### Transcription

Brazil’s Oil Production and Trade with the United StatesDavid Riker 1U.S. International Trade Commission, Office of EconomicsOctober 18, 2012ABSTRACTSignificant new discoveries of offshore oil could reshape the Brazilian economy, including its pattern oftrade with the United States and the rest of the world. In this paper, we estimate a set of econometricmodels that relate each country’s international trade in crude and refined petroleum, upstream inputs intopetroleum production, transportation equipment that uses petroleum products, and other bilateral importand exports to the country’s volume of domestic crude oil production. Then we use the models to projectthe likely impact of the anticipated boom in Brazilian crude oil production on Brazil’s trade with theUnited States.1Email: [email protected]. This paper is part of the Brazil Research Initiative in the Country and RegionalAnalysis Division of the Office of Economics. The author thanks Arona Butcher, Justino De La Cruz, CynthiaForeso, and participants in an Office of Economics seminar for their helpful comments at earlier stages of thisresearch.1

Brazil’s Oil Production and Trade with the United StatesDavid RikerI.IntroductionSignificant new discoveries of offshore oil could reshape the Brazilian economy, including itspattern of trade with the United States and the rest of the world. The Tupi field was discovered in 2007by a consortium of Brazilian and foreign companies (Petrobras, BG Group, and Petrogal). The newdiscoveries lie off the southern coast of Brazil, south of Rio de Janeiro and southeast of Sao Paulo. Theseoil and natural gas deposits lie deep below the salt layer of the ocean floor. The resulting high pressuresmake the resources difficult to recover, but production is still expected to be profitable. The estimatedrecoverable resources are vast, and their full development will require significant improvements inBrazil’s energy transportation infrastructure.2The economic importance of these oil discoveries has been widely recognized in projections offuture global energy markets. For example, the U.S. Department of Energy’s International EnergyOutlook 2011 projects that production in Brazil will increase by more than 50 percent between 2011 and2020. Brazil is projected to move into the top tier of the world’s non-OPEC oil producers.3The goals of this paper are to estimate the statistical relationship between each country’s volumeof crude oil production and its pattern of international trade in several categories of products, and then touse the statistical models to translate the projected increase in Brazilian oil production into a projectedchange in Brazil’s trade with the United States. In the rest of this Introduction, we provide an economicframework that identifies many of the ways that domestic oil production can affect international trade.These economic effects underlie the statistical relationships that we estimate.2U.S. Department of Energy (2012) discusses the relevant constraints on expanding crude oil production in Brazil.3However, even under these optimistic projections, Brazilian crude oil production is expected to remain less thanfour percent of global crude oil production.2

We expect that the increase in the volume of domestic crude oil production will affect U.S.imports from Brazil in three different ways. The increase is likely to have the most direct effect on U.S.imports of crude oil and refinery products from Brazil, but it may also have less direct effects on U.S.imports of other products by lowering the fuel costs of transporting imports and exports.4 The magnitudeof these secondary effects will likely depend on the extent of regulation of energy prices in Brazil, theinternational arbitrage of fuel prices, possibilities for substitution among fuels, and the distribution ofpetroleum revenues. For example, the regulation of energy prices in Brazil limits the relative costadvantages of a more abundant domestic supply of liquid fuels. On the other hand, the boom in Brazil’senergy sector could limit the expansion of other sectors of the Brazilian economy and thus limit theexpansion of Brazil’s exports to the United States. This economic phenomenon is often referred to asDutch disease, based on the seminal work on booming sectors and international trade in Corden andNeary (1982) and Corden (1984). More recent theoretical and empirical analyses of Dutch diseaseinclude Acosta, Lartey, and Mandelman (2009), Beverelli, Dell Erba, and Rocha (2011), and Rajan andSubramanian (2011).The increase in the volume of domestic crude oil production could also have significant effects ontrade in the opposite direction, from the United States to Brazil. It is likely to have the most direct effecton U.S. exports of drilling equipment from the United States, since the demand for these products inBrazil is derived from oil production.5 It is also likely to have a positive effect on U.S. exports oftransportation equipment to Brazil, since the demand for transportation equipment generally increases asliquid fuel costs decline.6 The projected increase in crude oil production may also have less direct effectson U.S. exports of other products that are not directly tied to petroleum. For example, we expect that the4The reduction in fuel costs may also reduce manufacturing costs in energy-intensive industries in Brazil.5The analysis of Dutch disease in Beverelli, Dell Erba, and Rocha (2011) specifically focuses on these types ofinput-output linkages between the natural resource sector and the manufacturing sector.6Transportation equipment includes railway vehicles, motor vehicles, aircraft, and ships.3

expansion in Brazil’s oil revenues will increase incomes in Brazil, and this will in turn increase Brazil’soverall expenditures on imports.7While economic theory and past studies of Dutch disease help us to identify the likely economiceffects, their magnitudes are fundamentally an empirical issue. Therefore, we estimate a set of statisticalmodels that relate a country’s international trade in crude oil and refinery products, upstream inputs intopetroleum production, transportation equipment, and other bilateral import and exports to the country’svolume of domestic oil production. 8The analysis in this paper is organized into two main sections. Section II presents a set ofeconometric models of the relationship between a country’s annual crude oil production and itsinternational trade in several categories of products. Section III combines these econometric models withprojections of Brazil’s future production of crude oil in order to project the impact of the anticipatedboom in Brazilian oil production on the country’s international trade with the United States. Section IVprovides concluding remarks.II.Econometric ModelsWe estimate a set of econometric models that quantify the effect of an increase in a country’s oilproduction on its international trade. We examine the impact on five distinct categories of trade: U.S.petroleum imports, U.S. imports of all other products, U.S. exports of drilling equipment, U.S. exports of7This is called the spending effect in the economics literature on Dutch disease, e.g., Corden (1984).8Sardorsky (2012) also presents a set of econometric models of the relationship between the energy sector andinternational trade of several South American countries. However, he focuses on the effect of total energyconsumption on international trade, rather than the effect of oil production.4

transportation equipment, and U.S. exports of all other products.9 We model the trade flows using thefollowing log-linear function:ܸ ௧ ൌ ߚ ሺܱ ܮܫ ௧ ሻఉభ ሺ ܲܦܩ ௧ ሻఉమ ሺ ܵܫܦ ሻఉయ ܼ௧ ߝ ௧The dependent variable of the models, ܸ ௧ , is the value of U.S. exports to country ܿ (or U.S. imports fromcountry ܿ ) in products from category ݅ in year ݐ . The variable ܱ ܮܫ ௧ represents crude oil production incountry ܿ in year ݐ . The econometric models include the annual gross domestic product of country ܿ( ܲܦܩ ௧ ), the international distance between the United States and country ܿ ( ܵܫܦ ሻ, and year effects thatare common across the countries (ܼ௧ ). In the models of U.S. imports, the year effects control for variationin the size of the U.S. market over time. In the model of U.S. exports, the year effects control forvariation in U.S. production costs over time. The error term ߝ ௧ includes any unobservable determinantsof trade and any error in the measurement of trade values.We estimate the parameters of the trade models using a data set that combines oil production dataand trade data for many countries over several years. (In Section III, we will use these econometricmodels to quantify the impact of the projected increase in oil production in a single country, Brazil, onthat country’s imports and exports.) The estimation data set is a panel that covers 95 countries over thefifteen year period from 1996 to 2010. The annual data on crude oil production are from the InternationalEnergy Statistics database of the U.S. Department of Energy’s Energy Information Administration(EIA).10 The annual data on U.S. imports and exports by product and by country are from the U.S.International Trade Commission’s Trade Dataweb.119For the sake of the analysis in this paper, petroleum products refers to HTSUS codes 2709001000 through2710999000; drilling equipment refers to HTSUS codes 7304110020 through 7304296175, 7305111030 through7305208000, 7306110010 through 7306298150, and 8430494000 through 8430498020; and transportationequipment refers to HTSUS codes 8601100000 through 8908000000.10These data on the production of conventional liquids are publicly available cfm.11These data are publicly available at http://dataweb.usitc.gov.5

We consider three different estimation techniques that rely on alternative assumptions about theerror term ߝ ௧ . We examine all three alternatives in order to evaluate the sensitivity of our results to theseassumptions. The first and simplest estimator is log-linear Ordinary Least Squares (OLS). This modelassumes that ߝ ௧ has a log normal distribution. Since the models are log-linear, the estimated coefficientsof the OLS models can be interpreted as elasticities. The greatest drawback of the log-linear OLSestimator is that observations with zero trade values are dropped from the estimation dataset, because thenatural log of zero is undefined. In addition, Santos Silva and Tenreyro (2006) demonstrate that OLSestimates of the coefficients of log-linearized models will be biased if there is heteroskedasticity in theerror term, as is often the case in cross-sectional estimation involving many countries. For these reasons,Santos Silva and Tenreyro recommend the use of Poisson models to estimate log-linearized economicmodels. In the Poisson model, the dependent variable is the value of the trade flow in levels rather thanlogs. The estimated coefficients can again be interpreted as elasticities. The Poisson model includes thesame set of explanatory variables but different assumptions about the distribution of the error term. Animportant limitation of the Poisson model is that it imposes the unrealistic assumption that the variance ofthe error term is equal to its mean. The third estimator that we consider, the negative binomial model, is ageneralization of the Poisson model that relaxes this restriction on the distribution of the error term. Weuse a likelihood ratio test to determine whether the Poisson model or the negative binomial model is moreappropriate in each case. As long as the dispersion parameter of the negative binomial model issignificantly different from zero, the negative binomial model is more appropriate. In all of the modelsthat we estimate, we assume that the volume of domestic oil production is an exogenous variable,determined for the most part by natural resource discoveries.We undertook the econometric analysis in two steps. First, we estimated unrestricted models thatincluded all of the explanatory variables. Based on this first set of results, we re-estimated the modelswith exclusion restrictions on one or more of the coefficients of the models. In most cases, the preferred6

specification involved an exclusion restriction. Tables 1, 2, 3, 4, and 5 present the estimated coefficientsof the preferred models.Table 1 reports the estimated coefficients of the models of U.S. imports of petroleum products,based on the log-linear OLS, Poisson, and negative binomial estimators. The dependent variable is thecustoms value of U.S. imports from each country. The volume of domestic oil production has asignificant positive effect on the value of these U.S. imports, with an elasticity that ranges from 0.7795(for the negative binomial model) to 1.0230 (for the OLS model). An elasticity less than one indicatesthat the increase in the value of the imports is less than proportional to the increase in the volume ofdomestic oil production. This is the case for the negative binomial estimate: for every 10% increase in oilproduction, there is a 7.8% increase in the value of the U.S. imports from the country. In the case of theOLS estimate, the elasticity is not significantly different from one. International distance has a significantnegative effect on the U.S. imports. The GDP of the country of origin is excluded from the models inTable 1, because it is not statistically significant in the unrestricted models. The likelihood ratio testrejects the Poisson model in favor of the negative binomial model, based on the test statistic reported inthe last row and last column of the table. The log-linear OLS model is the least reliable of the threemodels, because it excludes the 192 observations with imports equal to zero.Table 2 reports the estimated coefficients of the preferred models of U.S. imports of all otherproducts, for each of the three estimators. The dependent variable is the customs value of U.S. importsfrom each country. The volume of domestic oil production has a significant positive effect on theseimports, with an elasticity that ranges from 0.1957 (for the OLS model) to 0.3189 (for the negativebinomial model). The GDP of the country of origin has a significant positive effect in these models ofU.S. imports, reflecting the size and productive capacity of the country of origin. This is consistent withconventional gravity models of international trade. International distance again has a significant negativeeffect on the value of imports. The model in Table 2 does not include year effects, since the yearindicator variables are not jointly significant in the unrestricted models. The negative binomial estimator7

is again the best of the three alternatives. Focusing on the negative binomial model, there is a significantpositive effect of the volume of domestic oil production on the imports but the elasticity is much smallerthan the elasticity for U.S. imports of petroleum products. We interpret the small effect on imports ofthese other products as the less direct effect of domestic energy production on the export competitivenessof the products, through the Dutch disease effect that we discussed in the Introduction.Table 3 reports the estimated coefficients of the models of U.S. exports of drilling equipment.These exports are more responsive to the volume of domestic crude oil production in the destinationcountry than the aggregate expenditures in the destination country (measured by its GDP). The elasticityestimate for the former is more than twice as large as the elasticity estimate for the latter in the negativebinomial model in Table 3. The elasticity estimates with respect to the volume of domestic oil productionrange from 0.4263 to 0.6124, depending on the estimator. The elasticity estimates with respect to GDPrange from 0.1647 to 0.2977. The year indicator variables and the GDP terms are both statisticallysignificant in the unrestricted models. This set of econometric estimates is consistent with the predictionsof economic theory that we discussed in the Introduction: since drilling equipment is a direct input intocrude oil production, we expect that U.S. exports of drilling equipment will be very responsive to thelevel of domestic crude oil production.Table 4 reports the estimated coefficients for the preferred models of U.S. exports oftransportation equipment. The elasticity of exports with respect to the volume of domestic oil productionis statistically significant for all three estimators, but it is relatively small. The elasticity estimates rangefrom 0.1178 to 0.1646. The models in Table 4 do not include year effects, since the year indicatorvariables are not jointly significant in the unrestricted models. We expect that these exports will beresponsive to the increased abundance and reduced cost of fuel in the oil-producing country, but the effectis less direct that the effect on drilling equipment exports. Comparing the elasticity estimates in Table 4to their counterparts in Table 3 (the model of U.S. exports of drilling equipment), the exports of8

transportation equipment are more responsive to the GDP of the destination country and less responsiveto the volume of domestic crude oil production.Finally, Table 5 is the preferred model for U.S. exports of all products other than drillingequipment and transportation equipment. These exports are less responsive to the volume of domestic oilproduction in the destination country. We expect that an increase in oil revenues will have a positiveeffect on a country’s import demand by increasing incomes. There is some support for this prediction inthe data, but the effect is small. This is not surprising, since it is an indirect effect and since the modelalready captures most or all of the income effect through the GDP term.12III.Projections for Oil Production and TradeNext, we utilize the projections of Brazil’s oil production from the 2011 edition of the EIA’sInternational Energy Outlook (IEO). The IEO provides projections for the production of conventionalliquids by country by year through 2035.13 The projections for the early years are based on the futuresupply schedules for specific projects already in development. The EIA projections of production beyondthe early years are modeled outcomes that correspond to a specific set of assumptions aboutmacroeconomic conditions, world liquids production, and energy efficiency. We focus on the IEO’sreference case. This case assumes that OPEC countries invest in production capacity to maintain theirforty percent share of worldwide liquids production. It assumes that OECD countries will be slow torecover from the recent global financial crisis, while non-OECD countries will experience much highergrowth rates. Finally, the case assumes that the transportation section continues to rely heavily onconventional liquid fuels.12Since this model already controls for national income through the GDP term, the coefficient on oil productionrepresents the incremental effect of the composition of income, between oil revenues and other sources.13The IEO defines conventional liquids as crude oil and lease condensate, natural gas plant liquids, and refinerygain. (http://www.eia.gov/oiaf/aeo/tablebrowser/#release IEO2011&subject 0-IEO2011&table 39IEO2011®ion 0-0&cases Reference-0504a 1630)9

The IEO compares its projections to an alternative set that are published in the InternationalEnergy Agency’s World Energy Outlook. There are significant differences between the two sets ofenergy market projections when it comes to nuclear and renewable energy, but both sets of projectionshave similar predictions for the production of liquid fuels, the economic variable addressed in our models.Table 6 reports the projected annual production of conventional liquids for Brazil and the entireworld through 2020. Brazil’s production of conventional liquids is projected to increase from 2.3 millionbarrels per day in 2011 to 3.5 million barrels in 2020, a 52 percent increase. The growth of production inBrazil is projected to exceed the growth of production in the rest of the world, resulting in an increase inBrazil’s share from 2.78 percent in 2011 to 3.90 percent in 2020.Figures 1, 2, 3, 4, and 5 report our estimates of the changes in international trade flows that willresult from the projected increase in Brazilian crude oil production. Like the IEO energy projections,these trade effects are not complete forecasts of the value of trade flows: they do not incorporatepredictions about all of the economic factors that affect the market equilibrium. They are projections thatisolate the incremental effect of the projected increase in volume of domestic crude oil production. Theytranslate the IEO projections into their impact on international trade between Brazil and the United Statesfor the different categories of traded products. The underlying calculations combine the projections inTable 6 with the estimated coefficients on crude oil production in Tables 1, 2, 3, 4 and 5.The five figures report the projected cumulative growth rate in trade at each year for the negativebinomial and OLS estimators, relative to the value of trade in 2011. According to the negative binomialmodel in Figure 1, the projected increase in oil production in Brazil will result in a 20.3 percent increasein U.S. imports of petroleum products from Brazil (over 2011 trade) by 2015, and a 40.7 percent increase(over the same base year) by 2020. The log-linear OLS model projects a larger effect in each year, with a40.7 percent increase over the base year by 2020. It is important to keep in mind that U.S. imports ofpetroleum products from Brazil accounted for less than two percent of total U.S. imports of petroleum10

products in 2011, so the projected percentage increase in total U.S. imports is much smaller. Accordingto the negative binomial model in Figure 2, the projected increase in oil production in Brazil will have amuch smaller impact, an 8.3 percent increase in U.S. imports of all other products from Brazil over thebase year by 2015, and a 16.6 percent increase by 2020. The effect based on the log-linear OLS model iseven smaller than the negative binomial model estimates, a 10.2 percent increase over the base year by2020.The models in Figure 3 indicate that the impact on the value of U.S. exports of drilling equipmentto Brazil will be relatively large: a 14.3 percent increase over 2011 trade by 2015, and a 28.5 percentincrease over the base year by 2020. The log-linear OLS model projects an even larger effect in eachyear, with a 32.0 percent increase over the base year by 2020. The projected effects on U.S. exports oftransportation equipment to Brazil are more moderate. The negative binomial model in Figure 4 indicatesthat the change in the value of all other U.S. exports to Brazil is a 4.3 percent increase over 2011 trade by2015, and an 8.6 percent increase over the base year by 2020. Finally, the projected effects on anaggregate of all other U.S. exports to Brazil are small. The negative binomial model in Figure 5 indicatesthat the change in the value of all other U.S. exports to Brazil is a 3.7 percent increase over 2011 trade by2015, and a 7.4 percent increase over the base year by 2020.IV.ConclusionsThe EIA and other industry experts project a significant increase in Brazil’s crude oil productionover the next decade as the recent offshore discoveries are developed. The econometric analysis of ourpanel of oil-producing countries indicates that each country’s domestic oil production generally has arelatively large effect on the country’s pattern of trade, especially in products that use liquid fuels orproducts that are used in oil production. There are smaller indirect effects on trade in other categories ofproducts. The exact magnitudes depend on the modeling assumptions that we adopt. Overall, the models11

suggest that the increase in oil production in Brazil will likely result in an expansion of trade betweenBrazil and the United States.ReferencesAcosta, Pablo A., Emmanuel K.K. Lartey, and Frederico S. Mandelman (2009): “Remittances and theDutch Disease.” Journal of International Economics 79: 102-116.Beverelli, Cosimo, Salvatore Dell Erbe, and Nadia Roche (2011): “Dutch Disease Revisited. OilDiscoveries and Movements of the Real Exchange Rate When Manufacturing is Resource-Intensive.”International Economics and Economic Policy 8: 139-153.Corden, W. Max (1984): “Booming Sector and Dutch Disease Economics: Survey and Consolidation.”Oxford Economic Papers 36: 359-380.Corden, W. Max and J. Peter Neary (1982): “Booming Sector and De-Industrialisation in a Small OpenEconomy.” Economic Journal 92: 825-848.Rajan, Raghuram G. and Arvind Subramian (2011): “Aid, Dutch Disease, and Manufacturing Growth.”Journal of Development Economics 94: 106-118.Sadorsky, Perry (2012): “Energy Consumption, Output and Trade in South America.” EnergyEconomics 34: 476-488.Santos Silva, J.M.C. and Silvana Tenreyro (2006): “The Log of Gravity.” Review of Economics andStatistics 88: 641-658.U.S. Department of Energy, Energy Information Administration (2011): International Energy Outlook2011.U.S. Department of Energy, Energy Information Administration (2012): Country Analysis Briefs: Brazil.12

Table 1: Econometric Models of U.S. Imports of Petroleum ProductsDependent variable: Customs value of U.S. imports of petroleum productsExplanatory VariablesLog-Linear OLSModelPoissonModelNegative BinomialModelProduction of Crude P of the Trade PartnerExcludedExcludedExcludedInternational 654(1.1300)*Year Effects, 1996-2009Included߯ ଶ 6.17*Included߯ ଶ 254.59*Included߯ ଶ 166.69*Number of Observations7239159150.52600.7537ܴଶ or Pseudo-ܴଶEstimate of ߙ Parameter8.2653(0.5117)Note: The GDP of the Trade Partner is excluded from this set of preferred models because they are notstatistically significant in most of the unrestricted model. An asterisk indicates statistical significance atthe 1% level.13

Table 2: Econometric Models of U.S. Imports of All Other ProductsDependent variable: Customs value of U.S. imports of all other productsExplanatory VariablesLog-Linear OLSModelPoissonModelNegative BinomialModelProduction of Crude P of the Trade )*International 5(0.9786)*Year Effects, 1996-2009ExcludedExcludedExcludedNumber of Observations8169159150.62150.8393ܴଶ or Pseudo-ܴଶEstimate of ߙ Parameter4.4424(0.3354)*Note: The year effects are excluded from this set of preferred models because they are not jointlysignificant in the unrestricted model. An asterisk indicates statistical significance at the 1% level.14

Table 3: Econometric Models of U.S. Drilling Equipment ExportsDependent variable: FAS value of U.S. drilling equipment exportsExplanatory VariablesLog-Linear OLSModelPoissonModelNegative BinomialModelProduction of Crude P of the Trade )*International 421(1.0556)*Year Effects, 1996-2009Included ܨ 6.98*Included߯ ଶ 225.35*Included߯ ଶ 111.94*Number of Observations7519159150.53670.7688ܴଶ or Pseudo-ܴଶEstimate of ߙ Parameter5.5740(0.3596)*Note: An asterisk indicates statistical significance at the 1% level.15

Table 4: Econometric Models of U.S. Exports of Transportation EquipmentDependent variable: FAS value of U.S. exports of transportation equipmentExplanatory VariablesLog-Linear OLSModelPoissonModelNegative BinomialModelProduction of Crude P of the Trade )*International (1.2764)*Year Effects, 1996-2009ExcludedExcludedExcludedNumber of Observations8039159150.62620.9042ܴଶ or Pseudo-ܴଶEstimate of ߙ Parameter4.1365(0.3062)*Note: An asterisk indicates statistical significance at the 1% level.16

Table 5: Econometric Models of U.S. Exports of All Other ProductsDependent variable: FAS value of U.S. exports of all other productsExplanatory VariablesLog-Linear OLSModelPoissonModelNegative BinomialModelProduction of Crude Oil0.0416(0.0288)0.1401(0.0201)*0.1413(0.0236)*GDP of the Trade )*International 1(1.1989)*Year Effects, 1996-2009ExcludedExcludedExcludedNumber of Observations8219159150.70240.8960ܴଶ or Pseudo-ܴଶEstimate of ߙ Parameter3.5558(0.2955)*Note: An asterisk indicates statistical significance at the 1% level.17

Table 6: International Energy Outlook 2011 Projection of Oil ProductionYearBrazilian Production(Millions of Barrels per Day)World Production(Millions of Barrels per Day)Brazilian 7920203.589.83.9018

Oct 18, 2012 · pattern of trade with the United States and the rest of the world. The Tupi field was discovered in 2007 by a consortium of Brazilian and foreign companies (Petrobras, BG Group, and Petrogal). The new discoveries lie off the southern coast of Brazil,