WPS8076 Policy Research Working Paper 8076 Energy Prices and International Trade Incorporating Input-Output Linkages H. Ron Chan Edward Manderson Fan Zhang South Asia Region Office of the Chief Economist May 2017 Policy Research Working Paper 8076 Abstract This paper examines the effect of energy costs on indus- analysis finds that ignoring input-output relationships can try export competitiveness. Most studies in the literature lead to significant over- or underestimates of the effect of use direct energy consumption (energy consumption at energy price shocks on exports, depending on intermediate the final stage of production) and domestic energy prices factor intensities and trade relationships. Using estimated to compute energy costs faced by domestic industries. trade elasticities, the study simulates the economic conse- Using multi-country input-output information, this study quences of energy cross-subsidies and carbon taxes. The measures the effect of aggregate energy costs on export per- results show that energy cross-subsidies that raise energy formance, where aggregate energy costs include not only tariffs on industry to support lower rates for households direct energy costs, but also indirect energy costs passed on and farmers in India could reduce the country’s net manu- through the upstream supply chain. This study develops a facturing exports by $6.1 billion a year. Similarly, a carbon theoretical trade model that incorporates tradable interme- tax that unilaterally increases energy prices by 10 percent diate goods to inform its empirical strategy. It then estimates in the European Union could reduce European Union- a reduced-form model using a panel data for 10 manu- wide net manufacturing exports by 1.9 percent annually. facturing sectors in 43 countries from 1991 to 2012. The This paper is a product of the Office of the Chief Economist, South Asia 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://econ.worldbank.org. The authors may be contacted at fzhang1@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 Energy Prices and International Trade: Incorporating Input-Output Linkages∗ H. Ron Chan† Edward Manderson‡ Fan Zhang§ University of Manchester University of Manchester World Bank Keywords: Energy cross-subsidization, carbon leakages, energy costs, exports, competitive- ness, input-output linkages JEL Classifications: Q4, Q5, F1 ∗ This paper is part of a broader analytical effort to assess the cost of power sector distortions in South Asia being conducted by the Office of the Chief Economist of the World Bank South Asia Region. We thank Vivien Foster, Martin Rama, Yadviga Viktorivna Semikolenova, Govinda Timilsina, Maria Vagliasindi, and seminar participants at the University of Toulouse for helpful comments on earlier drafts. Jiarui Wang provided excellent research assistance. Financial support from the Trade and Competitiveness Multi-Donor Trust Fund and the Partnership for South Asia Trust Fund is greatly appreciated. † School of Social Sciences, University of Manchester, Arthur Lewis Building-3.078, Oxford Road, Manchester, M13 9PL, United Kingdom. Contact: ron.chan@manchester.ac.uk ‡ School of Social Sciences, University of Manchester, Arthur Lewis Building-3.078, Oxford Road, Manchester, M13 9PL, United Kingdom. Contact: edward.manderson@manchester.ac.uk § World Bank, 1818 H Street NW, Washington, DC, USA. Contact: fzhang1@worldbank.org 1 Introduction Changes in energy prices are usually thought to have implications for the competitiveness of economies. Policies leading to higher energy prices therefore tend to be subjected to close scrutiny and much debate. Prominent examples include climate change policies such as environmental taxes and cap and trade systems. There are concerns that these poli- cies, when unilaterally enforced, could hinder the ability of domestic industry to compete in export markets, especially in the short term and for energy-intensive sectors. Through their effects on energy prices, they may even induce firms to shift production to countries where emission constraints are less stringent, a response that may lead to an increase in global carbon emissions – the so-called carbon leakage effect. Another policy that may lead to unintended consequences for trade through energy price adjustments is energy cross- subsidization. Many countries, especially in Eastern Europe and South Asia, impose higher energy tariffs on industry to support lower rates for residential or agricultural consumers. Energy cross-subsidies may therefore hamper the international competitiveness of indus- tries, although this has received little attention in the discussion on the economic incidence of energy subsidies. In this paper, we examine the effect of energy costs on manufacturing exports.1 Most stud- ies in the literature have used direct energy consumption (energy consumption at the final stage of production) and domestic energy prices to compute energy costs faced by domestic industries. We contribute to the literature by considering both direct and indirect energy costs (energy consumption at intermediate stages of production). Taking into account indi- rect energy costs is important because even if a sector is not energy intensive on the basis of direct energy consumption, it may still be vulnerable to energy price shocks if it uses energy-intensive intermediate inputs. Furthermore, a sector that uses imported intermedi- ate goods may also be indirectly affected by energy price changes in the origin countries. Our paper relates to two strands of empirical literature. One strand consists of a small number of recent studies that analyze the relationship between energy prices and interna- tional trade using econometric techniques. For example, Aldy and Pizer (2015) use historical electricity price data for the United States to investigate how production and net imports 1 Changesin energy prices could also affect transport costs, which in turn has implications for international trade flows (Bridgman, 2008; Chen and Hsu, 2012). Analysis on this aspect is beyond the scope of this paper. 2 change in response to energy prices. At the global level, Sato and Dechezleprêtre (2015) use energy price indices from 42 countries during 1996–2011 to study the impact of changes in energy price differences between trading partners on trade flows. Based on direct energy costs alone, these studies find a significant yet small negative effect of higher energy costs on export performance. Using input-output information, another strand of literature on the trade-energy nexus em- phasizes the importance of accounting for the indirect energy consumption of intermediate inputs. For example, Bordigoni, Hita and Le Blanc (2012) find that the energy content em- bodied in imported intermediate inputs is close to the total direct energy consumption of European industry. A significant part of the embodied energy costs of manufactured prod- ucts in Europe is therefore not affected by domestic energy price changes. Sato et al. (2015) evaluate the impact of embodied energy on energy security. They find that the geographic diversity of embodied energy imports is much greater than that of direct energy imports, and that there is considerable variation across countries in the diversification of embodied energy imports. This paper builds on both strands of literature by combining an econometric model of inter- national trade with multi-country input-output information to measure the effect of energy costs aggregated over all stages of production on export performance. We first develop a theoretical framework based on a Melitz-type trade model in which each country produces one variety in each sector and there are multiple countries and sectors with heterogeneous productivity. We incorporate tradable intermediate goods in the model to inform our em- pirical strategy. We then compute the embodied cost (direct plus indirect cost) of factor inputs (electricity, natural gas and labor) using the OECD inter-country input-output ta- bles for 10 sectors of production. The key parameters of interest are estimated by Poisson pseudo maximum likelihood using panel data for 43 countries from 1991 to 2012. We control for exporter-by-importer fixed effects, exporter-year fixed effects, importer-year fixed effects and sector-year fixed effects to sweep out variations due to bilateral trade resistance (such as distance, and differences in language and culture) and time-varying shocks at country and sector levels (such as exchange rate fluctuations and regulatory and technology changes). The causal effect of energy costs on trade is identified by country-sector-year variation in- cluding differences in electricity prices, sector-specific energy intensity, and inter-country 3 inter-sectoral production linkages. We also conduct various sensitivity analyses, such as considering alternative levels of sector disaggregation, accounting for the delayed response of trade to changes in input costs, allowing for heterogeneous effects of energy costs based on characteristics of the trading partners, and controlling for country-sector-year fixed ef- fects to avoid potential bias from possible omitted variables (such as adjustment in firms’ markup). We find a robust and asymmetric effect of the energy costs of exporting and importing countries on trade. A one percent increase in the exporter’s (importer’s) aggregate electric- ity costs is associated on average with an approximately 0.07 percent decrease (0.10 percent increase) in exports for a sector of average electricity intensity. Aggregate natural gas costs also have a significant impact on trade, although of a smaller magnitude. Using the esti- mated trade elasticities, we simulate the economic consequences of energy cross-subsidies in Bulgaria and India and those of a carbon tax in the European Union. We find that en- ergy cross-subsidies imposing a 15 percent implicit electricity tax on industry in India could reduce its net manufacturing exports by about $6.1 billion a year. Similarly, a carbon tax unilaterally increasing energy prices by 10 percent in the European Union could reduce EU-wide net manufacturing exports by 1.9 percent annually. Our results suggest that input-output relationships are important for understanding the energy-trade relationship. Some sectors use significantly more indirect energy (at inter- mediate stages) than direct energy. For these sectors, when intermediate goods are mostly non-traded, a country’s own energy price change may have a substantial multiplier effect on its exports through domestic sectoral linkages. A case in point is India’s machinery sector, whose indirect energy intensity is almost five times its direct energy intensity. Because India is largely self-sufficient in intermediate inputs, a model that ignores inter-sectoral linkages would underestimate the impact of higher energy costs on the net exports of the country’s machinery sector by a factor of almost four. Conversely, if indirect energy is mostly im- ported, changes in domestic energy prices would be relatively less important for a sector’s exports. This is the case for Bulgaria. For an EU-wide carbon tax, we find that a model accounting only for direct energy consumption would underestimate the impact of higher energy prices on EU-wide net manufacturing exports by a factor of two. This reflects the intensive trading of intermediate goods within the European Union, which increases each 4 sector’s exposure to higher energy prices through regional multiplier effects. The rest of this paper is organised as follows. Section 2 outlines the theoretical framework. Section 3 discusses the empirical approach and data. Section 4 presents empirical and simulation results. Section 5 concludes. 2 Theoretical Framework The objective of this section is to motivate an empirical model in which we can test the effect of energy prices on trade. We build a Melitz-type model in which there are multi- ple countries and sectors, and in which firms are heterogeneous in productivity. Most of the discussions here follows Eaton and Kortum (2002); Anderson and van Wincoop (2003, 2004); Chaney (2008); Arkolakis, Costinot and Rodríguez-Clare (2012); Costinot, Donaldson and Komunjer (2012); Head and Mayer (2014); Costinot and Rodríguez-Clare (2014); and Caliendo and Parro (2015). We outline the basic model with multiple sectors (and factors) and then move on to include tradable intermediate goods. 2.1 Many Sectors and Factors We consider a partial equilibrium framework in which there are N countries and K sectors, and we treat input prices as exogenous. In the following, subscript i ( j) refers to country i ( j) and subscript k to sector k. Following Costinot, Donaldson and Komunjer (2012), Costinot and Rodríguez-Clare (2014), and others, the representative consumer has a two- tier utility function in which the upper tier follows Cobb-Douglas and the lower tier follows constant elasticity of substitution (CES) utility function. Formally, the utility function for the representative consumer in country j is given by: K ∏ C j,k βk Cj = (1) k =1 where: σk σk −1 σk −1 C j,k = c j,k ( ω ) σk dω (2) ω ∈Ω 5 where β k is the income share of sector k, ω represents each variety, Ω is the entire variety space, and σk is the elasticity of substitution for varieties within sector k. Each firm needs to pay a fixed cost for each variety, has increasing returns, and has no extra cost to diversify horizontally. Therefore, firms in each country produce that country’s own variety; in other words, each variety is sourced from one country only. We can therefore rewrite (2) as: σk N σk −1 σk −1 C j,k = ∑ ω ∈Ωij,k c j,k ( ω ) σk dω (3) i =1 where Ωij,k is the variety space in sector k that country j imports from country i; both Ω and Ωij,k are endogenously determined, as described below. Assume that the representative consumer in country j receives an income of Yj . The total value of exports (of good k with variety ω ) from country i to country j is given by: k −1 1−σk xij,k (ω ) = Pjσ ,k β k Yj ( pij,k ( ω )) (4) where Pj,k is the price index for sector k in country j, defined as: 1 N 1−σk ∑ 1−σk Pj,k = Pij ,k (5) i =1 and: 1 1−σk 1−σk Pij,k = pij,k (ω ) dω (6) Ωij,k where pij,k (ω ) is the price charged by each firm ω . Each firm pays a fixed cost of exports (and production) f ij , and a variable cost of production depends on the cost of M factors. In equilibrium, each firm chooses to export to only one country because of increasing returns. The total cost for a firm in country i, producing quantity q and shipping to country j, is given by: 2 M q ∏ w m ,i α m,k TCij,k (q) = f ij + τ (7) ϕ ij,k m =1 where m = 1, 2, ..., M denotes each primary factor; τ is iceberg transport cost between i and 2 The resulting cost function (7) is an equilibrium product of a producer with a Cobb-Douglas production function. 6 j; ϕ is the productivity term drawn from a continuous distribution function Gk ( ϕ); αm,k is the factor intensity in m, which is allowed to differ from sector to sector and meets the condition ∑m αm,k = 1; wm,i is the input price for factor m in country i; and the firm is assumed to be a price taker in the factor markets. Assume that all factors are used in a perfectly competitive, nontradable, homogenous-good sector in all countries and that the returns to factors, w, M α are predetermined. Define Wi,k ≡ ∏m m,k =1 wm,i as the marginal cost. Since the demand for each variety in sector k is isoelastic (with price elasticity of σk ), monopolistically competitive firms price a markup over marginal cost: σk τij pij,k = W (8) σk − 1 ϕ i,k Following Helpman, Melitz and Yeaple (2004), Chaney (2008), and Costinot and Rodríguez- Clare (2014), we assume that firms’ productivity follows a Pareto distribution: Gk ( ϕ) = 1 − ϕ−γk (9) where γk is the shape parameter. Because of the presence of a fixed cost of exporting, low- productivity firms will choose not to produce and export (extensive margins). Applying the demand and price functions described in equations (4) and (8) and setting profits equal to zero, we therefore have the cutoff productivity: 1 ∗ f ij σk −1 τij Wi,k ϕij ,k = Aσk · (10) β k Yj Pj,k where Aσk is a parameter that depends on σk only. The aggregate exports from country i to country j in sector k are given by: ∞ Xij,k = xij,k (ω )dGk (ω ) (11) ∗ ϕij ,k Plugging equations (4), (8), (9) and (10) into equation (11), we have: − γs γs τij Wi,s γs Xij,s = Aσs ,γs β s Yj σs −1 f ij (1− σs −1 ) (12) Pj,s 7 Taking a logarithmic transformation on both sides gives:3 γk γk log Xij,k = log β k + log Yj − γk log Wi,k + γk log Pj,k + 1− log f ij − γk log τij σk − 1 σk − 1 (13) The first term on the right-hand side of equation (13) takes into account the demand-side factors while the fourth term takes into account the bilateral trade cost between countries i and j. We hope to control for these two factors using with the first factor controlled by an importer-year fixed effect and the last one by an exporter-importer fixed effect. We will explain more on this in Section 3. The second term, log Wi,k , takes into account the input costs facing sector k in country i. Namely: M log Wi,k = ∑ αm,k log(wm,i ) (14) m =1 which is the main variable of interest in our empirical model. The third term, log Pj,k , represents the multilateral resistance term first defined in Anderson and van Wincoop (2003). The price index in an importing country, as illustrated in equation (5), depends on the prices (and thus costs and trade barriers) of country j and all other countries. In the empirical model, we use factor input costs of importing countries as a proxy for log Pj,k . Note that even though the model suggests that the coefficients of log Wi,k and log Pj,k have the same size (and opposite signs), the effects of exporters’ and importers’ energy costs on trade are likely to differ, because we are using importers’ costs to approxi- mate for the price index. Our empirical model described in Section 3 allows for asymmetric effects of exporters’ and importers’ costs on trade. 2.2 Intermediate Goods In the previous subsection, each sector uses a bundle of factors at different factor intensities. However, there are no linkages among sectors. Each sector produces its good using M fac- tors and each sector can be analyzed separately. In this subsection, we relax this restriction and allow each sector to use all the factors as well as intermediate goods from other sectors. Following the literature, we assume that there are a number of perfectly competitive firms 3 log( A σs ,γs ) is a constant that captured by sector(-by-year) fixed effect. It is omitted in equation (13) to simplify the notation. 8 that produce a composite intermediate good Hk . The aggregation of this intermediate good Hk originated from sector k is produced in exactly the same way as the final consumption good from sector k, that is: σk σk −1 σk −1 Hi,k = h i ,k ( ω ) σk dω (15) ω ∈Ωk The composite intermediate good producer in country i that produces H by procuring in- termediate goods (domestically and abroad) has the following demand for each variety of goods: −σk p i ,k ( ω ) h i ,k ( ω ) = Hi,k (16) Pi,k Here the price index Pi,k represents both the price index for the composite intermediate goods and the price index for final (consumption) goods, as the same set of firms export their varieties to the composite intermediate goods producer and to country i for final goods. Following the literature, we also assume that the producer of a variety produces the same proportion of that variety for both final goods and intermediate goods. The production of a variety in sector k depends on all M factors and all intermediate goods from the other K sectors. Therefore, in the presence of intermediate goods, the total production and export cost, described in equation (7), is modified as follows: φ 1− φ M K q αM αS TCij,k (q) = f ij + τij,k ϕ ∏ wmm ,i ∏ Pi,ll (17) m =1 l =1 Wi,k The assumption on intermediate goods aggregation leads to the same price elasticity of demand for intermediate goods as that for final goods. We therefore have the same format of markup-over-marginal-cost relationship as described in equation (8); the only difference is that marginal cost (Wi,k ) now also depends on the price of intermediate goods in the form of a Cobb-Douglas cost function. After incorporating intermediate goods, the estimation model will look almost exactly like equation (13) except that we will now replace Wi,k , which represents the direct costs of primary inputs, with Wi,k , which represents the aggregate (direct plus indirect) costs of the factor inputs. The estimation equations for each of the K sectors derived from equation (13) 9 are now interlinked and are no longer independent. In other words, price indices { Pi,k } depend on the input costs of other sectors. In the next section, we explain how to use an input-output matrix to compute the aggregate cost of factor inputs which is a function of factor input prices and factor intensities. 3 Empirical Strategy In this section, we first estimate the theoretical model developed in Section 2.1 in which each sector uses its own factors of production but there are no linkages between sectors. In this case, we consider the impact of only the the direct costs of factor inputs on trade. We then extend this analysis by estimating the theoretical model developed in Section 2.2, where each sector purchases intermediate goods produced by other sectors as described by input-output linkages. This more realistic approach allows us to estimate the effect of aggregate energy costs on export performance. 3.1 Baseline Model Our baseline model measures the effects of direct electricity, natural gas and labor costs on bilateral exports at the sector level.4 We estimate the model using the Poisson pseudo- maximum likelihood (PPML) estimator proposed by Santos Silva and Tenreyro (2006). This is a weighted nonlinear least squares estimator, and although it is a count data estimator, it can be used for continuous variables (such as exports) because it does not require the data to strictly follow a Poisson distribution. We use this estimator because it delivers consistent estimates of gravity equations under very general conditions and performs well even when the dependent variable displays a relatively large proportion of zero observations. Using the example of a gravity model, Santos Silva and Tenreyro (2006) demonstrate that the PPML estimator outperforms a log-linear specification estimated by ordinary least squares by providing estimates that are robust to different patterns of heteroskedasticity. Our baseline empirical model is derived directly from the theoretical analysis. From equa- 4 Ideally, we would also like to examine the effect of coal prices on international trade. However, coal prices are determined mostly through privately-negotiated contracts and are largely not publicly available. 10 tion (13), the model to be estimated takes the following form:   Exportsijkt = exp ∑ α1s log( ElectricityCostskt ) + α2s log( NGCostskt ) {i, j } s= +α3s log( LaborCostskt ) + α4 log(SectorOutputikt ) + δij + δit + δjt + δkt + ε ijkt (18) where Exportsijkt represents annual bilateral trade flows between exporting country i and importing country j in sector k in year t. ElectricityCost, NGCost, and LaborCost measure the direct unit input costs of electricity, natural gas, and labor respectively. These cost terms are calculated by interacting input prices with intensities, defined as input expenditure per unit of value added. Following our theoretical framework, we expect negative coefficients for the input costs of exporters (α1i , α2i , and α3i ), and positive coefficients for the input costs of importers (α1 j , α2 j , and α3 j ). Equation (18) also includes SectorOutput as a measure of the gross economic output of sector k in exporting country i. It controls for the possibility that the volume of exports is greater the bigger the scale of the sector. Thus we expect α4 to be positive. An advantage of our panel specification is that it allows for the inclusion of various fixed effects that can help mitigate biases from omitted variables. In particular, δij represents fixed effects that control for time-invariant determinants of trade flows between each coun- try pair. These include distance, common border, and differences in religion, culture, and language between trading partners. δit and δjt are exporter-year and importer-year fixed effects, respectively, which capture all time-varying country-specific factors that affect the trade flows of the exporting and importing countries. These include exchange rate regimes, infrastructure investments, and regulatory changes that affect export performance. δkt rep- resents sector-time effects that capture sector-specific shocks to bilateral trade. Finally, ε ijkt denotes an error term. Our estimation strategy relies on energy prices having an effect on trade that is conditional on the energy intensities of the sector. Since energy prices are observed at the national level, it is the interaction with the electricity intensity and natural gas intensity that generates inter-sectoral variation and allows the coefficients α1i , α1 j and α2i ,α2 j to be identified. 11 Our econometric model includes separate input cost terms for exporters and importers to allow for potential asymmetric effects of the energy costs of the exporting and importing countries on bilateral trade. The theoretical framework suggests that there may be asym- metries in the magnitudes of the estimated cost impacts. The reason is that domestic costs of production are expected to have a direct impact on the exports of country i, while input costs in country j will indirectly affect exports from country i to country j by affecting the prevailing product price in country i. In our econometric model, we test for statistically significant differences between (the absolute values of) α1i and α1 j for electricity costs and between (the absolute values of) α2i and α2 j for natural gas costs. 3.2 Input-Output Extended Model The econometric model developed in the previous subsection estimates the effect of direct energy costs on trade. Higher factor prices may also have an indirect effect on trade flows by raising the cost of production of intermediate inputs, if these costs are then passed on by input suppliers. Furthermore, if a sector’s intermediate inputs are imported, then both domestic and foreign factor prices and the factor intensities of intermediate and direct factor inputs would affect the domestic costs of production. Because we observe input- output linkages across countries, we assume that each sector in each country produces a specific intermediate good. Each sector uses outputs from other sectors (as well as its own output) as intermediate inputs; that is, there are N × K intermediate goods. In this case the Cobb-Douglas cost function for sector k in country i can be written as: K N Ci,k = Wi,kk ∏ ∏ Pj,s β s, j,k α (19) s =1 j =1 where W is the cost of direct factor inputs and Pj,s is the price of intermediate inputs pro- duced by sector s in country j, and αs, j,k is sector k’s intensity of the intermediate good produced by sector s in country j . Thus in total there are NK + 1 inputs to production. Taking logarithms on both sides of equation (19) gives: log Ci,k = β k log Wi,k + ∑ ∑ αs, j,k log Pj,s (20) s j 12 The price of intermediate goods from sector s in country j ( Pj,s ) is assumed to be the sum of production costs and a markup, that is, Pj,s = (1 + η j,s )Cj,s , where η j,s is the markup: log Pj,s = log CjP ,s + log(1 + η j,s ) (21) Substituting equation (21) for the price term in equation (20) gives: log Ci,k = β k · log Wi,k + ∑ ∑ αs, j,k log Cj,s + log(1 + η j,s ) (22) s j Equation (22) can be rewritten in matrix form as the following: CNK×1 = β NK×1 · WNK×1 + A NK× NK CNK×1 + Ψ NK×1 (23) where the elements within A include the α terms (i.e. input-output linkages among sectors) and the elements within Ψ include the markup terms. The first item on the right-hand side of this equation calculates the dot product of β and W . Assuming that the mark-up Ψ is stable over time and can be controlled for by a set of fixed effects, we can rewrite equation (23) as the following by dropping the markup term: −1 Ci = 1 − A β·W (24) Using the input-output relationship (matrix A), the factor intensities of all sectors (vector β) and a vector of input prices W , we are able to compute the direct plus indirect cost of factor inputs. In the input-output extended model, we replace the direct cost terms (calculated as the interaction terms of factor intensities and input prices) with the direct plus indirect costs described in equation (24). It is worth emphasizing that by using input- output tables, we assume that intersectoral linkages are exogenous – a (small) change in input prices would not shift the input-output relationship drastically. We do not account for substitution possibilities among primary factors and intermediate inputs. In this way, we should interpret the results from the input-output extended model as an estimation of short-term effects. 13 3.3 Data We estimate our empirical model using energy and bilateral trade data for 10 manufacturing sectors in 43 countries from 1991 to 2012. The full set of countries included in our analysis is listed in Table 1. We use data on electricity and natural gas prices from the International Energy Agency (IEA) Energy Prices and Taxes database, the Energy Regulators Regional Association (ERRA) Tar- iff Database, and various government and media reports. The IEA database records after-tax prices for residential and industrial consumers, while the ERRA database records after-tax prices for residential and non-residential consumers. For a subset of nine countries covered by both the IEA and the ERRA, we cross-check the price data from the two datasets and find that the data are fairly consistent. We use IEA data on industrial energy prices wher- ever available. We use ERRA data on nonresidential energy prices for one country (Saudi Arabia) that is not covered by the IEA. Table 2 lists the sources of energy price data for each country. Energy consumption data are from the IEA World Energy Statistics and Balances database. This database reports final electricity and natural gas consumption for 10 manufacturing sectors at the two-digit level of the international Standard Industrial Classification, Revision 3.1 (ISIC Rev. 3.1). One of the sectors is labelled “non-specified". According to IEA docu- mentation, this sector includes rubber and rubber products as well as any manufacturing activities not covered by the other nine sectors. Because there may be measurement errors associated with the energy intensity of this miscellaneous category, we conduct a robustness check by excluding the “non-specified” sector from the sample.5 We gather data on manufacturing value added, employment costs and number of employees from the United Nations Industrial Development Organization (UNIDO) Industrial Statis- tics Database (INDSTAT2). The sector-specific wages are calculated by dividing employment costs by the total number of total employees. Factor intensities are measured as the value of factor inputs divided by value added.6 5 The IEA notes that for some countries, “non-specified” could be a sector average where a sector-wise breakdown is not possible. 6 About 1.5 percent of the observations have factor intensity values below zero or greater than one. We dropped these observations as outliers. 14 The multicountry inter-sectoral input-output relationships are obtained from the OECD Inter-Country Input-Output Database (ICIO). The ICIO tables are available for the years 1995, 2000, 2005, 2008, 2009, 2010 and 2011. As our sample spans the period from 1991 to 2012, we use the 1995 edition of ICIO table for samples in 1991-1995, the 2000 edition for 1996-2000, the 2005 edition for 2001 to 2005, and the 2011 edition for 2011 and 2012. While both the IEA and the ICIO tables rely on ISIC Rev 3.1 to classify industries, ICIO sectors are slightly more disaggregated than IEA sectors because some IEA sectors are aggregated from multiple ISIC two-digit sectors. For the main analysis, we aggregate ICIO sectors to match the IEA data. We also apply different levels of sector disaggregation for robustness checks. Table 3 shows the correspondence between IEA and ICIO sectors. Bilateral trade data are obtained from the United Nations COMTRADE Database. Trade data are reported at the six-digit Harmonized Commodity Description and Coding System (HS) level. Using the concordance provided by Pierce and Schott (2012), we convert the data from the HS level to the ISIC Rev. 3.1. We always use the trade values reported by the exporters. Manufacturing value added, employment costs, export value and IEA and ERRA energy prices are reported in current prices denominated in US dollars. Energy prices from gov- ernment reports are reported in current prices denominated in local currencies. We convert local currencies into U.S. dollars using the exchange rates reported by the World Bank. We convert values for different years into 2010 prices using country-specific GDP deflators obtained from the International Monetary Fund. 3.4 Descriptive Statistics Table 4 provides summary statistics for the key variables. Table 5 shows how intensively each row sector uses inputs from each column sector. The numbers are the weighted aver- age shares of factor inputs across all countries in the sample in 2010. While the production of most sectors relies largely on inputs from subsectors within their own two-digit classifi- cation, there are strong interdependencies among the machinery, transport equipment and metals sectors.7 For example, for the manufacture of transport equipment, 12 percent of 7 In addition, because the “non-specified’" sector includes rubber and plastic products, a large share of its inputs are from the chemicals sector. 15 inputs are from the metals sector, and 24 percent from the machinery sector. Table 6 lists average direct factor intensity and aggregate factor intensity by sector. It shows that for sectors that rely heavily on intermediate goods from energy-intensive sectors, there is a substantial difference between direct and indirect energy intensities. For example, the indirect energy intensity of the machinery and transport equipment sectors is two to three times higher than their direct energy intensity. Ignoring indirect energy consumption would therefore significantly underestimate the true energy cost faced by these sectors. There is also significant variation in countries’ trade dependency on intermediate goods, as shown in Figure 1. In Bulgaria, for example, about 50 percent of intermediate goods for production are imported and close to 60 percent of the outputs produced as intermediate goods are destined for the export market. In India, in contrast, less than 20 percent of intermediate goods are imported or exported. 4 Empirical and Simulation results 4.1 Regression results Table 7 reports the main estimation results. Columns (1) and (3) correspond to regressions for the baseline model, which accounts for direct factor costs only. Columns (2) and (4) correspond to regressions for the input-output extended model, which accounts for direct and indirect energy costs based on equation (24). In addition, columns (1) and (2) show results for the full sample estimation, while columns (3) and (4) show results for regressions that exclude the “non-specified" sector. Because our cost variables involve interaction terms of factor prices and intensities, the size of the impact of factor price changes on exports depends on factor intensities. To ease interpretation of the coefficients, each factor intensity is scaled by its sample average so that the value is equal to 1 for a sector of average factor intensity. This means that the coefficients of factor costs in the baseline model can be inter- preted as the estimated percentage change in bilateral exports for a 1 percent increase in the factor price for a sector of average factor intensity. For consistency, factor intensities in the input-output extended model are scaled by the same sample mean. Table 7 shows that bilateral exports are negatively correlated with origin countries’ electric- 16 ity costs and positively correlated with destination countries’ electricity costs. The coeffi- cients are statistically significant in all models. Evaluated at the average electricity intensity, the baseline model suggests that a 1 percent increase in exporters’ electricity cost is asso- ciated with a reduction in exports of 0.11 percent. In contrast, the input-output extended model predicts a smaller export reduction of 0.07 percent. The estimated export elasticity of importers’ electricity cost is 0.14 based on the baseline model, and 0.10 based on the input-output extended models. Excluding the “non-specified" sector from the regression produces similar results. It is not surprising that the estimated trade elasticities are different between the baseline and input-output extended models. In the input-output model, the electricity cost term measures direct plus indirect electricity costs. Thus a 1 percent increase in this cost could reflect an increase in domestic electricity costs or an increase in the electricity costs of up- stream trading partners that supply intermediate inputs – or both. In the baseline model, in contrast, a 1 percent increase in electricity costs is entirely an increase in the domestic electricity price. The coefficients of electricity costs in the two models may therefore differ substantially for a sector if a large share of its electricity costs are embodied in intermedi- ate inputs. Such differences become more salient when we conduct simulation analysis for different countries and sectors. The trade elasticities of natural gas costs are smaller than those of electricity costs. The input-output extended model predicts that a 1 percent increase in aggregate natural gas cost for origin countries would cause a 0.04 percent reduction in exports, while a 1 percent increase for destination countries’ would cause a 0.04 percent increase in exports. These coefficients are statistically significant. The coefficients estimated by the baseline model are of the expected sign but are not statistically significant. Coefficients on the labor cost terms are all significant and of the expected sign. The effect of labor costs on trade is in general much greater than the effect of energy costs. The export elasticity of origin countries’ labor costs is between -0.16 and -0.22 across various models, while the elasticity of destination countries’ labor costs ranges between 0.15 and 0.22. Table 7 also reveals evidence of asymmetric effects of exporter and importer factor costs, especially for electricity. In particular, the importer factor costs are found to have a larger 17 absolute effect on bilateral trade than the exporter factor costs across all models and the difference (of their magnitudes) is statistically significant. Finally, consistent with predic- tions of the gravity model, estimates of all models show that sector output is significantly positively correlated with export and the elasticity is close to 1. 4.2 Robustness Check In this subsection, we conduct sensitivity analysis using alternative model specifications. First, we consider the possibility that sector output may be endogenous to export value, especially for export-oriented sectors. As a robustness check, we move sector output to the left-hand side of equation (18) and redefine our dependent variable as export intensity, that is, the ratio of exports to sector output. The results are reported in Table 8. The resulting coefficients are almost identical to those of the main model. This is expected, since the estimated trade elasticity of sector output in the main model is close to 1. Second, we consider the possibility that bilateral trade may not respond to price changes immediately as a result of market imperfections or adjustment costs. There could also be contemporaneous feedback from exports to energy prices if export-oriented sectors dom- inate domestic industrial energy consumption. To account for delayed responses and the potential endogeneity of energy prices, we control for one-year lagged explanatory variables rather than contemporaneous variables. The resulting coefficients, shown in Table 9 are very similar in significance and size. Third, we consider alternative levels of sector disaggregation. In previous regressions, we mapped export, ICIO, and value added data into the 10 manufacturing sectors used by the IEA so as to match energy consumption data. To account for potential aggregation bias in the factor content of trade (Feenstra and Hanson, 1999), we reestimate trade elasticities according to the sector aggregation used by the ICIO tables (15 sectors) as well as a sector aggregation based on the three-digit ISIC Rev. 3.1 (54 industries). Because energy consump- tion data are not available at these more disaggregated levels, we assume that the energy intensity of industries at the two-digit and three-digit levels of ISIC Rev 3.1 is the same if they belong to the same IEA sector. Regression results based on the alternative disaggrega- tion levels, reported in Tables 10 and 11, are fairly close to the main results, although with 18 a lower goodness of fit. Fourth, we consider alternative treatment of firms’ markup. Our estimation strategy relies on the assumption that markup is stable over time so that it is absorbed in the country pair fixed effects. If this assumption is violated, our estimation of the input-output model could be biased because markup could be correlated with both energy cost and trade value. To control for potential variation in markup over time, we estimate an alternative model with two additional sets of fixed effects: importer-sector-year fixed effects and exporter-sector- year fixed effects. Now variations in the factor input costs of exporters and importers, as well as in the sector outputs of exporters, would also be absorbed in these fixed effects. To identify the effects of energy costs on trade value, we control for a cost gap term (exporter cost minus importer cost) rather than separate exporter and importer cost terms. The re- gression results using cost gap terms, with and without this alternative set of fixed effects, are reported in Tables 12 and 13 respectively. The coefficients of electricity, natural gas and labor cost gap terms in the input-output extended model are all negative and statistically significant. In addition, their sizes are well within the range of the importer and exporter trade elasticities from the main model. These results suggest that potential biases from the omitted variable of firms’ markup may not be a concern. Finally, we investigate potential heterogeneous effects of factor costs on trade according to the characteristics of pairs of trading partners. In particular, we include in equation (18) interaction terms between factor costs and the following variables: a dummy variable equal to 1 if the trading partners are contiguous and 0 otherwise; a dummy variable equal to 1 if the trading partners belong to the same trading bloc (such as European Union or NAFTA) and 0 otherwise; a dummy variable equal to 1 if the trading partners are both in the euro zone and 0 otherwise; a continuous variable measuring the absolute distance between the trading partners; and a continuous variable measuring the relative distance between the trading partners.8 To simplify notation, we report estimation results with controls for cost gap terms rather than separate cost terms for exporters’ and importers’ factor inputs. The 8 Relative distance is a measure of GDP-weighted distance from the perspective of exporting country, that is, Distij Yj RelDistij = × ∑k Distik YW where Yj is the importing country’s GDP, YW is the world’s GDP, Distij is the great circle distance (measured in km) between two countries i and j. 19 results are reported in Table 14. The results indicate that while including interaction terms for contiguous countries and trading blocs does not change the estimation results from the input-output extended model, it tends to increase the absolute value of trade elasticities of direct factor costs in the baseline model. That is, when two countries share a common border or belong to the same trading bloc, the effect of the energy or labor cost gap is increased. Another finding from Table 14 is that distance could exacerbate the effect of energy costs on trade. According to the input- output extended model, a one-mile increase in the relative or absolute distance between two trading partners is associated with a 0.01 percentage point increase in the effect of the electricity cost gap. In other words, the greater the distance the more important is the effect of electricity cost on trade. 4.3 Simulation Analysis Next we analyze the economic consequences of two policies using the estimated trade elas- ticities reported in Section 4.1.9 The first policy concerns the use of energy cross-subsidies and its impact on industrial competitiveness. Energy cross-subsidies place an implicit tax on industry. In India, for example, the electricity price for industry is on average 15 percent higher than the cost of supply, to support lower tariffs for residential and agriculture con- sumers. In this counterfactual, we ask how manufacturing exports in Bulgaria and India would change with the removal of a 15 percent implicit tax on electricity. Bulgaria and India are selected for the analysis because they represent two different patterns of trade in intermediate goods, as discussed in Section 3.4: India has a strikingly low level of imports and exports of intermediate goods, while Bulgaria is much more engaged in global value chains (see Figure 1). The simulation analysis shows that a country’s level of participation in global value chains significantly affects the correlation between domestic energy price shocks and exports. The second counterfactual considers the implications of carbon pricing for international trade. The EU Emissions Trading System created the world’s first and largest carbon market. A number of countries outside Europe, such as China and the United States, have adopted 9 Counterfactuals based on alternative estimates of trade elasticities reported in Section 4.2 are similar and do not change our main findings. 20 similar programs. Carbon trading schemes put a price on carbon emissions and would increase the cost of fossil-fuel based energy in countries where they are in force. There is concern that a unilateral adoption of carbon regulation could lead to carbon leakage. Indeed, the European Commission considers that energy- and trade-intensive sectors face a higher risk of carbon leakage. To protect their competitiveness, these sectors are granted a higher share of free emissions allowances. To gauge the potential magnitude of carbon leakage, we simulate the impact of a 10 percent unilateral price increase for electricity in the European Union on EU manufacturing exports. Table 15 describes the impact of removing energy cross-subsidies on export and net export values. According to the input-output extended model, removing a 15 percent implicit elec- tricity tax on industry would increase India’s net manufacturing exports by 9.5 percent, or around $6.1 billion a year. Since India is quite self-sufficient in the production of interme- diate inputs, multiplier effects of energy price shocks are accumulated along the supply chain. The baseline model, which considers energy requirements only at the final stage of production, therefore underestimates the impact of energy price shocks by a factor of almost two. The pattern becomes more conspicuous for the machinery and transport equipment sectors, for which indirect energy consumption accounts for a large share of total energy consumption. For the machinery sector, ignoring indirect energy consumption would result in an underestimation of the impact of energy price shocks by a factor of four. The story is different for Bulgaria. Because Bulgaria has extensive exchanges of intermedi- ate inputs with other countries, the competitive advantage for its exporters from a decline in energy prices is partially offset by the decline in the input costs of its trading partners. The estimation from the input-output extended model indicates an increase in net manufactur- ing exports of 5.7 percent, while the estimation from the baseline model shows an increase of 7.0 percent. At the sectoral level, the difference between the baseline and input-output estimations for the machinery sector is somewhat smaller than that for India. The reason is that Bulgaria’s greater reliance on inputs from outside the country means smaller multiplier effects within the domestic supply chain. Table 16 describes the effects of a 10 percent across-the-board increase in electricity prices in the European Union on the exports of EU and non-EU countries. Exports decline for all sectors in the European Union. And for each sector, the input-output model predicts a larger 21 decline than the baseline model does. The impact on net exports ranges from 0.9 percent for the transport equipment sector (with the lowest aggregate energy intensity) to 4.4 percent for the basic metals sector (with the highest). For the European Union as a whole, the input- output model predicts an overall impact of a 1.9 percent reduction in net manufacturing exports annually. This is almost three times the size of the reduction predicted by the baseline model. The difference reflects multiplier effects within the European Union where there is intensive trade in intermediate goods between EU firms. 5 Conclusion This paper seeks to contribute to the literature on energy prices and international trade by taking into account the effect of embodied energy costs (costs of direct plus indirect energy consumption) on exports. We start with the premise that energy price shocks affect not only the cost of direct energy consumption (at the final stage of production) but also the cost of indirect energy consumption embodied in intermediate inputs. In addition, with the rapid expansion of global value chains, domestic energy prices may not tell the full story if a firm relies heavily on imported intermediate goods that are energy intensive. We develop a theoretical framework that incorporates tradable intermediate inputs in a Meliz-type trade model and apply the theoretical model to energy, trade, and multi-country input-output data for 10 manufacturing sectors in 43 countries from 1991 to 2012. We identify a robust effect of energy costs on export performance. We use the estimated trade elasticities to simulate two counterfactuals: the removal of energy-cross subsidies in Bulgaria and India, and a unilateral 10 percent increase in electricity prices in the European Union. We find that removing cross-subsidies which impose a 15 percent implicit electricity tax on industry in India could increase its net manufacturing exports by as much as 9.5 percent, or $6.1 billion a year. A carbon tax that unilaterally increases electricity prices by 10 percent in the European Union could cause an EU-wide reduction in net manufacturing exports of 1.9 percent annually. The simulation analysis also yields a nuanced understanding of the causal link between energy price shocks and exports. First, even if a sector is not energy intensive on the basis of direct energy consumption, it may still be vulnerable to energy price shocks if 22 it uses energy-intensive intermediate inputs. Second, if a country is not actively engaged in global value chains, domestic price shocks could have a multiplier effect on its overall competitiveness. However, it is worth noting that in cases where the input-output model predicts a larger effect for energy costs, the effect is still small in magnitude compared to that of labor costs. The predicted trade elasticity of labor costs is several times higher than that of energy costs in all cases. Our results should be interpreted as an upper-bound estimation because we do not ac- count for substitution possibilities among primary factors and intermediate inputs, as well as the substitution of intermediates imports among countries. Analysis of long-run dy- namic impacts such as those based on a computable general equilibrium model could yield new insights into the effect of energy price changes on industrial competitiveness in the international market. 23 References Aldy, Joseph E., and William A. Pizer. 2015. “The Competitiveness Impacts of Cli- mate Change Mitigation Policies.” Journal of the Association of Environmental and Resource Economists, 2(4): 565–595. Anderson, James, and Eric van Wincoop. 2003. “Gravity with Gravitas: A Solution to the Border Puzzle.” American Economic Review, 93(1): 170–192. Anderson, James, and Eric van Wincoop. 2004. “Trade Costs.” Journal of Economic Literature, 42(3): 691–751. Arkolakis, Costas, Arnaud Costinot, and Andrés Rodríguez-Clare. 2012. “New Trade Models, Same Old Gains?” American Economic Review, 102(1): 94–130. Bordigoni, Mathieu, Alain Hita, and Gilles Le Blanc. 2012. “Role of Embodied Energy in the European Manufacturing Industry: Application to Short-Term Impacts of a Carbon Tax.” Energy Policy, 43: 335–350. Bridgman, Benjamin. 2008. “Energy Prices and the Expansion of World Trade.” Review of Economic Dynamics, 11(4): 906–916. Caliendo, Lorenzo, and Fernando Parro. 2015. “Estimates of the Trade and Welfare Effects of NAFTA.” Review of Economic Studies, 82(1): 1–44. Chaney, Thomas. 2008. “Distorted Gravity: The Intensive and Extensive Margins of Inter- national Trade.” American Economic Review, 98(4): 1707–21. Chen, Shiu-Sheng, and Kai-Wei Hsu. 2012. “Reverse globalization: Does high oil price volatility discourage international trade?” Energy Economics, 34(5): 1634–1643. Costinot, Arnaud, and Andrés Rodríguez-Clare. 2014. “Trade Theory with Numbers: Quantifying the Consequences of Globalization.” Vol. 4 of Handbook of International Eco- nomics, Chapter 4. Elsevier. Costinot, Arnaud, Dave Donaldson, and Ivana Komunjer. 2012. “What Goods Do Coun- tries Trade? A Quantitative Exploration of Ricardo’s Ideas.” Review of Economic Studies, 79(2): 581–608. 24 Eaton, Jonathan, and Samuel Kortum. 2002. “Technology, Geography, and Trade.” Econo- metrica, 70(5): 1741–1779. Feenstra, Robert C., and Gordon H. Hanson. 1999. “The Impact of Outsourcing and High- Technology Capital on Wages: Estimates for the United States, 1979-1990.” The Quarterly Journal of Economics, 114(3): 907–940. Head, Keith, and Thierry Mayer. 2014. “Gravity Equations: Workhorse, Toolkit, and Cook- book.” Vol. 4 of Handbook of International Economics, Chapter 3. Elsevier. Helpman, Elhanan, Marc J. Melitz, and Stephen Ross Yeaple. 2004. “Export Versus FDI with Heterogeneous Firms.” American Economic Review, 94(1): 300–316. Pierce, Justin R., and Peter K. Schott. 2012. “Concording U.S. Harmonized System Codes over Time.” Journal of Official Statistics, 28(1): 53–68. Santos Silva, Joäo, and Silvana Tenreyro. 2006. “The Log of Gravity.” The Review of Eco- nomics and Statistics, 88(4): 641–658. Sato, Masahiro, Ali Kharrazi, Hirofumi Nakayama, Steven Kraines, and Masaru Yarime. 2015. “The Resilience of Embodied Energy Networks: A Critical Dimension for Sustain- able Development Goals (SDGs).” Global Environmental Research, 19(2): 187–198. Sato, Misato, and Antoine Dechezleprêtre. 2015. “Asymmetric Industrial Energy Prices and International Trade.” Energy Economics, 52(S1): S130–S141. 25 Figure 1: Share of Imported Intermediate Goods Imported or Exported, 2010 26 OECD Non-OECD Australia Brazil Austria Bulgaria Belgium-Luxembourg Colombia Canada Croatia Czech Republic India Denmark Indonesia Estonia Lithuania Finland Malaysia France Philippines Germany Russian Federation Greece Saudi Arabia Hungary South Africa Ireland Thailand Italy Vietnam Japan Korea, Rep. Mexico Netherlands New Zealand Poland Portugal Slovak Republic Slovenia Spain Sweden Switzerland Turkey United Kingdom United States Table 1: Countries in the Dataset 27 Country Data sources OECD countries IEA Non-OECD countries: Brazil IEA Bulgaria IEA Colombia IEA Croatia IEA India Various sources, including Ministry of Petroleum and Natual Gas and Planning Commission Indonesia Ministry of Energy and Mineral Resources Lithuania IEA Malaysia Tenaga Nasional (TNB), Petronas Philippines Meralco, FirstGen Russian Federation IEA Saudi Arabia ERRA South Africa IEA Thailand Energy Policy and Planning Office (Ministry of En- ergy) Vietnam Various sources including media reports Table 2: Sources of Energy Prices Data 28 Sectors in IEA Sectors in ICIO ISIC Rev 3.1 Non-ferrous metals Basic metals 27 (plus iron and steel) Chemicals and petrochemicals Chemicals and chemical products 24 Non-metallic minerals Other non-metallic mineral products 26 Transport equipment Motor vehicles, trailers, and semi-trailers 34 Other transport equipment 35 Machinery Fabricated metal products 28 Machinery and equipment, nec 29 Computer, electronic and optical equipment 30,32,33 Electrical machinery and apparatus, nec 31 Food and tobacco Food products, beverages and tobacco 15,16 Paper, pulp and print Pulp, paper, paper products, printing and 21,22 publishing Wood and wood products Wood and products of wood and cork 20 Textiles and leather Textiles, textile products, leather and 17,18,19 footwear Non-Specified Rubber and plastics products 25 Manufacturing nec; recycling 36,37 Table 3: List of Sectors in Our Dataset Obs. Mean Std. Dev. Min Max Value of exportsa 104711 0.52 2.68 0 120.99 Electricity prices, per MWh 104711 108.22 54.28 25.78 550.80 Electicity consumptionb 104711 10.42 23.53 0 263.12 Natural gas prices, per MWh 104711 29.33 14.85 4.07 79.20 Natural gas consumptionb 104711 10.92 39.51 0 881.10 Average annual wage 104711 27142 27273 697 1011431 Number of employees 104711 295991 532874 300 5995462 Value addeda 104711 21.63 59.66 0.0093 849.31 Sector outputa 104711 60.14 133.38 0.047 1577.3 Note: All prices or values are in constant 2010 US dollars. Table 4: Summary statistics a In billions of US dollars. b In millions of megawatt-hours. 29 Food Textiles Wood Paper Chemicals Minerals Metals Machinery Transport Others Food 69.5 0.5 0.6 7.7 4.2 2.1 0.8 6.4 0.8 7.4 Textiles 4.1 72.0 0.3 1.9 13.2 0.4 0.3 2.9 0.5 4.5 Wood 0.9 3.2 68.3 2.9 9.1 1.9 2.2 6.3 0.9 4.2 Paper 1.2 2.2 4.1 64.3 11.6 0.4 0.9 7.2 1.1 7.0 Chemicals 4.6 0.9 0.4 3.8 74.4 1.7 1.6 5.7 0.7 6.3 Minerals 1.2 1.4 2.5 5.3 13.8 49.3 5.7 12.3 1.6 7.0 30 Metals 0.5 0.6 2.7 1.0 3.6 2.0 67.4 17.0 1.4 3.8 Machinery 0.4 0.5 0.5 1.4 4.1 2.0 24.6 57.3 2.1 7.0 Transport 0.2 1.0 0.5 0.7 2.3 1.1 12.2 24.1 49.6 8.2 Others 0.9 5.0 3.8 2.9 34.5 1.3 5.9 10.8 1.8 33.0 Note: The table shows, for each row sector, the percentage of inputs that comes from each column sector. All figures are based on weighted average transactions in the 2010 OECD ICIO tables. Table 5: Input-Output Relationships Electricity Natural gas Labor Sector Direct Aggregate Direct Aggregate Direct Aggregate Food 0.040 0.054 0.016 0.019 0.354 0.371 Textile 0.043 0.056 0.011 0.016 0.509 0.468 Wood 0.065 0.073 0.007 0.013 0.448 0.432 Paper 0.101 0.094 0.022 0.022 0.405 0.403 Chemicals 0.112 0.103 0.040 0.036 0.309 0.337 Minerals 0.089 0.093 0.052 0.039 0.382 0.390 Metal 0.263 0.194 0.053 0.040 0.392 0.402 Machinery 0.021 0.093 0.004 0.020 0.440 0.422 Transport 0.022 0.062 0.005 0.014 0.438 0.430 Others 0.106 0.101 0.029 0.029 0.449 0.396 Note: “Aggregate" columns show the factor intensities after taking into account both direct and indirect costs. Table 6: Mean Factor Intensities by Sector 31 (1) (2) (3) (4) Variable Baseline IO Baseline IO Exporter electricity cost -0.1073∗∗∗ -0.0745∗∗∗ -0.1250∗∗∗ -0.0759∗∗∗ (0.0241) (0.0185) (0.0271) (0.0199) Importer electricity cost 0.1358∗∗∗ 0.0970∗∗∗ 0.1613∗∗∗ 0.1010∗∗∗ (0.0259) (0.0156) (0.0286) (0.0175) Exporter gas cost -0.0241 -0.0439∗∗∗ -0.0633∗ -0.0441∗∗∗ (0.0225) (0.0123) (0.0355) (0.0129) Importer gas cost 0.0116 0.0430∗∗∗ 0.0573∗ 0.0428∗∗∗ (0.0188) (0.0068) (0.0308) (0.0069) Exporter labor cost -0.1621∗∗∗ -0.2080∗∗∗ -0.1697∗∗∗ -0.2219∗∗∗ (0.0438) (0.0416) (0.0454) (0.0451) Importer labor cost 0.1495∗∗∗ 0.2168∗∗∗ 0.1565∗∗∗ 0.2211∗∗∗ (0.0408) (0.0313) (0.0421) (0.0346) Sector output 1.0168∗∗∗ 1.0286∗∗∗ 1.0189∗∗∗ 1.0308∗∗∗ (0.1118) (0.1054) (0.1130) (0.1077) Importer-exporter pair FE Yes Yes Yes Yes Exporter-year FE Yes Yes Yes Yes Importer-year FE Yes Yes Yes Yes Sector-year FE Yes Yes Yes Yes Observations 104711 104711 94777 94777 R2 0.8514 0.8467 0.8531 0.8461 Note: Standard errors are in parentheses and are clustered at the importer-exporter pair level. The dependent variable is the real value of exports. All models are estimated in Poisson pseudo-maximum likelihood (PPML) and include the set of importer-exporter pair fixed effects, importer-year fixed effects, exporter-year fixed effects, and sector-year fixed effects. Columns (2) and (4) apply the input-output (IO) extended model to compute the embodied costs of all three inputs (electricity, natural gas, and labor). Columns (3) and (4) exclude the “non-specified" sector. ∗ Significance at the 10 percent level. ∗∗ Significance at the 5 percent level. ∗∗∗ Significance at the 1 percent level. Table 7: Main Regression Results 32 (1) (2) (3) (4) Variable Baseline IO Baseline IO Exporter electricity cost -0.0820∗∗∗ -0.0810∗∗∗ -0.0983∗∗∗ -0.0879∗∗∗ (0.0215) (0.0215) (0.0275) (0.0228) Importer electricity cost 0.0890∗∗∗ 0.0669∗∗ 0.1052∗∗∗ 0.0698∗∗ (0.0223) (0.0311) (0.0278) (0.0315) Exporter gas cost -0.0017 -0.0321∗∗∗ -0.0116 -0.0334∗∗∗ (0.0138) (0.0119) (0.0213) (0.0122) Importer gas cost -0.0011 0.0346∗∗∗ 0.0097 0.0354∗∗∗ (0.0126) (0.0126) (0.0200) (0.0124) Exporter labor cost -0.2988∗∗∗ -0.1967∗∗∗ -0.3111∗∗∗ -0.2131∗∗∗ (0.0506) (0.0464) (0.0518) (0.0482) Importer labor cost 0.2844∗∗∗ 0.1566∗∗ 0.2943∗∗∗ 0.1598∗∗ (0.0451) (0.0661) (0.0467) (0.0676) Importer-exporter pair FE Yes Yes Yes Yes Exporter-year FE Yes Yes Yes Yes Importer-year FE Yes Yes Yes Yes Sector-year FE Yes Yes Yes Yes Observations 104711 104711 94777 94777 R2 0.6404 0.6279 0.6356 0.6210 Note: Standard errors are in parentheses and are clustered at the importer-exporter pair level. The dependent variable is the ratio of the value of exports to the value of sec- tor output. All models are estimated in Poisson pseudo-maximum likelihood (PPML) and include the set of importer-exporter pair fixed effects, importer-year fixed effects, exporter-year fixed effects, and sector-year fixed effects. Columns (2) and (4) apply the input-output (IO) extended model to compute the embodied costs of all three inputs (electricity, natural gas, and labor). Columns (3) and (4) exclude the “non-specified" sec- tor. ∗ Significance at the 10 percent level. ∗∗ Significance at the 5 percent level. ∗∗∗ Significance at the 1 percent level. Table 8: Results with Exports/Sector Output as the Dependent Variable 33 (1) (2) (3) (4) Variable Baseline IO Baseline IO Exporter electricity cost -0.1291∗∗∗ -0.0813∗∗∗ -0.1514∗∗∗ -0.0835∗∗∗ (0.0264) (0.0208) (0.0304) (0.0225) Importer electricity cost 0.1567∗∗∗ 0.0887∗∗∗ 0.1861∗∗∗ 0.0925∗∗∗ (0.0277) (0.0153) (0.0314) (0.0169) Exporter gas cost -0.0423 -0.0460∗∗∗ -0.0752∗ -0.0459∗∗∗ (0.0297) (0.0136) (0.0392) (0.0143) Importer gas cost 0.0271 0.0411∗∗∗ 0.0670∗ 0.0407∗∗∗ (0.0252) (0.0080) (0.0344) (0.0081) Exporter labor cost -0.1411∗∗∗ -0.2081∗∗∗ -0.1471∗∗∗ -0.2203∗∗∗ (0.0418) (0.0466) (0.0430) (0.0504) Importer labor cost 0.1391∗∗∗ 0.2042∗∗∗ 0.1446∗∗∗ 0.2078∗∗∗ (0.0413) (0.0337) (0.0424) (0.0364) Sector output 1.0224∗∗∗ 1.0271∗∗∗ 1.0244∗∗∗ 1.0290∗∗∗ (0.0979) (0.0932) (0.0983) (0.0952) Importer-exporter pair FE Yes Yes Yes Yes Exporter-year FE Yes Yes Yes Yes Importer-year FE Yes Yes Yes Yes Sector-year FE Yes Yes Yes Yes Observations 104711 104711 94777 94777 R2 0.8682 0.8636 0.8696 0.8632 Note: Standard errors are in parentheses and are clustered at the importer-exporter pair level. The dependent variable is the real value of exports. All control variables are one-year lags. All models are estimated in Poisson pseudo-maximum likelihood (PPML) and include the set of importer-exporter pair fixed effects, importer-year fixed effects, exporter-year fixed effects, and sector-year fixed effects. Columns (2) and (4) apply the input-output (IO) extended model to compute the embodied costs of all three inputs (electricity, natural gas, and labor). Columns (3) and (4) exclude the “non-specified" sector. ∗ Significance at the 10 percent level. ∗∗ Significance at the 5 percent level. ∗∗∗ Significance at the 1 percent level. Table 9: Results with Lagged Controls 34 (1) (2) (3) (4) Variable Baseline IO Baseline IO Exporter electricity cost -0.0883∗∗∗ -0.0328∗ -0.0965∗∗∗ -0.0250 (0.0204) (0.0196) (0.0221) (0.0190) Importer electricity cost 0.1108∗∗∗ 0.1198∗∗∗ 0.1240∗∗∗ 0.1166∗∗∗ (0.0216) (0.0153) (0.0232) (0.0146) Exporter gas cost -0.0234 -0.0221∗∗ -0.0528∗∗ -0.0181 (0.0167) (0.0111) (0.0235) (0.0113) Importer gas cost 0.0141 0.0514∗∗∗ 0.0480∗∗ 0.0488∗∗∗ (0.0143) (0.0063) (0.0208) (0.0065) Exporter labor cost -0.1945∗∗∗ -0.0409 -0.1885∗∗∗ -0.0210 (0.0394) (0.0592) (0.0396) (0.0621) Importer labor cost 0.1567∗∗∗ 0.4205∗∗∗ 0.1487∗∗∗ 0.4362∗∗∗ (0.0410) (0.0432) (0.0405) (0.0445) Sector output 0.8774∗∗∗ 0.8872∗∗∗ 0.8790∗∗∗ 0.8909∗∗∗ (0.0897) (0.0788) (0.0910) (0.0808) Importer-exporter pair FE Yes Yes Yes Yes Exporter-year FE Yes Yes Yes Yes Importer-year FE Yes Yes Yes Yes Sector-year FE Yes Yes Yes Yes Observations 153740 153740 134589 134589 R2 0.7497 0.7512 0.7502 0.7506 Note: Standard errors are in parentheses and are clustered at the importer-exporter pair level. The dependent variable is the real value of exports. All models are estimated in Poisson pseudo-maximum likelihood (PPML) and include the set of importer-exporter pair fixed effects, importer-year fixed effects, exporter-year fixed effects, and sector-year fixed effects. Columns (2) and (4) apply the input-output (IO) extended model to compute the embodied costs of all three inputs (electricity, natural gas, and labor). Columns (3) and (4) exclude the “non-specified" sector. ∗ Significance at the 10 percent level. ∗∗ Significance at the 5 percent level. ∗∗∗ Significance at the 1 percent level. Table 10: Results Using Data Aggregated to the ICIO Classification 35 (1) (2) (3) (4) Variable Baseline IO Baseline IO Exporter electricity cost -0.0722∗∗∗ -0.0464∗∗∗ -0.0811∗∗∗ -0.0397∗∗∗ (0.0163) (0.0139) (0.0179) (0.0138) Importer electricity cost 0.0905∗∗∗ 0.0974∗∗∗ 0.1040∗∗∗ 0.0970∗∗∗ (0.0171) (0.0110) (0.0187) (0.0110) Exporter gas cost -0.0168 -0.0256∗∗∗ -0.0411∗∗ -0.0220∗∗∗ (0.0131) (0.0075) (0.0183) (0.0077) Importer gas cost 0.0102 0.0395∗∗∗ 0.0387∗∗ 0.0383∗∗∗ (0.0112) (0.0054) (0.0160) (0.0054) Exporter labor cost -0.2457∗∗∗ -0.1518∗∗∗ -0.2418∗∗∗ -0.1395∗∗ (0.0470) (0.0548) (0.0480) (0.0597) Importer labor cost 0.1860∗∗∗ 0.3863∗∗∗ 0.1790∗∗∗ 0.4082∗∗∗ (0.0458) (0.0382) (0.0464) (0.0404) Sector output 0.8366∗∗∗ 0.8691∗∗∗ 0.8390∗∗∗ 0.8737∗∗∗ (0.0653) (0.0638) (0.0661) (0.0648) Importer-exporter pair FE Yes Yes Yes Yes Exporter-year FE Yes Yes Yes Yes Importer-year FE Yes Yes Yes Yes Sector-year FE Yes Yes Yes Yes Observations 513526 513526 475224 475224 R2 0.4448 0.4417 0.4420 0.4379 Note: Standard errors are in parentheses and are clustered at the importer-exporter pair level. The dependent variable is the real value of exports. All models are estimated in Poisson pseudo-maximum likelihood (PPML) and include the set of importer-exporter pair fixed effects, importer-year fixed effects, exporter-year fixed effects, and sector-year fixed effects. Columns (2) and (4) apply the input-output (IO) extended model to compute the embodied costs of all three inputs (electricity, natural gas, and labor). Columns (3) and (4) exclude the “non-specified" sector. The aggregation is based on ISIC Rev. 3.1. ∗ Significance at the 10 percent level. ∗∗ Significance at the 5 percent level. ∗∗∗ Significance at the 1 percent level. Table 11: Results Using Data Aggregated to the ISIC Three-Digit Classification 36 (1) (2) (3) (4) Variable Baseline IO Baseline IO Electricity cost gap -0.0945∗∗∗ -0.0859∗∗∗ -0.1116∗∗∗ -0.0889∗∗∗ (0.0273) (0.0131) (0.0349) (0.0147) Natural gas cost gap -0.0360∗ -0.0438∗∗∗ -0.0789∗∗ -0.0438∗∗∗ (0.0210) (0.0072) (0.0336) (0.0076) Labor cost gap -0.1517∗∗∗ -0.2146∗∗∗ -0.1521∗∗∗ -0.2244∗∗∗ (0.0411) (0.0287) (0.0418) (0.0315) Sector output 1.0259∗∗∗ 1.0273∗∗∗ 1.0265∗∗∗ 1.0296∗∗∗ (0.1059) (0.1059) (0.1062) (0.1080) Importer-exporter pair FE Yes Yes Yes Yes Exporter-year FE Yes Yes Yes Yes Importer-year FE Yes Yes Yes Yes Sector-year FE Yes Yes Yes Yes Observations 104711 104711 94777 94777 R2 0.8497 0.8468 0.8505 0.8465 Note: Standard errors are in parentheses and are clustered at the importer-exporter pair level. The dependent variable is the real value of exports. All models are estimated in Poisson pseudo-maximum likelihood (PPML) and include the set of importer-exporter pair fixed effects, importer-year fixed effects, exporter-year fixed effects, and sector-year fixed effects. ‘Cost gap’ is the exporter cost (in log) minus the importer cost (in log). Columns (2) and (4) apply the input-output (IO) extended model to compute the em- bodied costs of all three inputs (electricity, natural gas, and labor). Columns (3) and (4) exclude the “non-specified" sector. ∗ Significance at the 10 percent level. ∗∗ Significance at the 5 percent level. ∗∗∗ Significance at the 1 percent level. Table 12: Results with Cost Gaps 37 (1) (2) (3) (4) Variable Baseline IO Baseline IO Electricity cost gap -0.0115 -0.0939∗∗∗ -0.0301 -0.0685∗∗∗ (0.0230) (0.0198) (0.0278) (0.0211) Natural gas cost gap -0.0292* -0.0435∗ -0.0789∗∗ -0.0266∗∗∗ (0.0152) (0.0173) (0.0089) (0.0100) Labor cost gap 0.0030 -0.2755∗∗∗ 0.0181 -0.0876∗ (0.0593) (0.0545) (0.0589) (0.0460) Importer-exporter pair FE Yes Yes Yes Yes Exporter-sector-year FE Yes Yes Yes Yes Importer-sector-year FE Yes Yes Yes Yes Observations 104711 104711 94777 94777 R2 0.9798 0.9798 0.9801 0.9800 Note: Standard errors are in parentheses and are clustered at the importer-exporter pair level. The dependent variable is the real value of exports. All models are estimated in Poisson pseudo-maximum likelihood (PPML) and include the set of importer-exporter pair fixed effects, exporter-sector-year fixed effects, and importer-sector-year fixed effects. ‘Cost gap’ is the exporter cost (in log) minus the importer cost (in log). Columns (2) and (4) apply the input-output (IO) extended model to compute the embodied costs of all three inputs (electricity, natural gas, and labor). Columns (3) and (4) exclude the “non-specified" sector. ∗ Significance at the 10 percent level. ∗∗ Significance at the 5 percent level. ∗∗∗ Significance at the 1 percent level. Table 13: Results with a Different Set of Fixed Effects 38 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Variable Baseline IO Baseline IO Baseline IO Baseline IO Baseline IO Elec cost gap -0.0877∗∗∗ -0.0801∗∗∗ -0.1025∗∗∗ -0.0823∗∗∗ -0.1066∗∗∗ -0.0864∗∗∗ -0.1073∗∗∗ -0.0853∗∗∗ -0.1026∗∗∗ -0.0884∗∗∗ (0.0237) (0.0130) (0.0308) (0.0121) (0.0312) (0.0141) (0.0388) (0.0134) (0.0310) (0.0131) NG cost gap -0.0405∗∗ -0.0421∗∗∗ -0.0147 -0.0446∗∗∗ -0.0314 -0.0438∗∗∗ -0.0237 -0.0459∗∗∗ -0.0332 -0.0459∗∗∗ (0.0192) (0.0070) (0.0142) (0.0071) (0.0203) (0.0085) (0.0181) (0.0089) (0.0203) (0.0084) Labor cost gap -0.1502∗∗∗ -0.2042∗∗∗ -0.1381∗∗∗ -0.2084∗∗∗ -0.1490∗∗∗ -0.2135∗∗∗ -0.1138∗∗ -0.2086∗∗∗ -0.1734∗∗∗ -0.2145∗∗∗ (0.0416) (0.0294) (0.0410) (0.0290) (0.0438) (0.0304) (0.0467) (0.0311) (0.0477) (0.0304) Sector output 1.0413∗∗∗ 1.0417∗∗∗ 1.0286∗∗∗ 1.0279∗∗∗ 1.0260∗∗∗ 1.0280∗∗∗ 1.0288∗∗∗ 1.0267∗∗∗ 1.0233∗∗∗ 1.0249∗∗∗ (0.1100) (0.1098) (0.1025) (0.1048) (0.1066) (0.1065) (0.1046) (0.1055) (0.1053) (0.1057) Elec cost gap × interaction -0.0092 -0.0083∗∗ 0.0530 -0.0176 0.0569 0.0009 0.0346 0.0001 -0.0292 -0.0157∗∗ (0.0254) (0.0039) (0.0399) (0.0154) (0.0352) (0.0157) (0.0425) (0.0136) (0.0299) (0.0061) 39 NG cost gap × interaction 0.0040 -0.0020 -0.1328∗∗ 0.0002 -0.0634 -0.0019 -0.0603 0.0036 0.0398∗∗∗ -0.0055 (0.0097) (0.0038) (0.0527) (0.0097) (0.0623) (0.0126) (0.0419) (0.0089) (0.0138) (0.0047) Labor cost gap × interaction -0.0394∗∗ -0.0067 -0.0652∗∗ -0.0308 -0.0542 -0.0158 -0.0976∗∗ -0.0092 -0.0346 -0.0171 (0.0182) (0.0073) (0.0309) (0.0230) (0.0581) (0.0347) (0.0455) (0.0276) (0.0267) (0.0118) Interactions Relative distance Contiguity Euro zone Trading blocs Absolute distance Note: Standard errors are in parentheses and are clustered at the importer-exporter pair level. The dependent variable is the real value of exports. All models are estimated in Poisson pseudo-maximum likelihood (PPML) and include the set of importer-exporter pair fixed effects, importer-year fixed effects, exporter-year fixed effects, and sector-year fixed effects. ‘Cost gap’ is the exporter cost (in log) minus the importer cost (in log). Even-numbered columns apply the input-output (IO) extended model to compute the embodied costs of all three inputs (electricity, natural gas, and labor). ∗ Significance at the 10 percent level. ∗∗ Significance at the 5 percent level. ∗∗∗ Significance at the 1 percent level. Table 14: Results with More Interactions India Bulgaria Baseline IO Baseline IO Exports Net exports Exports Net exports Exports Net exports Exports Net exports Overall impact 3.42% 5.45% 4.33% 9.52% 4.37% 7.04% 2.96% 5.65% Wood 5.23% 6.54% 4.50% 10.47% 3.18% 6.03% 2.43% 5.69% Chemicals 2.78% 5.61% 3.75% 8.77% 7.72% 12.26% 2.02% 4.72% Non-metallic minerals 4.51% 7.70% 5.01% 11.69% 3.47% 8.09% 2.61% 6.10% Basic metals 8.29% 16.21% 7.47% 17.46% 11.63% 23.89% 7.27% 16.95% 40 Food products 2.23% 2.62% 3.63% 8.44% 1.76% 3.52% 1.71% 3.99% Textiles 4.47% 5.83% 4.47% 10.44% 1.12% 2.94% 0.82% 1.92% Pulp and paper 3.36% 8.64% 3.77% 8.81% 2.77% 7.93% 2.32% 5.42% Machinery 0.70% 1.58% 3.29% 7.70% 1.03% 1.94% 1.91% 4.46% Transport 1.30% 2.10% 3.29% 7.72% 1.19% 2.24% 1.19% 2.77% Note: The table lists predicted changes in exports and net exports in 2010 for India and Bulgaria. The simulation is done using results in columns (3) and (4) in Table 7 (i.e. exclduing the “non-specified" sector) from the baseline and input-output extended models respectively, based on observed input-output relationships and actual trading volumes in 2010. Table 15: Simulation Results: Effect on Manufacturing Exports in India and Bulgaria from a Unilateral 15 Percent Decrease in Electricity Price EU Non-EU Baseline IO Baseline IO Exports Net exports Exports Net exports Exports Net exports Exports Net exports Overall impact -0.19% -0.71% -0.53% -1.91% 0.33% 1.02% 0.58% 1.03% Wood -0.06% -0.46% -0.40% -1.37% 0.09% 0.53% 0.37% 0.64% Chemicals -0.45% -1.11% -0.86% -3.18% 0.31% 1.85% 1.09% 2.11% Non-metallic minerals -0.51% -1.25% -0.75% -2.77% 0.34% 2.18% 0.70% 1.85% Basic metals -0.40% -2.72% -1.31% -4.44% 1.37% 3.20% 1.27% 2.02% Food products -0.05% -0.40% -0.71% -2.04% 0.19% 0.71% 0.57% 0.98% 41 Textiles -0.07% -0.56% -0.43% -1.40% 0.61% 1.14% 0.66% 0.99% Pulp and paper -0.20% -1.21% -0.71% -2.43% 0.54% 2.20% 0.72% 1.64% Machinery -0.09% -0.27% -0.28% -1.07% 0.12% 0.35% 0.29% 0.51% Transport -0.09% -0.23% -0.25% -0.91% 0.03% 0.29% 0.24% 0.43% Note: The simulation assumes a 10 percent increase in both electricity and natural gas prices across the European Union. The simulation is done using results in columns (3) and (4) in Table 7 (i.e. exclduing the “non-specified" sector) from the baseline and input-output extended models respectively, based on observed input-output relationships and actual trading volumes in 2010. Non-EU averages are based on the non-EU countries in the sample. Table 16: Simulation Results: Effect on EU and Non-EU Manufacturing Exports from a 10 Percent Increase in Energy Prices in the European Union