Policy Research Working Paper 10926 Effective Fuel Price in Reducing Emission Intensity A Panel Analysis for Brazil Ayan Qu Macroeconomics, Trade and Investment Global Practice & Africa Region September 2024 Policy Research Working Paper 10926 Abstract This paper studies how effective an incremental change in fixed effects models based on statistical results. The findings the price of fuel, a proxy for fuel carbon tax, is in reduc- show that (1) the price of diesel has the most significant ing the emission intensity of road transportation in Brazil and robust impact on reducing emission intensity; (2) the through panel analysis at the federative unit level from 2010 short-run and long-run elasticities of the price of diesel to 2020, after offering descriptive insights into Brazil’s auto- are −0.74 and −2.06, respectively; and (3) both entity and motive fuel market with respect to its products, actors, and time effects are significant, with the year of 2020 having a external factors. The paper postulates multiple variations consistent effect in reducing emission intensity across the of panel analysis models and focuses on dynamic two-way estimated models. This paper is a product of the Macroeconomics, Trade and Investment Global Practice and the Office of the Chief Economist, Africa Region. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http:// www.worldbank.org/prwp. The author may be contacted at aqu@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Effective Fuel Price in Reducing Emission Intensity: A Panel Analysis for Brazil Ayan Qu Keywords: fuel prices; emission intensity; transportation; panel analysis; Brazil JEL Classification: Q41; Q58; H23; C23 Acknowledgments The author would like to thank Marek Hanusch and Daniel Navia Simon for guidance, and Jon Strand, Gabriela Mundaca, Heron Marcos Teixeira Rios, and Vanessa Rahal Canado for comments/discussions/contributions. Errors are the responsibilities of the author. The author may be contacted at aqu@worldbank.org. I. Introduction Transportation is a major source of greenhouse gas (GHG) emissions in Brazil, while continued technological development and renewable energy transition present great opportunities for decarbonizing the sector. After agriculture and forestry (and their land uses), transportation is a major source of the country’s GHG emissions, accounting for about half of emissions from energy consumption (SEEG 2022). The country’s clean electricity matrix explains low emission intensity of electric power (Brazil CCDR; World Bank 2021), but it has not fully benefited the transportation sector that has only a tiny share of electric vehicles (0.02% in 2018, but it has increased rapidly to 3% in 2023 according to International Energy Agency). Meanwhile, the country is a global pioneer in biofuels (in particular, ethanol and biodiesel) that have lower emission factors than fossil fuels and whose production from biomass, if not resulting from deforestation, contributes to emission mitigation. Total GHG emissions from transportation are dominated by those from diesel and gasoline combustion. Climate inequality is another concern that also manifests inside Brazil, while the biofuel market contributes to convergence of income between the agriculture and energy sectors. The less developed regions (e.g., the North and Northeast regions) emit less emissions (Figure 1) but bear higher costs both in terms of current fuel price (Map 1) and potential future climate damage. Emission reduction measures, such as a carbon tax, are progressive and could help address this climate inequality by increasing the cost in regions that enjoy higher income and emit more. Meanwhile, development of the biofuel market expands the demand for agricultural products such as sugarcane and soybean, and in turn the revenue of farmers, thus helping bridge the income gap across sectors. Indeed, the ratio of agricultural productivity to whole economy productivity has been steadily improving since the 1980s, which coincided with the introduction of ethanol, and especially since the 2010s after the introduction of biodiesel. As carbon tax plays an important role in incentivizing clean energy transition and addressing climate inequality, this paper offers empirical evidence to answer the research question of how effective carbon tax, proxied by incremental change in fuel prices, is in achieving reduction of emission intensity in Brazil. We utilize variations in fuel prices and emission intensities of the transportation sector at the federative unit (FU) level and in annual frequency to conduct panel analysis while controlling for entity- and time- specific effects in achieving our estimation goals. The results offer support that a marginal increase in fuel prices, especially the diesel price, could help reduce emission intensity in both the short and long-run. In addition, emission intensity is persistent, and both entity and time effects are significant. Notably, the year 2020 demonstrated a consistent fixed effect in reducing emission intensity across estimated models. The rest of the paper is organized as follows. Section II presents stylized facts on Brazil’s fuel markets including the products, actors, and external factors. Section III reviews related literature. Section IV describes the data and metadata. Section V explains the empirical strategy in estimating the short-run and long-run emission intensity elasticities of fuel prices. Section VI discusses the results and policy implications. The last section concludes. 2 Figure 1. Regions with Higher Income Emit More GHG a. Distribution of GHG emissions (eCO2 t, in ln) b. Distribution of Income (million R$, in ln) 20 16 19 15 18 14 17 13 16 12 15 11 14 10 13 9 12 8 11 7 10 6 North Northeast Central-west Southeast South North Northeast Central-west Southeast South Data Source: SEEG. Data Source: System of Regional Account, IBGE. Note: Values of data points for both indicators are aggregates at FU level from 2010 to 2020. Map 1. Regions with Lower Income Pay Higher Fuel Prices Note: Values are average retail fuel prices of diesel B, gasoline C, and ethanol in 2022. Data source: Brazilian National Agency of Petroleum, Natural Gas and Biofuels (ANP). 3 II. Stylized Facts of Brazil’s Fuel Market for Transportation This section presents key characteristics of Brazil’s complex automotive fuel markets pertaining to products, actors, technology, and policy that might have an impact on emission intensity, the target of our analysis, thus informing our estimation strategy. II.1. Products: Volumes and Prices Brazil has a diversified fuel market for transportation where multiple fuels co-exist including notable shares of renewable biofuels. Four major types of retail fuels (i.e., diesel oil B, gasoline C, hydrous ethanol, and compressed natural gas or CNG) are on the market, some of which are mixtures of different source fuels (Table 1). While diesel oil and gasoline have been the most consumed fuels, ethanol produced from sugarcane entered the market as early as the 1970s after the first oil crisis and the launch of the Pró- Álcool program.1 Ethanol was initially used as anhydrous ethanol mixed into gasoline, and later also as hydrous ethanol used independently in neat ethanol-powered vehicles or flex-fuel vehicles. The combined volume of anhydrous and hydrous ethanol consumed (in toe) has now grown to around two-thirds that of gasoline A, the major component of retail gasoline C. Starting from the mid-2000s, biodiesel entered the market and has been mixed with diesel in increasing proportion to form diesel oil B, and a new variant of diesel oil B S10 with much less sulfur content was introduced after 2012. In addition, CNG has existed since the 1990s, though still in relatively small amounts and not available for sale in all FUs. Table 1. Multiple Types of Retail Automotive Fuels Co-exist in Brazil Type of Fuel Characteristics Applications Diesel oil B Mix of diesel oil A and biodiesel Heavy and light vehicles produced before (around 10-13% in recent years); 2012 Sulfur content of 500mg/kg Diesel oil B S10 Mix of diesel oil A and biodiesel; Heavy and light vehicles produced after Sulfur content of 10mg/kg 2012 with post-treatment technologies to reduce emissions Gasoline C Mix of gasoline A and anhydrous Light vehicles and motorcycles including ethanol (around 25-27% in recent flex-fuel vehicles years); Sulfur content of 50mg/kg Hydrous Ethanol Around 95% ethanol and the rest Flex-fuel2 and neat ethanol vehicles water CNG A blend of light compounds Vehicles converted to use CNG (that can run consisting of carbon and hydrogen on both CNG and gasoline) Sources: Petrobras, ANP, and CETESB. 1 The first use of ethanol fuel in Brazil dates back decades earlier, though mostly experimental and the usage was sporadic after World War II due to cheap oil prices. 2 88% of the new licensed vehicles had this technology in 2016 according to the Brazilian National Agency for Petroleum, Natural Gas and Biofuels (ANP). 4 Sales volumes of different fuels present various patterns of growth (Figure 2), reflecting influences of economic conditions, domestic policies, international markets, and technological invention/adoption. Diesel oil consumption has been increasing more steadily across time, with temporary dips during recessions. Gasoline consumption has been growing with medium-term fluctuations, partially reflecting world commodity cycles/oil crisis. Hydrous ethanol grew rapidly in the 1980s, lost momentum in the 1990s, before trending up again amid short-term fluctuations, likely resulting from policy incentives and the flex-fuel technology introduced in 2003 that enables ethanol to substitute gasoline. The growth of natural gas has been stagnant since 2007/2008 and was surpassed by biodiesel in 2013, the latter of which continues to grow thanks partially to policy support. Overall, the total volume of these fuels grew until the mid-2010s when it started fluctuating, closely mirroring the country’s real GDP. Figure 2. Varied Paths of Consumption across Time a. Accumulated/Stacked Values (103 toe) b. Time Series of Each Fuel (103 toe) 90000 40000 80000 35000 70000 30000 60000 25000 50000 20000 40000 15000 30000 10000 20000 5000 10000 0 1970 1974 1978 1982 1986 1990 1994 1998 2002 2006 2010 2014 2018 0 1970 1973 1976 1979 1982 1985 1988 1991 1994 1997 2000 2003 2006 2009 2012 2015 2018 2021 Diesel Oil Biodiesel Gasoline Anhydrous Ethanol Diesel Oil Biodiesel Gasoline Hydrous Ethanol Natural Gas Anhydrous Ethanol Hydrous Ethanol Natural Gas Data Source: Brazilian Energy Balance (BEN), EPE. Fuel prices, another factor that affects fuel consumption and its resultant emissions, present both spatial and temporal variations, which we will exploit to understand the effect of an incremental change in fuel price as a proxy for fuel tax. Final consumer prices in Brazil consist of producer billing price, federal and state taxes, and cost of distribution. These retail fuel prices vary by fuel type, year, and region (Figure 3). The price of ethanol is more dispersed, probably because it is supplied by numerous ethanol plants. The north and northeast regions are less developed but tend to have higher fuel prices than the richer southeast regions, partially due to logistics cost, resulting in a regressive fuel price system. Notably, FUs with the highest average fuel prices in 2022 are all in the north region, namely Acre, Roraima, Pará, and Rondônia, followed by four FUs in the northeast region, while São Paolo has the lowest average fuel price (Map 1). In addition, we also see rises in nominal fuel prices in recent years before dropping towards the end of 2022, as federal fuel tax rates were temporarily reduced to zero to mitigate the impact of inflation. 5 Figure 3. Dispersed Distribution of Fuel Prices (R$/l) a. Fuel Prices (R$/l) Distributions by Time b. Fuel Prices (R$/l) Distributions by Region Data Source: ANP. Note: Values are nominal prices. II.2. Actors: Producers and Consumers The country’s fossil fuel production is concentrated, while biofuel production is dispersed. Petrobras, the publicly held company whose controlling stakeholder is the federal government, has largely held a monopolistic position in the production of oil and gas until recent years when the market has gradually opened to new players. Still in 2020, Petrobras realized R$ 272 billion net revenue, over 10 times that of the second player Shell BrasilPetroleo. A few other producers take a smaller share of the market. After exploration and production, the crude oil is refined by the country’s 18 oil refineries (ANP) before going to distributors and retailers. The production of biofuels, on the other hand, starts from biomass farms and is produced by 383 ethanol plants and 49 biodiesel producers (ANP). In Brazil, biodiesel is mostly produced from soybeans (around 70%), and the rest from animal fat (e.g., beef fallow) and vegetable oil (e.g., cottonseed oil). Ethanol is produced mostly from sugarcane (around 85%) and the rest mostly from corn and sugar beets. While ethanol represents the largest share of total biofuel production (around 80%), biodiesel production has grown faster in recent years. Automotive fuels are directly consumed by households or businesses that own vehicles, nevertheless to a lesser degree they might affect the broader population through public/shared transportation. The country’s more than 115 million vehicles are shared among the total population of more than 213 million and total households of more than 75 million, or around 0.54 per capita and 1.5 per household respectively, according to data from IBGE for recent years. The country has a young population on average, and slightly over 37% of the country’s population is below the age of 18 (the minimum legal age for obtaining a driver license) and would not own a vehicle. The regional distribution of vehicles is also uneven, with almost half 6 of the vehicles registered in the southwest region and only six percent in the north region (Senatran 2021). In other words, an increase in the automotive fuel price through tax would have less adverse impact on the more vulnerable population that does not own a vehicle. II.3. External Factors: Technologies and Policies Beyond price, development of biofuels is also enabled by technological development and incentivized by government policies including through fiscal instruments. On the production side, sugarcane is one of the oldest crops in Brazil. It has been cultivated for sugar since the 16th century, and ethanol is produced as a byproduct during that process. Nowadays, distillers in Brazil could alter the relative production of sugar and ethanol, increasing the flexibility of ethanol production in response to the market. On the demand side, the introduction of automobiles in the early 20th century opened the opportunity for the usage of ethanol fuel, which was encouraged by a few local governments. It was not until the first oil crisis, and the launch of the first national program promoting ethanol (i.e., “Pró-Álcool”) in 1975, when Brazil saw the first wave of ethanol fuel production. Innovation and tax exemption on vehicles that could run on pure hydrous ethanol, and policy mandate on mixing anhydrous ethanol into gasoline, are all key factors in driving the growth of ethanol through increasing demand. The volume of ethanol fuels reached a similar level as that of gasoline towards the end of the 1980s, though it was outpaced by faster growth of gasoline thereafter. The flex-fuel technology introduced in 2003 that allows a vehicle to consume a blend of gasoline and ethanol at different proportions, tax differentials between fossil fuels and biofuels, and the mandate of mixing biodiesel into diesel through the National Program of Production and Use of Biodiesel (PNPB) have helped explain recent growth of biofuels. The availability of distribution infrastructure across Brazil is also an enabling asset for the biofuel market. The recently established Brazilian Biofuel Policy (RenovaBio), in particular, targets carbon intensity of the fuel matrix. As an integral part of the national energy policy, RenovaBio (Law No. 13576) was passed at the end of 2017 to promote biofuels (with emphasis on reliable supply) and reduce emissions from fuels for an initial period of 10 years. In achieving these goals, the policy relies on a combination of instruments. Mandatory annual targets for reducing emissions of the fuel matrix would be imposed on fuel distributors. The compliance of the targets would be verified by the number of Decarbonization Credits which are partially based on the amount of biofuel produced/imported/traded and could be traded in organized markets. For producers/importers of biofuels, certification of efficient production based on life-cycle assessment should be issued. Certain revenues also enjoy income tax benefits/exemptions. 7 II.4. Target: Emission Intensity Emission intensity of income, defined here as the aggregate emissions (t) from road transportation divided by aggregate real income, is the macro target for our analysis. This indicator measures how much emissions are generated per unit income increase, a smaller value of which is generally preferred. Emission intensity represents the technological factor T in the I=PAT identity that contributes to the environmental impact from economic activities. The emission intensity in Brazil has a wide range from 20 to 120 (t CO2e/million real income R$) across Brazil with more developed FUs such as the Federal District, Rio de Janeiro, and Sao Paulo having lower emission intensity (Figure 4). What explains the heterogeneity of emission intensity – to what extent it is due to fuel price or other FU/time-specific factors that confound the real relationship between fuel price and emission intensity – will be analyzed, and how effective a fuel tax could be in reducing emission intensity will be discussed in the following sections. Figure 4. Varied Emission Intensity (ton Emissions / million R$ Real Income) across FU and Time a. Distribution by FU b. Distribution by Year Source: IBGE, SEEG, and author’s calculations. 8 III. Literature Review Broadly speaking, this paper contributes to the discussion on getting the energy price right through a “cost-effectiveness” approach. While existing energy taxes are not necessarily based on carbon emissions, they could effectively reduce emissions and other negative externalities. This approach examines energy prices/taxes against the result of achieving a climate-related outcome (Parry, Heine, Lis, and Li 2014; Mooij, Keen, and Parry 2012). Theoretically, imposing a tax on an energy product would increase its price and reduce the quantity consumed, and in turn emissions. The quantity reduced depends on the elasticities of supply and demand, which vary across groups of population. The “cost-effectiveness” approach relies on empirical results in determining the effectiveness of raising the energy price through tax. A recent study by Köppl and Schratzenstaller (2022) reviewed empirical effects of carbon taxes in relation to a multiple impact matrix including environmental and macroeconomic impacts. A consensus is emerging that an appropriately designed carbon tax will dampen emissions (or at least their growth) without jeopardizing economic growth. Studies that examine actual/ex-post outcomes of carbon prices have a relatively short history with limited geographical coverage (mostly for Europe) but are attracting increasing attention. Another recent meta-review of ex-post analysis from Green (2021), on the other hand, revealed that actual reduction in emissions from carbon pricing is limited though it varies considerably across sectors, and a carbon tax performs better than Emissions Trading Schemes (ETS). Within this approach, studies on the impact of fuel price on emission intensity are sparse but emerging. Carbon tax increases fuel price and affects consumer behaviors through the price signal. In most countries where explicit carbon tax does not exist, the impact of marginal fuel price change due to fuel tax could be a proxy of the carbon tax effect. A few recent studies have used energy price change as a proxy to study the impact of carbon tax (Coste, Cali, Cantore, and Heine 2019; Mundaca, Strand and Young 2021). Most studies on the effectiveness of fuel prices or carbon taxes focus on their environmental or economic impact only or separately. Few studies are available that target a combined economic and environmental performance matrix, such as emissions intensity, i.e., the ratio of emissions to a non-emission measure, usually GDP. This is surprising as emissions intensities are measures of both theoretical and practical importance, and has been adopted as targets at the firm, sector, and country levels (Herzog, Pershing, and Baumert 2006) including as Nationally Determined Contributions (NDCs) target. In this stream, Ryan et al. (2009) examines the impact of fiscal measures on fleet emission intensity for EU15 countries and found that fuel and vehicle taxes, rather than voluntary agreements of automobile manufacturers, have an impact on emission intensity. Brännlund et al. (2014) analyzed carbon intensity of Swedish industry using firm-level data and found that it responds to both carbon tax and fuel price. A few more studies examine the relationship between energy price and energy intensity (Filipović et al. 2015; Grubb et al. 2018). To the best of our knowledge, this might be the first study on the impact of fuel prices on emission intensity of the transportation sector for Brazil utilizing insights from panel analysis. We benefit from the World Bank’s recent Economic Memorandum for Brazil (Hanusch Ed. 2023) in understanding the context of Brazil’s economic development and environmental challenge nexus. While most studies on Brazil’s fuel price elasticities of demand adopted time series analysis at national aggregate level, we are informed by a few studies that employed panel analysis at the state or municipality level (da Silva et al. 2009; Santos 2012; Uchôa, et al. 2019; Heine et al. 2021). Another recent paper studies the environmental impact of fuel prices in Brazil using time series analysis that examines aggregate CO2 at the national level (Filho et al. 2021). Our empirical strategy of panel analysis is also inspired by studies on the economic- environmental relationship for other regions/world including Baltagi and Griffin (1983) and Yao (2021). 9 IV. Data and Metadata We compiled a panel dataset at the Federative Unit (FU) level (including 26 states and 1 federal district) for all FUs and at annual frequency across the overlapping periods from 2010 to 2020 that are available from all data sources. All data are collected from credible national sources including Brazilian Institute of Geography and Statistics (IBGE), Brazilian National Agency of Petroleum, Natural Gas and Biofuels (ANP), Energy Research Office (EPE) of the Brazilian Ministry of Mines and Energy (MME), and Greenhouse Gas Emission and Removal Estimating System (SEEG). Fuel prices for our analysis are final consumer prices at retail stations of automotive fuels. A typical retail price of fuel in Brazil consists of cost to producers (mostly Petrobras) or importers, federal and state taxes (i.e., CIDE, PIS, COFINS and ICMS), cost of biofuel that is mixed into the fossil fuel, and cost to distributors and retailers. The retail prices are surveyed in ANP’s Fuel Price Survey (LPC) that cover prices of gasoline C, hydrous ethanol, diesel oil B, vehicular natural gas (CNG), as well as Diesel oil B S10 that was introduced at the end of 2012. The survey was conducted weekly around 459 locations across all Brazilian FUs (ANP, 2020). The LPC is the most comprehensive and authoritative survey of fuel prices in the country, and it has been used for developing public policies (ANP, 2020). We obtained data for our analysis from the historical series aggregated at the monthly frequency and FU level, and combined datasets from different time periods, resulting in continuous series from 2001 to 2022. Prices at FU level is weighted by sales of distributors after October 2004. We take a simple average of the monthly values to aggregate it into quarterly and annual series. We use GDP compiled from the income perspective for our analysis. Total income of an administrative region consists of employee remuneration (salary plus social contribution), operation surplus, mixed income, and taxes (net of subsidies). We obtained country, region, and FU level GDP/income data from the country’s System of Regional Accounts (SCR) 2020 that is produced by the Brazilian Institute of Geography and Statistics (IBGE) in collaboration with the State Statistical Offices, State Government Departments, and the Superintendence of the Manaus Free Zone. The SCR, revised in 2015, is compatible with international methodologies of the System of National Accounts 2008 (IBGE 2020). We collected three income series, i.e., total income, employee remuneration, and other income (operation surplus plus mixed income). The data is available at annual frequency from 2010 to 2020. Estimates of greenhouse gas (GHG) emissions are focused on those that arise from fuel combustion for road transportation covering all vehicles (e.g., cars, trucks, buses, and motorcycles) that respond to prices of automobile fuels. During combustion, carbon (C) stored in the fuel is oxidized and released as carbon dioxide (CO2) and, to the extent of incomplete combustion, methane (CH4), carbon monoxide (CO) and non-methane volatile organic compounds (NMVOCs). In addition, minor emissions from nitrogen (N2) oxidation result in nitrogen oxides (NOx) and/or nitrous oxide (N2O) depending on process temperature (de Azevedo, T., Costa Junior, C., Brandão Junior, A. et al., 2018). We obtained emission data from the latest version of System for Estimating Greenhouse Gas Emissions (SEEG version 10, 2022). An initiative of Climate Observatory, SEEG estimates are based on Brazilian Inventory (BI) of Anthropogenic Emissions and Removals of Greenhouse Gases from the Ministry of Science, Technology, and Innovation (MCTI) that are consistent with guidelines of Intergovernmental Panel on Climate Change (IPCC). SEEG reproduces and validates BI data, and improves its coverage and granularity (de Azevedo, T., Costa Junior, C., Brandão Junior, A. et al., 2018). For our analysis, we 10 aggregated emissions (across types of vehicles) from the road transportation subsector (while preserving the breakdowns in types of fuels and GHG emissions). The emissions are calculated in general as a multiplication of fuel consumption as available from EPE’s Brazilian Energy Balance (BEN) with adjustments in vehicle usage and emission factors as published by the MCTI. The emission factors are fuel specific and those for GHG gases other than CO2 take process/activity and technology into additional consideration. SEEG allocates national emissions to subnational levels including to all 27 FUs based on actual fuel consumption for the transportation sector. Our collection starts from 2006 when data is almost complete for all FUs and fuel/vehicle types till 2021. To our knowledge, this is the most comprehensive dataset on Brazil’s GHG emissions in terms of time periods and level of disaggregation. V. Empirical Strategy We focus on the relationship between fuel prices and emission intensity. Fuel prices are subject to policy intervention, such as through fuel/carbon taxes and state-owned fuel producers. To remove the effect of inflation, we will analyze deflated/real fuel prices. Emission Intensity, defined here as aggregate emissions in CO2 equivalent from road transportation divided by aggregate real income of an administrative region, could be a target of the government that represents a balanced view of economic development and climate mitigation. This emission intensity measure also serves to scale the gross emission values so that it is more comparable across our units of analysis and the heteroskedasticity of the residuals would be reduced. Based on literature and our assumptions, the emission intensity should respond to fuel prices, such as through reducing fuel consumption and substitution effects across fuels due to price differences. Given the complexity of Brazil’s fuel markets, the primary question this paper will answer is how effective fuel prices are in reducing emission intensity in Brazil. We will analyze both the effect of the general fuel price level and price differentials between fuels on emission intensity. Additionally, we will also understand the evolution and heterogeneity of emission intensity in Brazil and whether the pandemic in 2020 has an impact. As price data are timelier with higher frequency than emission and GDP data, our model could also be used to nowcast emission intensity. V.1. Summary Statistics We transform the source data into our analyzed variables to start with, and Table 2 summarizes panel statistics of our analyzed variables. We first deflate nominal income and fuel prices to obtain real income and fuel prices. The deflator was calculated from national level GDP at current and constant prices from World Development Indicators, with the deflator equals to 1 in 2010, the starting period of our analysis. We applied these year-specific deflators to the income and prices of each FU in the corresponding year. In other words, the prices and incomes used for our analysis are constant at the price level of 2010. Second, we average the fuel prices of gasoline C, diesel B, and hydrous ethanol to represent the general price level of automotive fuels that captures price information from all major fuels, as well as modelling individual fuel price. To derive the emission intensity (t CO2e per million R$), we divided the aggregate CO2 equivalent (t) emissions from road transportation by real income. The CO2 equivalent considers all GHG emissions from road transportation based on their global warming potential as equivalent to CO2, while 11 CO2 accounts for most of transportation sector's GHG emissions. All calculations are computed at FU level for each year. Table 2. Summary Panel Statistics of Analyzed Variables Variable Mean Std. Dev. Min Max Observations overall 2.10 0.14 1.78 2.61 N= 297 Average between 0.10 1.89 2.42 n= 27 Fuel Price within 0.10 1.87 2.38 T= 11 overall 1.96 0.15 1.71 2.59 N= 297 Diesel between 0.11 1.83 2.36 n= 27 Price within 0.10 1.73 2.26 T= 11 overall 2.42 0.17 1.89 2.99 N= 297 Gasoline between 0.09 2.27 2.71 n= 27 Price within 0.14 1.98 2.94 T= 11 overall 1.91 0.20 1.42 2.41 N= 297 Ethanol between 0.17 1.53 2.20 n= 27 Price within 0.10 1.64 2.17 T= 11 overall 56.56 20.34 16.24 132.59 N = 297 Emission between 20.19 18.72 122.16 n = 27 Intensity within 4.48 44.27 77.93 T = 11 Note: Values for prices are in the unit of R$/l (at constant price) and emission intensity in ton CO2e per million R$. V.2. Panel Regression We apply panel regression with entity and time effects, a more efficient analysis as compared with cross- sectional or time series analysis, in uncovering the relationship between fuel price and emission intensity. The panel analysis increases sample size by adding an additional dimension of variation and reduces omitted variable bias by controlling for individual heterogeneity, resulting in more reliable parameter estimates (Baltagi, 2021). Utilizing the panel dataset is especially valuable as the Brazilian economy is considerably heterogeneous as manifested through the variance of emission intensity and fuel price across FU/time. Additionally, the panel analysis offers insights on entity and time effects to provide a more comprehensive understanding of emission intensity across space and time. a. Functional Form We apply linear regressions with both double level and double log specifications. In designing the functional form of our regression, a first glimpse of the variables in a scatter plot and in time series (Figure 12 5) suggests that the relationship between emission intensity and price for each FU is largely linear and both series are largely stationary in our sample, so we would stay with the linear regression. The different specifications of variables would help us generate interpretations of coefficient at different scales: the double level specification would tell us the effect of a unit increase in the actual real price (R$/l) on the actual change in emission intensity (t CO2e/million R$), which would help generate more concrete policy implications including on carbon price for policy makers; while the double log specification would tell us the elasticity of price without unit which could be compared with other research on elasticities. b. Model Specification and Estimation We control both entity and time effects as either fixed or random effects. The entity fixed effects would control for FU-specific characteristics that do not vary across time, such as geographical characteristics, cultural preferences, and institutional quality of subnational governments. The time fixed effects would control for time-specific shocks that apply for the entire Brazil, such as commodity market fluctuations, federal policies, and the pandemic. As we do not know whether the individual effects are correlated with price or random, we will apply both fixed effects and random effects models unless the Hausman’s test (1978) renders the random effects model inappropriate. The fixed effect models would be estimated with the within regression estimator while the random effects the GLS estimator. To understand the effect of both the general price level and the combined effects of individual prices, we conduct both univariate (single price) and multivariate (multiple prices) regressions after controlling for the entity and time effects. In the univariate regression, we focus on the relationship between average fuel price and emission intensity. To have a more comprehensive understanding and as robustness check, we will substitute the average price with each individual fuel price to understand the relative importance of different fuel prices, as the price of an individual fuel not only affects its own emissions but also emissions of other fuels through substitution effect. In the multivariate regression, we include prices of all three major fuels as regressors. It should be noted that the fuel prices are correlated which lead to the issue of multicollinearity, so we turn the prices of gasoline and ethanol into relative prices (relative to diesel) to mitigate this issue and better understand the substitution effects among the fuels. To differentiate short-run and long-run effects, we will estimate both static and dynamic models. In the dynamic models, we include lagged value of the dependent variable (i.e., the Koyck lag) as a regressor. The static model will tell us the marginal effect of price on emission intensity, while the dynamic model will help us understand the dynamic short-run marginal effect and the long-run cumulated effect of price on emission intensity. The double-log specification of the dynamic model will help us derive the short-run and long-run elasticities of fuel price. In mathematical equations, we propose the following static and dynamic models: yit = α + βP′it + µi + λt + vit = 1, …, = 1, …, (1) yit = α + βP′it + γyit−1 + µi + λt + vit = 1, …, = 1, …, (2) with i denoting FU and t denoting year. Y is the dependent variable emission intensity, α is the scaler, Pit is the itth observation on explanatory variables which is average fuel price/diesel price/gasoline 13 price/ethanol price in the univariate model, and the reference fuel price and relative prices of other fuels in the multivariate model. The residual has three components: µ is the time-invariant entity effect, λ the entity-constant time effects, and v the remainder disturbance that varies with entity and time. Equation (1) is the static postulation, and Equation (2) the dynamic postulation. The double-log models would use the natural logarithm of both y and P, where β represents short-run elasticity and β/(1-γ) represents long- run elasticity. Figure 5. Negatively Correlated Emission Intensity and Real Fuel Price in Brazil Emission Intensity (lhs) Deflated Fuel Price (rhs) 48 2.8 47 2.6 46 2.4 45 2.2 44 2.0 43 1.8 42 1.6 41 40 1.4 39 1.2 38 1.0 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 Emission Intensity (CO2e ton/million R$) Price of Diesel oil B (R$/l) Price of Gasoline C (R$/l) 14 VI. Results and Discussion After testing multiple models, we focused on dynamic models with two-way fixed effects to reveal the relationship between fuel prices and emission intensities based on statistical results. The results suggest that diesel price has the most significant impact on emission intensity of the transportation sector in Brazil. Increasing diesel price through tax by R$1/l (excluding inflation) might help reduce emission intensity of road transportation by 20 (t CO2e/million R$) on average. Insignificant effect of ethanol prices could be explained by its substitution effect with fossil fuels which works in the opposite direction in reducing emission intensity. Emission intensity of the transportation sector in Brazil is elastic to fuel prices in the long run, implying levying carbon tax should be effective in reducing emission intensity as it also induces technological and/or behavioral changes that have long-term effect. Furthermore, both entity and time effects are significant, suggesting other factors beyond prices also matter for emission intensities and could be the subjects for research in the future. VI.1. Model Selection Initial statistical tests suggest that we should focus on dynamic models with two-way fixed effects. Entity- specific effects are significant and explain most of the variations in emission intensity. A simple pooled OLS without controlling for entity fixed effects performs poorly in explaining emission intensity, and r- square improves notably after controlling for entity fixed effects. Rho statistics in models with entity effects are in the range 0.8-0.9, suggesting significant entity effects as well. Joint test on the year dummies suggests that time effects are also significant. Comparing static and dynamic models with both entity and time effects suggests that dynamic models achieve higher adjusted r-square. Among dynamic models, Hausman’s test on the one way fixed and random effect models suggests that we should not use random models, so we will only report results from fixed effects models. Modified Wald test for groupwise heteroskedasticity in fixed effect regression model suggests that there’s presence of heteroskedasticity, thus we will report heteroskedasticity-robust standard errors. VI.2. Insights Diesel price has the most significant effect on emission intensity, while the average price level of all fuels is also in reverse relationship with emission intensity. In other words, increasing fuel prices correspond with lower emission intensity. Among single price (“univariate”) models (controlling for lagged value of emission intensity, entity and time fixed effects), coefficients of the price variable are always negative, but only significant at 95% confidence level when diesel price is the explanatory variable and at 90% confidence level when average fuel price is the explanatory variable. The coefficient of gasoline price also turns significant in the double log model. The diesel price model also generates the highest within r-square. Table 3 and Table 4 summarize the results of single price models in double level and double log variations. 15 Diesel price remains significant at 95% confidence level in the multiple price (“multivariate”) model, whether used as the reference fuel or as relative prices to other fuels (Table 5). Several factors might explain the responsiveness of emission intensity to different fuel prices. Besides being the most consumed fuel with relatively higher emission factor, diesel is used mostly in heavy vehicles for commercial purposes, implying that corporations could be more responsive to fuel prices than households. Ethanol price tends to be insignificant, as the substitution effect works in the opposite direction to own price effect in reducing emissions (i.e., increasing ethanol price reduces ethanol consumption but also increases gasoline consumption). When we control the relative prices of diesel and gasoline to ethanol (and in turn the substitution effect), the reverse impact of ethanol price on emission intensity also turns significant. Increasing diesel price through tax by R$1/l (excluding inflation) might help reduce emission intensity of road transportation by 20 (t CO2e/million R$) on average and at Brazil’s current income level that could translate into 80 million t CO2e reduction. In percentage terms, a 10% increase in diesel price could reduce emission intensity by 7.4% in the current year with accumulated long-run effect amounting to 20%. It should be noted that the relationship estimated apply within the price range of the sample, and we do not know whether at extremely high/low prices the linear relationship still holds. Figure 6 suggests that predictive margins are wider at lower/higher ends of the price range. Nevertheless, a marginal increase from the current price level, or around the real price level of 2R$/l, should be effective as estimated. It should also be noted that increases in diesel prices in the past have led to the trucker’s strike between May 21 and 31, 2018, in Brazil, and that event was estimated to have resulted in improved air quality and reduced deaths (Leirião et al. 2020). Emission intensity of the transportation sector in Brazil is elastic to fuel prices in the long run (within a decade). While a 1% increase in fuel price tends to induce less than 1% reduction in emission intensity in the short run, its accumulated effect in the long-run increases for all prices and to above 1% for diesel and average price. The persistent effect from price could result from behavior change or the adoption of more efficient vehicles, the effect of which will be carried over into the future. Short-run and long-run elasticities of diesel price are presented in Table 6. Beyond price and lagged effects, both time and entity effects are notable. Emission intensity increases significantly in 2012 and decreases significantly in 2020, when the pandemic happened. The time effects from these two years are robust across all single price models at 95% confidence level. In the diesel single price model, more time effects turn significant including 2012, 2013, 2014, 2016, 2019, 2020. Except for 2020, emission intensity has been increasing (when controlling for previous year’s emission intensity) in previous years unfortunately. The more developed FUs such as the federal district, Rio de Janeiro, and Sao Paolo tend to have lower emission intensity due to their entity-specific effects, rather than resulting from the relative lower fuel prices in these regions. Tocantins is almost an outlier in having much higher emission intensity, followed by Rondônia. 16 Table 3. Dynamic Single Price Models in Level with Fixed Entity and Time Effects Single Price Models Emission Intensity Average Gasoline Diesel Ethanol (Dep. Var) Fuel Price Price Price Price Price -13.93* -5.41 -20.34** -3.05 Emission Intensity (t-1) 0.64*** .65*** 0.58*** 0.65*** _cons 50.53*** .45*** 62.06*** 25.66** legend: * p<0.1; ** p<0.05; *** p<0.01 N 270 270 270 270 F 48.33 58.62 46.99 54.22 p 0.00 0.00 0.00 0.00 r2_within 0.60 0.59 0.62 0.59 r2_adjusted 0.58 0.57 0.60 0.57 r2_between 0.99 1.00 0.96 1.00 r2_overall 0.96 0.97 0.94 0.97 Year 2012 2.23 3.12 2.25 3.73 2013 -0.44 0.37 1.32 1.11 2014 -0.25 0.62 2.23 1.68 2015 -2.19 -1.66 1.38 -1.2 2016 0.33 0.07 2.61 0.94 2017 -0.67 -0.73 1.73 0.26 2018 -1.52 -2.29 1.56 -2.55 2019 0.33 -0.11 3.84 0.1 2020 -3.89 -3.12 -2.2 -2.24 FU AL -4.26 -1.41 -8.88 -0.66 AM -9.55 -6.79 -14.28 -5.99 AP 0.01 1.22 -1.63 3.16 BA -0.15 3.27 -4.42 3.7 CE -3.66 -1 -7.5 -0.23 DF -15.42 -12.15 -20.61 -11.44 ES -5.41 -2.67 -11 -1.55 GO 1.37 5.35 -1.03 5.05 MA 1.41 4.1 -2.48 5.45 MG -2.62 1.25 -5.79 1.2 MS 1 3.82 -0.16 4.44 MT 6.9 10.23 8.52 9.61 PA 2.28 3.76 -0.6 4.64 PB -3.28 0.31 -6.98 1.37 PE -5.14 -1.68 -9.2 -0.87 PI 3.65 6.16 1.05 7.3 PR -2.44 2.11 -6.16 2.49 RJ -12.15 -8.99 -18.88 -8.72 RN -2.52 0.24 -6.99 1.12 RO 10.72 12.04 9.26 12.62 RR -1.16 0.02 -4.3 1.54 RS -2.75 -0.54 -8.46 0.74 SC -2.15 0.64 -6.51 2.08 SE -2.49 0.37 -6.8 1.35 SP -13.83 -8.58 -17.72 -8.23 TO 23.32 25.43 21.87 25.59 Note: Significance based on robust error. 17 Table 4. Dynamic Single Price Models in Ln with Fixed Entity and Time Effects Single Price Models Emission Intensity in ln Average Gasoline Diesel Ethanol (Dep. Var) Fuel Price Price Price Price Price in ln -0.51* -0.27** -0.74** -0.04 Emission Intensity in ln (t-1) 0.70*** 0.71*** 0.64*** 0.71*** _cons 1.61*** 1.40*** 1.89*** 1.17*** legend: * p<0.1; ** p<0.05; *** p<0.01 N 270 270 270 270 F 50.23 58.45 56.37 57.59 p 0 0 0 0 r2_within 0.65 0.64 0.67 0.64 r2_adjusted 0.63 0.63 0.66 0.62 r2_between 0.99 1 0.97 1 r2_overall 0.97 0.98 0.95 0.98 Year 2012 0.03 0.05 0.03 0.06 2013 -0.02 -0.01 0.01 0.01 2014 -0.02 -0.01 0.03 0.02 2015 -0.05 -0.04 0.02 -0.02 2016 0 0 0.04 0.01 2017 -0.02 -0.03 0.02 0 2018 -0.04 -0.06 0.01 -0.06 2019 0 -0.01 0.06 0 2020 -0.08 -0.07 -0.05 -0.04 FU AL -0.07 -0.03 -0.16 -0.01 AM -0.18 -0.13 -0.27 -0.11 AP -0.01 0 -0.04 0.05 BA -0.01 0.04 -0.09 0.06 CE -0.06 -0.02 -0.13 0 DF -0.37 -0.31 -0.48 -0.28 ES -0.09 -0.05 -0.19 -0.02 GO 0.01 0.07 -0.04 0.08 MA 0.01 0.05 -0.06 0.08 MG -0.05 0.01 -0.11 0.03 MS 0 0.04 -0.02 0.07 MT 0.07 0.12 0.09 0.13 PA 0.03 0.05 -0.02 0.07 PB -0.06 -0.01 -0.13 0.03 PE -0.08 -0.03 -0.16 0 PI 0.03 0.07 -0.01 0.1 PR -0.05 0.02 -0.12 0.05 RJ -0.26 -0.2 -0.39 -0.19 RN -0.04 0 -0.12 0.02 RO 0.13 0.14 0.1 0.16 RR -0.01 0 -0.07 0.03 RS -0.04 -0.01 -0.14 0.01 SC -0.04 0 -0.12 0.04 SE -0.04 0 -0.12 0.03 SP -0.27 -0.18 -0.34 -0.15 TO 0.22 0.25 0.17 0.26 Note: Significance based on robust error. 18 Table 5. Estimations from Dynamic Multiple Price Model in Level with Fixed Entity and Time Effects Emission Intensity Coef. Robust t P>t (Dep. Var) Std. Err. Diesel Price -20.706 9.490 2.180 0.038 ** Model I Gasoline Relative Price -3.152 9.076 0.350 0.731 Ethanol Relative Price 2.181 7.769 0.280 0.781 Emission Intensity (t-1) 0.585 0.066 8.900 0.000 *** _cons 64.665 21.955 2.950 0.007 *** Gasoline Price -17.464 7.843 2.230 0.035 ** Model II Ethanol Relative Price 3.885 8.719 0.450 0.660 Diesel Relative Price -50.350 22.605 2.230 0.035 ** Emission Intensity (t-1) 0.574 0.072 7.940 0.000 *** _cons 102.976 43.899 2.350 0.027 *** Ethanol Price -20.469 8.590 2.380 0.025 ** Model III Gasoline Relative Price -0.849 9.210 0.090 0.927 Diesel Relative Price -37.141 17.475 2.130 0.043 ** Emission Intensity (t-1) 0.580 0.061 9.520 0.000 *** _cons 101.123 29.662 3.410 0.002 *** Notes: Models vary by reference fuel and relative prices are calculated against reference fuel in the respective model. Significance is based on robust error. Table 6. Short-run and Long-run Price Elasticities Average Gasoline Diesel Ethanol Fuel Price Price Price Price Short-run elasticity -0.51 -0.27 -0.74 -0.04 Long-run elasticity -1.70 -0.93 -2.06 -0.14 Figure 6. Predictive Margin within 95% Confidence Intervals of Diesel Single Price Model 70 Emission Intensity, t CO2e per million R$ 40 50 30 60 19 VII. Conclusion This paper investigated how effective an incremental change in fuel price, a proxy for fuel/carbon tax, is in reducing the emission intensity of the transportation sector in Brazil through panel analysis. We first studied Brazil’s fuel market with regards to its products, actors, and external factors to contextualize our analysis. The country has a complex market of automotive fuels with considerable heterogeneity across space and time. We then leveraged econometric methods for panel analysis to uncover the relationship between fuel price and emission intensity, after selecting the most appropriate models among multiple alternatives. Our results suggest that the price of diesel has the most significant and robust impact on emission intensity. The short-run and long-run elasticities of the price of diesel are -0.74 and -2.06, respectively, in Brazil. Beyond the price and lagged effects, both entity and time effects are significant. Notably, the year 2020, when the pandemic happened, had a consistent fixed effect in reducing emission intensity across the estimated models. 20 References ANP. 2020. The Fuel Price Survey from the Brazilian National Agency of Petroleum, Natural Gas and Biofuels (ANP). Accessed March 2023. https://www.gov.br/anp/pt-br/assuntos/precos-e-defesa- da-concorrencia/precos/precos-revenda-e-de-distribuicao-combustiveis Baltagi, Badi H., and James M. Griffin. 1983. "Gasoline demand in the OECD: An application of pooling and testing procedures." European Economic Review 22, no. 2 (1983): 117-137. Baltagi, B.H. (2021). Econometric Analysis of Panel Data. Syracuse, NY: Syracuse University. Brännlund, Runar, Tommy Lundgren & Per-Olov Marklund. 2014. “Carbon Intensity in Production and the Effects of Climate Policy—Evidence from Swedish Industry.” Energy Policy (Volume 67): 844- 857. CCDR. 2023. Country Climate and Development Report for Brazil: Options for Greener and Climate-smart Growth. Washington, D.C.: World Bank. Coste, Antoine, Massimiliano Cali, and Dirk Heine. 2018. "Staying Competitive: Productivity Effects of Environmental Taxes." da Silva, Giovani Ferreira, Gisele Ferreira Tiryaki, and Luiz Antônio Magalhães Pontes. 2009. "The impact of a growing ethanol market on the demand elasticity for gasoline in Brazil." In Annual International Association for Energy Economics Conference, 32., San Francisco. USAEE San Francisco, 2009. De Azevedo, Tasso Rezende, Ciniro Costa Junior, Amintas Brandão Junior, Marcelo dos Santos Cremer, Marina Piatto, David Shiling Tsai, Paulo Barreto et al. 2018. "SEEG initiative estimates of Brazilian greenhouse gas emissions from 1970 to 2015." Scientific data 5, no. 1: 1-43. Heine, Dirk, João Paulo Braga, James Cust, Ira Dorband, Rodrigo Garcia, Alexis Rivera Ballesteros, Christian Schoder, Stephen Stretton, Tong Zhang, and Karlygash Zhunussova. 2021. Fiscal Foundations for Decarbonization: Carbon Charges in Brazil. Unpublished manuscript. Filho, da Rocha Lima Roberto Ivo, Thereza Cristina Nogueira de Aquino, and Adriano Marçal Nogueira Neto. 2021. "Fuel price control in Brazil: environmental impacts." Environment, Development and Sustainability 23: 9811-9826. Filipović, Sanja, Miroslav Verbič, Mirjana Radovanović. (2015). “Determinants of Energy Intensity in the European Union: A Panel Data Analysis.” Energy (Volume 92, Part 3): 547-555. Green, Jessica (2021). “Does Carbon Pricing Reduce Emissions? A Review of Ex-post Analyses.” Environmental Research Letters 16, no. 4: 043004. Grubb, M. Bashmakov, I., Drummond, P., Myshak, A., Hughes, N. Biancardi, A., Agnolucci, P. and Lowe, R. 2018. An Exploration of Energy Cost, Ranges, Limits and Adjustment Process. UCL Institute for Sustainable Resources. Hanusch, Marek Ed. 2022. A Balancing Act for Brazil’s States of the Legal Amazon: An Economic Memorandum. Washington, DC: World Bank. 21 Hausman, Jerry A. 1978. “Specification Tests in Econometrics.” Econometrica 46: 1251–1271. Timothy, Herzog, P. Jonathan, and B. Kevin. 2006. "Target-intensity: an analysis of greenhouse gas intensity targets." Washington DC: World Resources Institute, Washington D.C., USA. IBGE. (2020). System of Regional Accounts from IBGE. Accessed in March 2023. https://www.ibge.gov.br/en/statistics/economic/national-accounts/16855-regional-accounts- of-brazil.html?=&t=o-que-e Köppl, Angela, and Margit Schratzenstaller. 2023. "Carbon taxation: A review of the empirical literature." Journal of Economic Surveys 37, no. 4: 1353-1388. Leirião, Luciana Ferreira Leite, Daniela Debone, Theotonio Pauliquevis, Nilton Manuel Évora do Rosário, and Simone Georges El Khouri Miraglia. 2020. "Environmental and public health effects of vehicle emissions in a large metropolis: Case study of a truck driver strike in Sao Paulo, Brazil." Atmospheric Pollution Research 11, no. 6: 24-31. Mooij, Ruud, Keen, Michael, and Parry, Ian. 2012. Fiscal Policy to Mitigate Climate Change. Washington, DC: International Monetary Fund. Mundaca, Gabriela, Jon Strand, and Ian R. Young. 2021. "Carbon pricing of international transport fuels: Impacts on carbon emissions and trade activity." Journal of environmental economics and management 110: 102517. Parry, Ian, Dirk Heine, Elisa Lis, and Shanjun Li. 2014. Getting Energy Prices Right: From Principal to Practice. Washington, DC: International Monetary Fund. Pigato, Miria A. (Ed.) 2019. Fiscal Policies for Development and Climate Action: International Development in Focus. Washington, DC: World Bank. Ryan, Lisa, Susana Ferreira, and Frank Convery. 2009. “The Impact of Fiscal and Other Measures on New Passenger Car Sales and CO2 Emissions Intensity: Evidence from Europe.” Energy Economics (Volume 31, Issue 3): 365-374. Santos, Gervásio F. 2013. "Fuel demand in Brazil in a dynamic panel data approach." Energy Economics 36: 229-240. Uchôa, Frederico, Cleiton Silva de Jesus, and Leonardo Chaves Borges Cardoso. 2020. "Fuel demand elasticities in Brazil: a panel data analysis with instrumental variables." International Journal of Energy Economics and Policy 10, no. 2: 450-457. World Bank. 2021. Renewing with Growth; Volver a crecer. LAC Semiannual Report. Washington, DC: World Bank. Yao, Jiaxiong. 2021. Electricity Consumption and Temperature: Evidence from Satellite Data. IMF Working Papers. 22