1 and 2 in Panel A (Panel B) display separate estimations of equations 8 and 9 (equations 6 and 7), respectively, by 2SLS (by GMM). Columns 3 and 4 in Panel A (Panel B) display joint estimations of equations 8 and 9 (equations 6 and 7) by 2SLS (by GMM), and in column 4 the coefficients of wholesale prices are restricted to be equal to 0.5. Estimations consider a station i located in municipality m in week t as the unit of analysis, from January 2002 to March 2008 [Observations = 117,941]. The relevant market for each fuel station is the 4-digit postal code (CEP). All prices are expressed in BRL real terms, deflated by the monthly IPCA, and the price of gasoline is multiplied by a factor of 0.7. Instruments used in the estimations consider only cig , cia , θat (share of ethanol cars in a municipality), θgt (share of gasoline cars in a municipality), IVmt (defined in Equation 4) and municipality fixed effects, as well as different interactions of these terms. For example, we use 1I{ctia < ctig }IVmt as an instrument for 1I{ptia < ptig }θft . Bootstrapped standard errors clustered by city-month in parentheses [Nclusters = 2,443]. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
46
ACCEPTED MANUSCRIPT Table B.2 presents the results we obtain if we use the narrow market definition. It shows that the coefficients in our preferred specification (column 4) have similar magnitudes to the ones presented in Table B.1. Our findings with respect to pass-through are
AC
CE
PT
ED
M
AN US
CR IP T
also similar in this case.
47
ACCEPTED MANUSCRIPT Table B.2: Structural Estimation - 5-digit CEP Market Definition (1) Panel A cG
αG
αA γA αF
AN US
γF
Panel B cG cA
γG
PT
ED
αA
CE
βA
βG
AC
Non-Linear Model - βF 0.386*** 0.552*** (0.030) (0.013) 0.513*** 0.496*** (0.008) (0.013) 0.389*** 0.195*** (0.045) (0.012) 1.035*** 0.987*** (0.022) (0.011) 0.521*** 0.350*** (0.015) (0.025) 0.973*** 1.003*** (0.015) (0.016) -3.356*** 0.843*** 0.542*** (0.026) (0.013) (0.019) 4.859*** 0.356*** 0.744*** (0.032) (0.014) (0.022) 1.180*** 1.069*** (0.017) (0.040) 1.011*** 1.075*** (0.004) (0.022) No Yes No Yes No No No No No
M
αG
γF
(4)
CR IP T
γG
αF
(3)
Linear Model - βA = βG = βF = 1 0.484*** 0.504*** 0.500 (0.015) (0.014) . 0.549*** 0.524*** 0.500 (0.008) (0.010) . 0.237*** 0.188*** 0.187*** (0.028) (0.018) (0.017) 0.975*** 0.985*** 0.992*** (0.041) (0.032) (0.012) 0.477*** 0.415*** 0.395*** (0.019) (0.015) (0.013) 0.726*** 0.807*** 0.864*** (0.024) (0.025) (0.011) -3.783** 0.798** 0.452* 0.655*** (1.806) (0.365) (0.272) (0.234) 4.450*** 0.353 0.691*** 0.549*** (1.427) (0.323) (0.238) (0.211)
cA
γA
(2)
Single Equation: Ethanol Single Equation: Gasoline Restricted Model
=1
0.265*** (0.041) 1.042*** (0.022) 0.312*** (0.046) 1.028*** (0.037) 0.726*** (0.021) 0.589*** (0.017) 1.069*** (0.021) 1.085*** (0.028) No No Yes
NOTES: Table displays estimated parameters from the structural model. Columns 1 and 2 in Panel A (Panel B) display separate estimations of equations 8 and 9 (equations 6 and 7), respectively, by 2SLS (by GMM). Columns 3 and 4 in Panel A (Panel B) display joint estimations of equations 8 and 9 (equations 6 and 7) by 2SLS (by GMM), and in column 4 the coefficients of wholesale prices are restricted to be equal to 0.5. Estimations consider a station i located in municipality m in week t as the unit of analysis, from January 2002 to March 2008 [Observations = 93,360]. The relevant market for each fuel station is the 5-digit postal code (CEP). All prices are expressed in BRL real terms, deflated by the monthly IPCA, and the price of gasoline is multiplied by a factor of 0.7. Instruments used in the estimations consider only cig , cia , θat (share of ethanol cars in a municipality), θgt (share of gasoline cars in a municipality), IVmt (defined in Equation 4) and municipality fixed effects, as well as different interactions of these terms. For example, we use 1I{ctia < ctig }IVmt as an instrument for 1I{ptia < ptig }θft . Bootstrapped standard errors clustered by city-month in parentheses [Nclusters = 2,407]. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
48