Policy Research Working Paper 11069 Energy Prices, Energy Intensity, and Firm Performance Reyes Aterido Mariana Iootty Martin Melecky Finance, Competitiveness and Innovation Global Department February 2025 Policy Research Working Paper 11069 Abstract This paper estimates the effect of electricity prices on firm similar firms in energy-non-intensive sectors. In parallel, performance, focusing on firm productivity, sales, and energy-intensive firms may increase sales and productiv- employment. Using the World Bank Business Pulse Survey ity but this result is robust to all alternative specifications. data for a sample of 24 emerging markets and developing Firms may increase sales while reducing employment after economies during 2019–23, the paper estimates the average energy price hikes, by adopting energy-efficient technolo- effect and the heterogeneous effects across industries of vary- gies and by passing through costs to consumers in inelastic ing energy intensity and firms that implemented (or did not markets while reducing employment in energy-intensive implement) energy efficiency measures (self-reported in the sectors due to cost pressures. These results highlight the Business Pulse Survey). The findings show that increasing adoption of energy efficiency measures by firms as an electricity prices by 1 percent reduces employment at firms important employment protection policy action to cope in energy-intensive industries that did not adopt energy with future volatility in energy (electricity) prices. efficiency measures by about 1.5 percent, compared with This paper is a product of the Finance, Competitiveness and Innovation Global Department. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at mmelecky@worldbank.org, miootty@worldbank.org, and raterido@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, Energy Intensity, and Firm Performance Reyes Aterido, Mariana Iootty, and Martin Melecky World Bank Keywords: World Bank Business Pulse Survey Data, Firm-level Panel Data, Energy Prices, Energy Intensity of Industries and Firms, Firm Performance, Productivity, Sales, Employment. JEL Classifica ion: D22, L25, Q41, Q43, O13. 1. Introduction The link between rising energy costs and firm performance has garnered increasing attention in academic and policy discussions due to the broader adoption of carbon pricing mechanisms and the gradual elimination of energy subsidies. As these policy changes elevate nonrenewable energy prices—a crucial input in the production processes of many industries—firms may experience notable performance implications. Higher energy costs generally affect profitability and operational efficiency by increasing production expenses, particularly in energy-intensive sectors. Firms with greater energy dependence often face heightened stock market volatility due to increased uncertainty and lower expected cash flows (Sadorsky 1999; Henriques & Sadorsky 2008). Moreover, rising energy costs may constrain corporate investment strategies, leading to reduced capital expenditures, especially in industries reliant on energy inputs (Kilian 2008; Bloom 2009). The increased accessibility of firm-level data has opened new avenues for empirical studies of the complex link between energy price fluctuations and firm performance. Cali et al. (2022) show that, in Indonesia and Mexico, increases in electricity prices harm manufacturing plants’ performance. However, fuel price hikes result in higher productivity and profits of manufacturing plants. Fuel prices incentivize plants to replace inefficient fuel-powered with more productive electricity-powered capital equipment. Their results help to re-evaluate the policy trade-off between reducing carbon emissions and improving economic performance, particularly in countries with large fuel subsidies, such as Indonesia and Mexico. Cali et al. (2023) studied firms from 12 sectors and 11 middle-income countries during 2002–2013, using World Bank Enterprise Survey data. The study did not consistently find that higher energy prices negatively affect economic performance; they may even enhance it in some cases. Firms might be able to offset increased energy costs through innovation or new market strategies. The impact of energy prices on firms is a multifaceted issue, depending on several characteristics, including energy intensity, firm size, ownership type, and previous experience with electricity outages—which made companies less sensitive to price increases, likely due to their familiarity with managing energy and input shortages. The dynamic effects of energy price shocks on firm performance have also been explored. Andre et al. (2023) analyze firm-level data from 21 OECD countries (1995–2020) and show that energy price shocks impact firm productivity through immediate cost increases and longer-term adjustments. The study highlights sectoral and firm-level heterogeneity, finding that energy price hikes initially reduce productivity, particularly in energy-intensive sectors and firms under financial constraints. However, medium-term productivity improvements often follow as firms adapt, typically through increased investment. These findings align with the "pollution haven" hypothesis, which suggests that higher energy costs erode competitiveness and may drive firms to relocate (Copeland & Taylor 2004), and the Porter hypothesis, which posits that higher carbon prices can enhance productivity by incentivizing efficiency and innovation (Porter & van der Linde 1995). This dual perspective underscores the nuanced impact of energy price shocks on firm outcomes, blending short-term losses with potential long-term gains. The recent energy price surge, triggered by the strong economic recovery following COVID-19 and the Russian invasion of Ukraine, has sparked renewed interest in the subject. Battistini et al. (2022) focus on EU economies during 2020-2022 and highlight a few interesting aspects. Energy price shocks disproportionately affect firms based on their energy dependence and hedging strategies. Energy- intensive sectors are particularly hard-hit, facing greater financial stress and profitability challenges due to heightened costs and supply chain disruptions. Spiking energy prices have triggered a significant 2 worsening of the energy terms of trade in the euro area. This has impacted companies by reducing their purchasing power and increasing operational costs, affecting their profitability and investment decisions. Ari et al. (2022) employed the IMF-World Bank CPAT model and found clear evidence of a regressive impact on European households’ finances. However, the effects on European businesses are less clear-cut, with mixed results regarding the loss of competitiveness. The increase in energy costs for energy-dependent and trade-sensitive industries within the European Union may not be disproportionately higher than for non-EU countries due to variations in natural gas reliance or the use of other energy products across different nations. When evaluating the effects of fluctuating energy prices, the role of fossil fuel subsidies must be carefully considered. Accurately pricing fossil fuels is essential for the efficient allocation of an economy's scarce resources and investment across sectors and activities. An efficient price incorporates both the supply costs and the environmental externalities of fuel consumption (Coady et al. 2019). However, substantial subsidies or underpricing distort market signals, leading to the overconsumption of fossil fuels, exacerbating global warming, and intensifying local environmental degradation. For instance, Black et al. (2023) report that global fossil fuel subsidies reached an unprecedented $7 trillion in 2023, as governments sought to shield consumers and businesses from energy price spikes triggered by the Russian Federation’s invasion of Ukraine and the post-pandemic economic recovery. These subsidies, while aimed at providing short-term relief, may have inadvertently delayed firms' and households' consideration of environmental costs in consumption and investment decisions. As a result, the adoption of energy-efficient technologies and sustainable practices has likely been postponed, undermining long-term climate goals (Bovenberg & de Mooij 1994; Parry et al. 2021). Against this backdrop, this paper contributes to the existing body of research by focusing on the impact of country-specific energy price fluctuations during 2019-2023 on firms in emerging economies using the World Bank Business Pulse Survey—a dataset not used yet for this type of analysis in the literature— together with the CPRS classification of Battiston et al. (2022) to gauge the risk exposure of sectors to (the intensity of treatment by) the energy price fluctuations. Furthermore, the paper contributes to the literature by shedding light on the specific context of emerging economies, employing rapid survey data to grasp the nuances of business cycles, using the quarterly country-specific electricity price tariffs for businesses, matching the latter with the month in which a firm was surveyed, and highlighting the differentiated effects of sectoral energy intensity in conjunction with firm-level energy efficiency measures as indicators of a firm's exposure to energy price fluctuations and preparedness to manage energy price shocks. Using a different data set with larger country coverage, our baseline findings confirm prior evidence from Cali et al. (2022, 2023) that electricity price increases, on average, lead to higher productivity and sales for firms. However, the magnitude of this positive effect is significantly reduced for firms that have not implemented energy efficiency measures, suggesting that such initiatives are critical for firms to adapt to evolving energy market trends. In addition, when the baseline results are subjected to a battery of robustness tests the results do not survive in all alternative specifications. By contrast, we find that rising electricity prices exert a negative effect on employment, particularly in energy-intensive industries where energy efficiency measures are absent. This result highlights the vulnerability of labor markets in sectors heavily reliant on energy (electricity) inputs, especially in developing economies where the adoption of efficiency-enhancing technologies often lags. 3 These findings align with evidence from Aldy and Pizer (2015), who show that energy-intensive firms are disproportionately affected by energy price increases due to their limited flexibility in reducing energy consumption. Similarly, Sato et al. (2019) emphasize the employment risks in sectors with high energy dependency, particularly in regions with limited policy support for energy efficiency. Our study’s integration of industry-level energy intensity with firm-level adaptations provides a novel contribution by offering deeper insights into how electricity price changes propagate through different economic layers. Compared to earlier studies, such as Martin et al. (2014), which primarily focused on advanced economies or treated energy price impacts in aggregate terms, this paper uniquely highlights the heterogeneity of these effects across firms in developing countries, where structural constraints and varying levels of technological adoption exacerbate disparities in energy price impacts. Our findings underline the importance of energy efficiency measures not only as a tool for enhancing productivity and sales, but also as a critical buffer against adverse employment effects associated with rising electricity prices. This aligns with evidence from Bloom et al. (2010), who emphasize that firm-level innovations, including energy efficiency improvements, are key to maintaining competitiveness during economic transitions. Our findings also complement those by Newell et al. (2021), which highlight the role of energy efficiency in mitigating the labor market disruptions associated with volatile energy prices, particularly in energy-intensive sectors. By focusing on developing economies, this study expands the literature on energy price shocks—which has predominantly centered on advanced economies—and addresses critical gaps in understanding the shocks’ differentiated effects. Consistent with Popp (2019), our results suggest that fostering widespread adoption of energy efficiency practices is essential for mitigating the employment risks of transitioning to greener energy systems. Particularly for energy-intensive industries in developing countries, these measures are crucial for safeguarding jobs and ensuring firms remain competitive amid fluctuating electricity prices and shifting global energy markets. The rest of the paper is organized as follows. Section 2 describes the employed data. Section 3 explains the estimation methodology and identification. Section 4 discusses the estimation results and their robustness checks. Section 5 concludes. 2. Data Our analysis uses data from the Business Pulse Surveys (BPS) developed by the World Bank to monitor the impact of the COVID-19 pandemic on the private sector. The questionnaire collects information on several dimensions of firm performance, spanning from basic economic indicators—such as the operating status of the business, year of establishment, sales, and employment—as well as firm-specific practices such as managerial practices, technology adoption, and implementation of energy-efficient measures. Firms were surveyed from April 2020 (about end-2019 performance) to June 2023 in several waves, providing a unique perspective on the private sector's response to the pandemic and the subsequent economic disturbances caused by the fast post-COVID-19 recovery, Russia’s invasion of Ukraine, and the associated energy price surges and fluctuations. Our sample comprises an unbalanced panel of 63,716 observations drawn from 24 countries across four regions (Eastern and Central Europe, Latin America and the Caribbean, South Asia and the Pacific, and Sub- 4 Saharan Africa). 1 Due to data availability issues, the analysis is limited to 16 countries. This reduction stems primarily from misreporting in the sales variable but is also due to incomplete data for control variables, and electricity prices; the latter is, for example, not available for Tajikistan. Furthermore, self-reported energy efficiency measures are not available in all countries. Consequently, regressions that include energy efficiency measures are based on data from 13 countries. The sample covers micro, small, medium, and large businesses across all main sectors (agriculture, manufacturing, retail, and other services, including construction). 2 Table A1 (a-c) in the Annex describes the sample distribution across years, sectors, and firm sizes at the country level. This breakdown is instrumental for categorizing sectors by their energy usage intensity. The sectors are meticulously disaggregated to the most granular level the sampling methodology allows. Within the sample, agriculture accounts for 6 percent of the firms, manufacturing for 25 percent, wholesale and retail for 24 percent, and a diverse array of other services for 45 percent. Within the service sector, food services and accommodation represent 23 percent, transportation and information and communication technology (ICT) account for 13 percent, and construction for 12 percent. The remaining firms are engaged in financial activities, real estate, education, health, and other sectors. Table A2 in the Annex shows the summary statistics of the main variables used in the analysis. The dependent variables used in the estimations are employment, sales, and sales per worker as a measure of labor productivity. 3,4 The median firm has sales of $12,312, 6 workers, 5 with a labor productivity of $1,950 per worker, and has been operating for 15 years. Figure 1 depicts the trends of our outcome variables, labor productivity, sales, and employment. To address the effects of energy price shocks on firm performance, our paper carefully manages four critical aspects of the data. The first aspect involves classifying sectors based on their energy intensity following the Climate Policy Relevant Sectors (CPRS) framework introduced by Battiston et al. (2017). The CPRS taxonomy identifies energy-intensive sectors through a multifaceted lens, considering (i) the emissions produced by each sector's economic activities, (ii) the sector's contribution to the Greenhouse Gas (GHG) emissions chain, (iii) the sector's engagement in specific policy processes, including its lobbying capacity; and (iv) the transition risk associated with the sector, which inversely relates to the level of fuel substitutability. A sector is deemed energy-intensive if its activities result in emissions, utilizes fuel (or a mix of fuels) as a primary input, and has a low potential for substituting these fuels. The sector classification by energy usage is systematically outlined in Table A3 in the Annex. 1 For each country, the sample frame was based on the statistical data from the National Statistical Committees of each country at the time of the survey's first implementation. For the later survey waves, the list of companies was updated using lists of business associations and internal lists of entrepreneurs from the survey firm. 2 It is also worth acknowledging that firm weights are unavailable, so the sample is not representative. However, the estimations control for firm characteristics, muting to some extent the composition effects. 3 Sales per worker is a widely used proxy for labor productivity, offering a practical measure of output per unit of labor input, especially when value-added data are unavailable. Bloom et al. (2010) emphasize its effectiveness in capturing labor efficiency and management practices, while Syverson (2011) notes that it may reflect other factors, such as market power or capital intensity, which could distort comparisons. Despite these limitations, Bartelsman et al. (2004) highlight its importance in developing economies, where sales data often provide the most accessible alternative for productivity analysis. 4 Sales and sales per worker are winsorized at the 5 and 95 percentiles. 5 Few firms in the Philippines have more than 100,000 workers. 5 The second element is measuring energy price shocks. We use the commercial electricity rate shown in Figure 1, compiled quarterly by Global Petrol Prices for each country. For the countries included in the sample, electricity rates, on average, fell during 2020 and 2021, reaching their nadir in the last quarter of 2021. Subsequently, there was a recovery in the rates during 2022. The price fluctuation ranged from $0.03 to $0.38 per kWh. In the case of oil prices, the range was between $33.7 and $113. The third aspect refers to fossil fuel subsidies. We measure this aspect using the IMF data (Black et al., 2023), which provide, for 170 countries, the estimates of explicit subsidies for fossil fuels, i.e., undercharging for the supply cost of fossil fuels. All but five countries in the dataset, among them Tajikistan, have no (zero) petroleum subsidies. In the case of explicit subsidies for electricity, only three countries in the dataset do not have subsidies. Considering countries with explicit subsidies, the petroleum subsidies range from 0.00004 to 3.58 percent of GDP, while electricity subsidies range from 0.04 to 9.76 percent of GDP. The distributions of such subsidies overtime are depicted in Figures 3 (panels a and b). The fourth data aspect concerns firms' adoption of energy efficiency practices. This is a crucial factor because it can significantly influence a firm's resilience to energy price fluctuations. To account for these practices, we draw on data from the BPS, which features a module focused on firms' energy efficiency measures. The survey queries firms on implementing technologies or methods to improve energy usage efficiency. Those who have adopted energy efficiency solutions are prompted to detail the specific technologies or practices in use. These practices encompass questions about LEED certification for buildings, adopting efficient lighting systems, adherence to ISO 14001 or 50001 standards, or engagement in carbon trading schemes. About 38 percent of the surveyed firms acknowledge that they do not employ energy efficiency technologies or practices. Yet, this figure masks considerable cross-country heterogeneity. For example, in Armenia, 99 percent of the surveyed firms reported using at least one energy-saving approach. By contrast, in Paraguay, a striking 86 percent of firms indicated they do not engage in any such practices. For a more detailed breakdown, see Table A2 in the Annex. 3. Estimation methodology To assess the impact of energy shocks on firm performance, we utilize the BPS sample for 24 developing countries over the 2019-2023 period, covering large fluctuations in country-level electricity prices. We hypothesize that a firm's susceptibility to the energy price shock is contingent on the shock's magnitude at a specific time, the firm's energy requirements for its operations that we assume are sector-specific, and any energy efficiency measures or lack thereof that can make the firm-level energy efficiency deviate from its sector’s average. As explained in the data section, the variations in the shock intensity are quantified through the monthly electricity data for each country matched to firm surveys by the month of the firms’ interviews. The BPS data's unique advantage lies in recording the precise survey date for each firm, introducing firm-level variation in shock exposure across countries. This aspect is critical as the shock's magnitude is shown to vary from month to month (Figure 2). The energy usage intensity is determined based on CPRS classification; sectors such as manufacturing, construction, and transportation are categorized as energy- intensive, in contrast to agriculture, retail, and other services, which are classified as less energy-intensive using the Battiston et al. (2022) methodology which is predetermined at the global level and not influenced by firm or sector energy technology adoption at the country level. 6 We evaluate firm performance using three metrics: labor productivity (sales per worker) and, separately, sales and employment. Labor productivity is used due to the unavailability of value-added data (Bartelsman et al., 2004). 6 We hypothesize that firms with a higher degree of exposure to energy price increases may experience a reduction in sales, are likely to downsize and cut jobs, and could experience a drop in productivity because firm costs may rise faster than revenues, and some supply chain disruptions may reduce firms’ ability to make profit-maximizing decisions. Additionally, we hypothesize that firms operating in energy-intensive (CPRS) sectors will be disproportionately affected, with the effect of energy prices on their performance further amplified. The baseline regression to assess the effect of energy prices on firm performance is described in equation (1) where we focus on the energy price effect associated with fluctuations in electricity prices that we observe at the country level and can better identify and use the global prices of oil only as a control variable: ,,, = ∙ , + ∙ + ∙ , ∙ + ∙ , + + + , + + ,,, (1) where ,,, is the performance of firm i in sector j in country k at time t, is the price of electricity that varies over time at the country level, is the sector-specific classification according to energy usage intensity, , is a vector of firm control characteristics, including age and size clusters (micro, small, medium, large) 7 as well as the monthly global price of oil, are sector-specific dummies to account for sector-specific characteristics other than energy usage intensity, are country-level fixed effects to control for country characteristics that are invariant in the short term, such as institutions or market structures, , are country-time fixed effects to control for time-varying macroeconomic conditions common for all firms in a given country that vary over time, are the time-fixed effects to control for time-varying global factors that affect all countries, sectors, and firms. The equation is estimated by OLS using robust standard errors. Weights are employed in the analysis to equalize the emphasis on smaller versus larger countries. Note that relevant firm characteristics are carefully considered for each regression: to reduce omitted variable concerns, size dummies are incorporated when assessing labor productivity or sales as the dependent variables, while sales quartiles are used when employment is the dependent variable. The age of firms is controlled for because older firms may utilize energy efficiency practices and generate outcomes different from younger firms. 8 The overall effect of electricity price fluctuations on a firm in the energy- intensive (CPRS) sector is then given by: + . 6 Bartelsman et al. (2004) highlight its importance in developing economies, where sales data often provide the most accessible alternative for productivity analysis. 7 Micro (1 to 5 employees), small (6 to 19), medium (20 to 99), large (100 or more employees); the omitted category in the regressions is “small”. When the dependent variable is employment, size is determined based on the quartiles of sales, with the lowest quartile serving as the omitted category. 8 The age of a firm significantly influences its performance, reflecting variations in experience, resource accumulation, and adaptability. Young firms often exhibit higher growth rates but face greater financial constraints and survival risks, as noted by Haltiwanger et al. (2013). In contrast, older firms benefit from established customer bases and operational efficiencies but may encounter innovation inertia (Huergo & Jaumandreu, 2004). Coad et al. (2016) find that firm age interacts with industry dynamics, where mature firms in competitive sectors must innovate to sustain performance. 7 Equation (2) introduces the energy inefficiency variable, , which identifies firms that do not implement any action to manage or reduce energy usage. Notice that the variable does not vary over time because it is only asked in the last wave of the follow-up surveys. There are three additional terms pertaining to : interacted with the electricity price, , , interacted with the energy intensity dummy, , and in a triple interaction with both the energy price and energy intensity vector. Because firms applying techniques to manage or reduce energy use will become less dependent on energy, we expect these firms to suffer less from the energy price increase. The interactions with electricity price, with the energy intense dummy, and the triple interaction will indicate whether firms that are more exposed to energy price shocks but have not introduced energy savings suffered more from the shock. We therefore expand equation (1) as follows: ,,, = ∙ , + ∙ , ∙ + ∙ , ∙ + ∙ , ∙ ∙ + ∙ + ∙ + ∙ ∙ + ∙ , + + + , + + ,,, (2) Note that we do not use firm-level fixed effects in this specification because they are perfectly correlated with the variable that varies only across firms. The overall effect of electricity price fluctuations on a firm in an energy-intensive sector that did not implement any energy efficiency measures will be thus given by: + + + . 4. Estimation results Table 1 reports the estimation results using labor productivity (sales per worker) and separately sales and employment as the dependent variables. The baseline estimation suggests that firms in energy-intensive sectors tend to be more productive and have higher sales (columns 1 and 2). Sales and productivity in energy-intensive industries can be higher due to economies of scale, technological advancements, and market dynamics. High capital intensity reduces per-unit costs as production scales (Hall & Weiss, 1967), while innovations enhance efficiency (Caves et al., 1982). Inelastic demand and energy-saving technologies further boost competitiveness (Pindyck, 1981; Aldy & Pizer, 2015). Similarly, firms in energy-intensive sectors tend to employ more full-time workers than firms of similar age, size, and sectoral characteristics in non-energy-intensive industries. This result could also be associated with the anecdotal evidence that more firms in energy-intensive industries are state-owned, with state- owned firms being known for over-employment. Evidence suggests that energy-intensive industries often have a higher proportion of state-owned enterprises (SOEs), which are commonly associated with overemployment. For instance, Lin, Cai, and Li (1998) highlight the prevalence of SOEs in sectors like steel production and coal mining, driven by social objectives such as employment stabilization rather than efficiency. Jefferson and Rawski (1994) and Dong and Putterman (2000) find that SOEs frequently maintain larger-than-necessary workforces, resulting in inefficiencies compared with privately-owned firms. This phenomenon is particularly evident in countries like China, where SOEs dominate energy-intensive sectors (Yusuf, Nabeshima, & Perkins, 2006), often prioritizing social stability over economic optimization. Increases in global oil prices are associated with a rise in firm labor productivity as firms increase sales and reduce employment. Firms may increase sales and reduce employment after oil price hikes in various ways, for instance, by adopting energy-efficient technologies that trigger productivity gains and passing- 8 through costs to consumers in inelastic markets (Aldy & Pizer, 2015; Pindyck, 1981). Simultaneously, cost increases might also lead to labor reductions, especially in energy-intensive sectors (Caves et al., 1982). For increases in electricity prices, we observe a negative effect on labor productivity in contrast to the positive effect observed for oil prices. Unlike global oil prices, electricity prices are country-specific and reflect the average commercial electricity rates by country, allowing for better identification. The negative association between electricity prices and labor productivity may be due to rising electricity prices increasing operating costs and reducing competitiveness, leading to a drop in sales, especially in energy- intensive sectors (Aldy & Pizer, 2015). Similarly, firms may cut production to manage costs in competitive markets, hindering output and profitability (Pindyck, 1981; Caves et al., 1982). Both petroleum and electricity subsidies are associated with lower firm employment through a strong negative effect. Petroleum and electricity subsidies can lower firm employment by encouraging capital- intensive production methods over labor-intensive ones. By reducing energy costs, subsidies incentivize firms to adopt automation and energy-efficient technologies, which reduces the need for workers. This shift towards automation is particularly evident in energy-intensive industries, where subsidies encourage investment in machinery instead of hiring labor (IMF, 2013). Additionally, subsidies can distort market signals, hindering the reallocation of resources to more labor-intensive sectors. As noted by the World Bank, these subsidies can lead to inefficiencies, promoting industries that rely more on capital than on labor, thus reducing overall employment opportunities (World Bank Group, 2024). By contrast, electricity subsidies—but not petroleum subsidies—are associated with increasing firm sales and productivity. Electricity subsidies can enhance firm sales and productivity by lowering operational costs, allowing firms to invest in energy-efficient technologies and increase production. This reduction in energy costs enables firms to allocate resources more effectively, boosting output (Stern, 2013). In contrast, petroleum subsidies often support capital-intensive industries, distorting market signals and hindering resource reallocation (IMF, 2013). As a result, while electricity subsidies promote efficiency, petroleum subsidies may not have the same effect on firm performance (Aldy & Pizer, 2015). Introducing interactions of energy prices and energy dependence Next, we introduce the double interaction of country-level electricity prices with industry-specific energy dependence (intensity) which helps us identify a more causal relationship than the estimated associations presented in Table 1. This is because we assume that country-level electricity prices are exogenous relative to firm-level performance and so is the industry-level energy-intensity measure (CPRS classification) based on global experience. The results of double interactions reported in Table 2 clarify the final effect of energy prices on firm performance. They reveal that increases in electricity and oil prices are associated with increased sales and productivity of firms in energy-intensive sectors. This finding is consistent with the earlier hypothesis, based on the results in Table 1, that increasing oil prices can raise labor productivity. Firms may increase sales while reducing employment after oil price hikes by adopting energy-efficient technologies that lead to productivity gains and by passing-through costs to consumers in inelastic markets while reducing employment in energy-intensive sectors due to cost pressures (Aldy & Pizer, 2015; Caves et al., 1982; Pindyck, 1981). In sum, our estimation results reveal a notable decline in employment across firms due to increasing electricity prices. This negative employment response is particularly accentuated in firms categorized as energy-intensive according to the CPRS classification and could be even more pronounced 9 if the policy response includes an increase in electricity subsidies. By contrast, rising electricity prices boost firm sales and productivity. These gains could be even more significant if a concurrent policy response introduces subsidies, which are fiscally costly and may transfer taxpayers' resources to firm owners and investors. Accounting for energy efficiency measures This section introduces the triple interaction among energy prices, industry-level energy intensity, and firm-level energy efficiency measures. It uses the instrumental variable (IV) approach to address a possible endogeneity of the firm-level energy-efficiency measures vis-à-vis the firm-level outcomes. This endogeneity may rise because well-performing firms may be more likely to implement energy-efficiency measures rather than only energy-efficiency measures affecting firm performance—raising reverse causality concerns. We use three candidate IV variables for the firm-level energy-efficiency measures. The validity of all three IV variables relies on the assumption that one firm taking an energy-efficiency measure cannot significantly influence all other firms of the same size across countries, firms in the same industry across countries, and firms in the same location (country). Namely, for each firm, we calculate the average likelihood of energy efficiency measure adoption across firms of similar size, excluding the given firms from the calculation, and the analogous averages for firms in the same subsector and in the same location, always excluding the given firm from the sample. Then, we generate three candidate IVs interacting for a given firm: (i) the size and sector averages, (ii) the size and subsector averages, and (iii) the size, subsector, and location averages. From (i)-(iii), the instruments are assumed to show greater relevance because the richer interaction of averages creates a synthetic firm that is more similar to the given firm. Table A4 in the Annex reports the results of the first-stage regression of firm-level efficiency measures on the thus constructed instrument for each firm. All three computed instruments are relevant and significant at the 1 percent level. The size-subsector-location instrument is the most significant, statistically and economically; therefore, we use it as our baseline IV in this section. We use the other two instruments in robustness checks. Table 3 reports the estimation results for equation (2) where we introduce the triple interaction of electricity prices, energy intensity of industries, and energy efficiency measures or lack thereof. The latter firm-level variable is instrumented by the size-sector-local average computed for each (excluding it from the computation sample). By controlling for double and triple interaction between global oil prices and country-level energy (commercial) electricity prices (oil and electricity), subsectoral energy intensity, and location (country) intensity, the estimation results reveal additional significant heterogeneities. We focus again on the interactions that include electricity prices because the data is more granular (i.e. country- level commercial electricity rates), and the energy efficiency measures reported at the firm level, which are mostly related to electricity consumption such as the LEED certification for buildings, adopting efficient lighting systems, adherence to ISO 14001 or 50001 standards. The interactions of energy prices and energy intensity remain significantly positive for sales and productivity—as was the case in the estimation results using double interactions only (and no triple interaction). When controlling for additional heterogeneity in the specification of equation (2), the double interaction of energy intensity and oil prices becomes significantly negative in the employment regression. The double interaction of energy intensity and electricity prices remains negative but loses significance at common levels. However, this loss of significance would be due to the significant heterogeneity identified by the double interaction of electricity prices and energy-efficiency measures and the triple interaction of 10 electricity prices, energy intensity, and energy-efficiency measures. The estimated coefficient on the double interaction between electricity prices and energy efficiency measures suggests that, after electricity price increases, firms that did not adopt energy efficiency measures reduce employment significantly more than other firms. The triple interaction is also estimated as significantly negative. Together with the double interaction, it delivers an effect that is economically almost three times as large as the respective double interaction effect in Table 2—suggesting that, after electricity prices increase, firms in energy-intensive sectors that did not adopt energy efficiency measures are those that markedly reduce employment. An increase in electricity prices by 1 percent can lead to a reduction in employment by about 1.5 percent in firms within energy-intensive industries that have not adopted energy efficiency measures. Conversely, these energy-intensive firms tend to see an increase in sales and productivity by approximately 1.4 percent, regardless of whether they have or have not implemented energy efficiency measures. These findings suggest that promoting the adoption of energy efficiency measures among firms could serve as an important employment protection policy during periods of electricity price volatility, which developing countries might consider prioritizing. The impact of energy subsidies, which could serve as a complementary policy to mitigate the effects of energy price increases, is further elucidated by incorporating double and triple interactions and considering a broader range of heterogeneity. Namely, the effect of electricity subsidies on sales and productivity remains positive, and the effect of petroleum subsidies on sales and productivity becomes significantly negative. Both electricity and petroleum negatively affect employment but only the effect of electricity is statistically significant, at the 1 percent level. Our results suggest that using energy subsidies to preserve employment during energy price shocks can be a counterproductive policy strategy. This result aligns with the literature and the possible negative effect of subsidies on the pace of structural adjustment and the long-term distortionary effect of employment. Such subsidies can distort market signals to both consumers and firms, leading to inefficient energy consumption, misdirected investments, misallocation of resources, and future resilience to energy price shocks (Coady et al., 2024; Aldasoro & Faia, 2024; Coady et al., 2024). This inefficiency can hinder economic growth and delay necessary adjustments to energy price changes. Furthermore, energy subsidies can strain public finances, limiting the government's ability to invest in other critical areas that support employment and economic growth (Clements et al., 2024). Finally, we explore the correlation between country-level energy subsidies and the firm-level adoption of energy efficiency measures. Table 4 reports the estimation results, which suggest that the short-term (one- year-lagged (t-1)) and medium-term (the preceding five-year average (t-1 … t-5)) subsidies significantly discourage firms from adopting energy efficiency measures—with the regression controlling for firm size, subsector, and country’s economic development. The medium-term effect of petroleum subsidies comes out as the strongest. However, even the short-term effect of petroleum and electricity subsidies is consistently significantly negative, highlighting that even short-term subsidies can delay the much-needed structural adjustment when possibly sustained energy price shocks hit countries, industries, and firms. These estimated correlations align with the literature highlighting that short- and medium-term subsidies negatively impact energy efficiency adoption (Stefanski, 2024; Davis, 2023). Subsidized firms often delay technological investments, worsening inefficiencies (Van den Bergh & Delarue, 2023). Subsidies risk long- term harm to growth and employment. 5. Robustness checks 11 This section carries out several robustness checks and reports the results. First, we test whether our results could change with the use of alternative instrumental variables and report the results in Table B1 in Annex B. Both alternative instruments—IV2 based on size-sector and IV3 based on size-subsector—confirm the robustness of our baseline results. Next, we perform a battery of other robustness checks and report them in Table B2. In columns 1-3, we include country-time fixed effects to check whether other country-level macroeconomic factors, other than energy prices, could have affected firm performance: productivity, sales, and employment, respectively. In columns 4-6, we cluster the standard errors by country-sector to allow for country- and sector-specific spillovers (the results do not change materially if we cluster only by country or only by sector). In columns 7-9, we drop the weights for country size to treat all firms in larger and smaller countries with equal importance and, in columns 10-12, we also drop India, the largest country in our sample. In columns 13-15, we replace the current monthly/quarterly price for electricity (and oil) with a 3-month lagged average (using a 6-month lagged average does not materially change the results). Overall, the impact of electricity price shocks on productivity and sales is not robust across the considered alternative specifications. By contrast, the result that electricity price shocks significantly decrease employment at firms in energy-intensive sectors that did not adopt energy efficiency measures remains robust across all alternative specifications. 6. Conclusion The paper examined the impact of electricity price changes on firms' labor productivity, sales, and employment across 24 developing countries from 2019 to 2023. It used business-specific country-level tariffs and considered the energy dependence of industries as per the methodology of Battiston et al. (2022), as well as the energy efficiency measures taken by individual firms. Our study corroborates the findings of Cali et al. (2022) and Cali et al. (2023), showing that, on average, firms may experience an increase in productivity and sales when electricity prices rise but these results are not sufficiently robust to alternative model specifications. By contrast, a rise in electricity prices significantly reduces employment, particularly impacting firms in energy-intensive industries that did not implement energy efficiency measures. These results underscore the critical role of integrating industry-level energy dependence with firm-level energy efficiency initiatives to fully comprehend the effects of energy price fluctuations on firm performance. To safeguard quality jobs amid energy price fluctuations, including during the green transition, policy makers must encourage widespread adoption of energy efficiency measures, especially by firms in energy-intensive sectors. 12 References Aldasoro, I., & Faia, E. (2024). Energy subsidies and economic efficiency: Market distortions in times of crisis. Economic Policy, 39(120), 711-738. Aldy, J. E., & Pizer, W. A. (2015). "The Competitiveness Impacts of Climate Change Mitigation Policies." Journal of the Association of Environmental and Resource Economists, 2(4), 565–595. Andre, C., Costa, H., Demmou, L. and G. Franco (2023). Rising Energy Prices and Productivity: Short Run Pain, Long term gain? OECD Economic Department Working Paper # 1755. Ari, A., Arregui, N., Black, S., Celasun, O., Iakova, D., Mineshima, A., Mylonas, V., Parry, I., Teodoru, I., & Zhunussova, K. (2022). Surging Energy Prices in Europe in the Aftermath of the War: How to Support the Vulnerable and Speed up the Transition Away from Fossil Fuels. IMF Working Papers, 2022/152 Bartelsman, E. J., Haltiwanger, J., & Scarpetta, S. (2004). Microeconomic Evidence of Creative Destruction in Industrial and Developing Countries. World Bank Policy Research Working Paper, No. 3464. Battistini, N, Bobasu, A and J. Gareis (2023). Who foots the bill? The uneven impact of the recent energy price shock. European Central Bank Economic Bulletin, issue 2/2023. Battiston S., Monasterolo I., van Ruijven B., Krey V. (2022) Mapping economic activities into climate scenarios and transition risk classes: the NACE-CPRS-IAM classification, NGFS Technical notes. Battiston, S., Mandel, A., Monasterolo, I., Schütze, F., & Visentin, G. (2017). A Climate stress-test of the financial system. Nature Climate Change, 7(4), 283–288. https://doi.org/doi:10.1038/nclimate3255) Black, S., Liu, A., Parry, I., and N. Vermon (2023). IMF Fossil Fuel Subsidies Data: 2023 Update. IMF Working Paper # P/23/169 Bloom, N. (2009). The Impact of Uncertainty Shocks. Econometrica, 77(3), 623–685. Bloom, N., Genakos, C., Martin, R., & Sadun, R. (2010). "Modern Management: Good for the Environment or Just Hot Air?" Economic Journal, 120(544), 551–572. Bovenberg, A. L., & de Mooij, R. A. (1994). Environmental Levies and Distortionary Taxation. American Economic Review, 84(4), 1085–1089. Cali, M., Cantore, N., Iacovone, L., Pereira-Lopez, M., Presidente, G. (2022). The perverse effects of low fuel prices on firms. Journal of Environmental Economics and Management 111. Cali, M., Cantore, N., Marin, G., Mazzanti, M., Nicolli, F., Presidente, G. (2023). Energy prices and the economic performance of firms in emerging countries. Structural Change and Economic Dynamics 66. Caves, D., Christensen, L., & Diewert, W. (1982). "Multilateral Comparisons of Output, Input, and Productivity Using Superlative Index Numbers." The Economic Journal, 92(365), 73–86. Clements, B., Gupta, S., & Keen, M. (2024). The Fiscal Costs of Energy Subsidies: A Macroeconomic Perspective. IMF Working Papers, 24(081). 13 Coad, A., Segarra, A., & Teruel, M. (2016). Innovation and Firm Growth: Does Firm Age Play a Role? Research Policy, 45(2), 387–400. Coady, D., Parry, I., & Shang, B. (2024). Energy Price Reforms and Efficiency: Lessons from IMF Country Case Studies. IMF Staff Papers, 24(027). Coady, D., Parry, I., Le, N., & Shang, B. (2019). Global Fossil Fuel Subsidies Remain Large: An Update Based on Country-Level Estimates. World Development, 91, 11–27. Copeland, B. R., & Taylor, M. S. (2004). Trade, Growth, and the Environment. Journal of Economic Literature, 42(1), 7–71. Davis, L. (2023). The hidden cost of energy subsidies: Evidence from firm investment data. Energy Economics, 54(3), 112-129. Dong, X., & Putterman, L. (2000). China's State-Owned Enterprises in the First Reform Decade: An Analysis of a Declining Monopsony. The Economics of Transition, 8(1), 151-178. Hall, M., & Weiss, L. (1967). Firm Size and Profitability. The Review of Economics and Statistics, 49(3), 319– 331. Haltiwanger, J., Jarmin, R. S., & Miranda, J. (2013). Who Creates Jobs? Small vs. Large vs. Young. Review of Economics and Statistics, 95(2), 347–361. Henriques, I., & Sadorsky, P. (2008). Oil Prices and the Stock Prices of Alternative Energy Companies. Energy Economics, 30(3), 998–1010. Huergo, E., & Jaumandreu, J. (2004). Firms' Age, Process Innovation, and Productivity Growth. International Journal of Industrial Organization, 22(4), 541–559. IMF. (2013). Energy Subsidy Reform: Lessons and Implications. International Monetary Fund. Jefferson, G. H., & Rawski, T. G. (1994). Enterprise Reform in Chinese Industry. The Journal of Economic Perspectives, 8(2), 47-70. Kilian, L. (2008). The economic effects of energy price shocks. Journal of Economic Literature 46 (4), 871– 909. Lin, J. Y., Cai, F., & Li, Z. (1998). Competition, Policy Burdens, and State-Owned Enterprise Reform. The American Economic Review, 88(2), 422-427. Martin, R., et al. (2014). Industry Compensation Under Relocation Risk: A Firm-Level Analysis of the EU Emissions Trading Scheme. American Economic Review, 104(8), 2482–2508. Parry, I., Black, S., & Vernon, N. (2021). Still Not Getting Energy Prices Right: A Global and Country Update of Fossil Fuel Subsidies. IMF Staff Climate Notes, 2021/005. Pindyck, R. S. (1981). The Economic Effects of a Decline in Energy Price Volatility. The Energy Journal, 2(1), 1–19. Porter, M. E., & van der Linde, C. (1995). Toward a New Conception of the Environment-Competitiveness Relationship. Journal of Economic Perspectives, 9(4), 97–118. 14 Rentschler, J., Kornejew, M. (2017). Energy price variation and competitiveness: firm-level evidence from Indonesia. Energy Economics 67, 242–254. Sadorsky, P. (1999). Oil Price Shocks and Stock Market Activity. Energy Economics, 21(5), 449–469. Sato, M., et al. (2019). Energy Costs, Employment, and Emissions: Exploring the Heterogeneity of Effects Across Firms. Environmental and Resource Economics, 72(4), 935–962. Stefanski, R. (2024). Subsidies and energy efficiency: A firm-level analysis. Journal of Environmental Economics, 67(2), 345-370. Stern, D. I. (2013). The Role of Energy in Economic Growth. Energy Economics, 40, 15–22. Syverson, C. (2011). What Determines Productivity? Journal of Economic Literature, 49(2), 326–365. Van den Bergh, K., & Delarue, E. (2023). Energy subsidies and technological inertia: A cross-country comparison. Energy Policy, 48(5), 587-601. World Bank. (2024). Energy Subsidies and Economic Growth. World Bank Group. Yusuf, S., Nabeshima, K., & Perkins, D. H. (2006). Under New Ownership: Privatizing China's State-Owned Enterprises. Stanford University Press. 15 Figures Figure 1. Average Labor Productivity, Sales, and Employment Source: World Bank Business Pulse Survey 2020-2023 Figure 2. Energy prices Source oil prices: OPEC, monthly average Source electricity prices: Global Petrol Prices https://www.globalpetrolprices.com Quarterly electricity price, businesses rate (country level) 16 Figure 3a. Explicit subsidies to petroleum Figure 3b. Explicit subsidies to electricity Source: Parryet al. (2021) International Monetary Fund, staff estimations. 17 Tables Table 1: Estimation results based on Equation (1) (1) (2) (3) Sales per worker Sales (log) Employment (log) Energy intense 0.240** 0.255*** 0.235*** (0.0955) (0.0964) (0.0320) Price oil, log 11.64*** 11.20*** -2.044*** (0.450) (0.452) (0.140) Price electricity, log -1.818*** -1.978*** -0.734*** (0.295) (0.302) (0.0987) Explicit petroleum subsidy (% GDP) 1.468 1.469 -0.587** (1.010) (1.031) (0.298) Explicit electricity subsidy (% GDP) 8.011*** 7.852*** -0.463* (1.132) (1.131) (0.269) Constant -54.12*** -50.26*** 7.677*** (2.350) (2.364) (0.731) Observations 21,629 21,629 21,629 R-squared 0.225 0.331 0.471 Country FE YES YES YES Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Weighted by country sample size Controls for size clusters[1], age, wave, and subsector [1] Constructed using employment in columns (1) and (2), and sales in column (3) 18 Table 2a: Estimation results based on Equation (2) with double interactions (1) (2) (3) Sales per worker (log) Sales (log) Employment (log) Energy intense -0.410 -0.543 -0.579** (0.787) (0.794) (0.257) Price oil, log 11.34*** 10.90*** -1.997*** (0.455) (0.457) (0.142) Price electricity, log -1.923*** -2.064*** -0.594*** (0.301) (0.308) (0.101) Explicit petroleum subsidy (% GDP) 1.309 1.316 -0.516* (1.006) (1.027) (0.302) Explicit electricity subsidy (% GDP) 8.107*** 7.950*** -0.469* (1.121) (1.121) (0.269) log Price oil*Energy intense 0.483*** 0.490*** -0.0398 (0.179) (0.179) (0.0542) log Price elect*Energy intense 0.689*** 0.633*** -0.473*** (0.172) (0.174) (0.0573) Constant -52.99*** -49.09*** 7.733*** (2.382) (2.399) (0.744) Observations 21,629 21,629 21,629 R-squared 0.226 0.332 0.474 Country FE YES YES YES Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Weighted by country sample size Controls for size clusters[1], age, wave, and subsector Clusters constructed using employment in columns (1) and (2), and sales in column (3) 19 Table 2b: Estimation results based on Equation (2) with double interactions including subsidies (1) (2) (3) Sales per worker Sales (log) Employment (log) (log) Energy intense -0.334 -0.468 -0.590** (0.787) (0.794) (0.257) Price oil, log 12.21*** 11.74*** -2.176*** (0.474) (0.475) (0.143) Price electricity, log -1.979*** -2.083*** -0.520*** (0.260) (0.266) (0.0923) Explicit petroleum subsidy (% GDP) 63.85*** 61.58*** -13.66*** (12.83) (13.58) (5.187) Explicit electricity subsidy (% GDP) -6.694*** -6.935*** 1.567*** (1.561) (1.599) (0.545) log Price oil*Energy intense 0.476*** 0.482*** -0.0394 (0.179) (0.180) (0.0542) log Price elect*Energy intense 0.712*** 0.655*** -0.478*** (0.171) (0.173) (0.0573) log Price elect*Subsidy elect (% GDP) -6.125*** -6.195*** 0.783*** (0.647) (0.662) (0.215) log Price petroleum*Subsidy fuel (% GDP) -13.59*** -13.09*** 2.868*** (2.712) (2.868) (1.090) Constant -58.61*** -54.55*** 8.842*** (2.391) (2.404) (0.752) Observations 21,629 21,629 21,629 R-squared 0.230 0.335 0.474 Country FE YES YES YES Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Weighted by country sample size Controls for size clusters[1], age, wave, and subsector [1] Constructed using employment in columns (1) and (2), and sales in column (3) 20 Table 3: Estimation results based on Equation (2) with the triple interaction using IV approach (1) (2) (3) Sales per Employm worker Sales (log) ent (log) (log) Energy intense 0.873 0.808 0.266 (0.923) (0.936) (0.301) Energy inefficient IV1 -0.745 -1.103 -4.469*** (1.141) (1.165) (0.419) Price oil, log 12.08*** 11.54*** -1.974*** (0.486) (0.488) (0.134) Price electricity, log -2.385*** -2.353*** 0.217 (0.507) (0.517) (0.166) Explicit petroleum subsidy (% GDP) -1.996*** -1.918*** -0.0712 (0.679) (0.679) (0.179) Explicit electricity subsidy (% GDP) 6.943*** 6.690*** -1.092*** (0.868) (0.871) (0.243) Energy intense*Energy inefficient IV1 -1.337 -2.146* -1.542*** (1.249) (1.275) (0.495) log Price oil*Energy intense 0.555*** 0.594*** -0.110** (0.191) (0.191) (0.0505) log Price elect*Energy intense 1.368*** 1.421*** -0.141 (0.322) (0.327) (0.118) log Price elect*Energy inefficient IV1 -0.833 -1.008* -0.835*** (0.548) (0.559) (0.192) log Price elect*Energy intense*Energy inefficient IV1 -0.401 -0.839 -0.702*** (0.601) (0.612) (0.229) Constant -48.26*** -43.55*** 10.10*** (2.628) (2.656) (0.787) Observations 19,237 19,237 19,237 R-squared 0.233 0.335 0.571 Country FE YES YES YES Firm-level adoption of efficiency measures instrumented by average adoption of the measures by other firms in the same sector, of similar size and location. Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Weighted by country sample size Controls for size clusters[1], age, wave, and subsector [1] Constructed using employment in columns (1) and (2), and sales in column (3) 21 Table 4: Correlation of country-level energy subsidies and firm-level adoption of energy efficiency measures (1) (2) Energy Efficiency Energy Efficiency Petroleum subsidies (average previous 5 years) 3.091*** (0.274) Electricity subsidies (average previous 5 years) 0.215*** (0.0131) Petroleum subsidies (lagged) 0.602*** (0.190) Electricity subsidies (lagged) 0.275*** (0.0187) Micro -0.300*** -0.449*** (0.0344) (0.0339) Medium 1.003*** 1.040*** (0.0394) (0.0390) Large 1.188*** 1.181*** (0.0678) (0.0679) Age (log) 0.287*** 0.279*** (0.0227) (0.0235) Agriculture 0.0340 -0.0498 (0.0618) (0.0647) Construction 0.232*** 0.165** (0.0802) (0.0801) Retail -0.162*** -0.164*** (0.0381) (0.0375) Transportation 0.0695 0.0113 (0.0803) (0.0792) Accommodation 0.138* 0.155** (0.0714) (0.0690) Restaurants -0.0186 -0.0413 (0.0557) (0.0540) IT 0.535*** 0.513*** (0.0929) (0.0933) Financial 0.741*** 0.751*** (0.0828) (0.0836) Education -0.0861 -0.232** (0.0911) (0.0925) Health 0.267** 0.196* (0.119) (0.117) Other Services -0.0513 -0.0858** (0.0416) (0.0413) Constant -1.230*** -1.029*** (0.0774) (0.0750) Observations 13,706 13,706 Pseudo R2 0.3142 0.3027 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Weighted by country sample size. Controls for GDP per capita; Omitted variables: small and manufacturing 22 Annex A Table A1a: Sample by Year 2020 2021 2022 2023 Total Argentina 524 525 1019 0 2068 Armenia 92 1041 707 0 1840 Bangladesh 483 523 1026 0 2032 Bulgaria 867 864 550 0 2281 Comoros 0 0 0 600 600 Croatia 39 52 379 0 470 Ghana 4193 2068 0 3157 9418 Greece 0 0 1123 0 1123 India 2539 2526 3087 0 8152 Kenya 3539 1567 1819 0 6925 Kyrgyzstan 299 224 1075 0 1598 Malawi 842 0 2033 0 2875 Malaysia 63 134 1500 0 1697 Nepal 1430 1503 1506 0 4439 Pakistan 354 821 1527 0 2702 Paraguay 204 221 413 0 838 Philippines 0 0 3839 0 3839 Poland 726 808 0 0 1534 Romania 823 593 666 0 2082 Senegal 322 0 0 407 729 Tajikistan 403 245 1031 0 1679 Tunisia 1074 0 0 1996 3070 Uzbekistan 264 201 1017 0 1482 Vietnam 92 98 53 0 243 Total 19172 14014 24370 6160 63716 23 Table A1b. Sample by Subsector Agric. Manuf. Constr. Retail Transp. Hotels Food IT Fin. Edu. Health Other Argentina 0 601 2 1092 153 0 0 49 0 0 0 169 Armenia 0 746 64 516 51 57 128 109 0 0 0 36 Bangladesh 103 1439 21 285 21 3 47 9 2 2 0 97 Bulgaria 109 359 261 456 93 76 66 168 266 71 86 270 Comoros 106 127 125 109 60 66 0 0 0 0 0 0 Croatia 4 56 37 64 10 8 7 20 23 0 1 137 Ghana 352 1963 514 1692 172 172 310 196 92 211 164 2652 Greece 28 75 55 315 25 43 158 27 64 57 47 229 India 29 3218 165 1644 162 762 461 266 531 23 42 849 Kenya 458 1045 729 1011 434 470 602 146 413 643 162 812 Kyrgyzstan 310 422 75 452 26 10 75 5 18 33 9 163 Malawi 128 381 66 1272 81 112 178 15 4 63 35 528 Malaysia 72 323 120 432 38 61 0 55 180 87 41 190 Nepal 850 985 91 1012 70 135 688 23 18 42 39 486 Pakistan 87 348 154 227 73 38 285 77 37 46 89 1241 Paraguay 0 147 0 212 47 23 24 27 53 38 40 227 Philippines 153 134 177 1067 117 57 768 100 93 103 96 825 Poland 21 503 163 422 18 27 44 33 79 1 32 187 Romania 29 396 248 474 157 16 201 24 29 9 53 446 Senegal 152 218 9 187 15 9 25 2 3 6 5 96 Tajikistan 352 295 141 579 21 16 28 13 28 11 28 167 Tunisia 0 1213 63 883 259 0 102 66 0 0 67 417 Uzbekistan 203 362 81 423 52 11 64 21 22 17 38 188 Vietnam 11 97 0 50 0 0 0 0 0 0 4 68 Total 3557 15453 3361 14876 2155 2172 4261 1451 1955 1463 1078 10480 24 Table A1c: Sample by Size Micro Small Medium Large Argentina 1083 476 393 116 Armenia 379 583 388 213 Bangladesh 685 797 416 132 Bulgaria 958 612 393 130 Comoros 434 130 20 8 Croatia 198 112 93 55 Ghana 6534 1988 540 77 Greece 705 310 85 23 India 304 2971 4031 846 Kenya 2643 1857 1109 452 Kyrgyzstan 542 572 264 32 Malawi 2183 480 169 43 Malaysia 337 292 502 566 Nepal 2985 982 251 88 Pakistan 1569 504 263 90 Paraguay 452 230 98 35 Philippines 2243 713 216 207 Poland 305 467 547 213 Romania 748 691 488 154 Senegal 300 294 90 29 Tajikistan 460 692 366 5 Tunisia 1196 785 569 472 Uzbekistan 443 553 348 30 Vietnam 32 97 53 61 Total 27718 17188 11692 4077 25 Table A2. Summary Statistics N Mean SD Min Max Sales (000 USD) 23639 9.1e+09 1.0e+12 0.01 1.4e+14 Employment 60675 119220 11767238 1 2.0e+09 Sales per worker (000 USD) 23487 1.4e+09 1.7e11 2.5e-7 2.4e+13 Age 50219 17.7 13.3 1 203 Monthly oil price (USD) 63716 76.1 27.8 33.7 113 Quarterly electricity price (USD) 55762 0.14 0.06 0.03 0 Energy efficient 48964 0.62 0.49 0 1 Explicit petroleum subsidy (% GDP) 63716 0.21 0.68 0 3.58 Explicit electricity subsidy (% GDP) 63716 1.02 1.98 0 9.76 Country: Argentina N Mean SD Min Max Sales (000 USD) 1320 965 3736 0 65770 Employment 2068 25 81 1 2124 Sales per worker (000 USD) 1320 23 83 0 2603 Age 2065 22 18 1 96 Monthly oil price (USD) 2068 85 30 44 113 Quarterly electricity price (USD 2068 0 0 0 0 Energy inefficient 0 . . . . Armenia Sales (000 USD) 415 1149 10192 0 204458 Employment 1563 58 193 1 4001 Sales per worker (000 USD) 402 21 56 0 964 Age 1569 17 10 2 85 Monthly oil price (USD) 1840 83 23 34 113 Quarterly electricity price (USD 1840 0 0 0 0 Energy inefficient 1662 0 0 0 1 Bangladesh Sales (000 USD) 1525 151 905 0 27248 Employment 2030 39 192 1 5101 Sales per worker (000 USD) 1525 11 43 0 730 Age 1999 19 12 2 113 Monthly oil price (USD) 2032 83 29 34 113 Quarterly electricity price (USD 2032 0 0 0 0 Energy inefficient 2032 1 0 0 1 Bulgaria 26 Sales (000 USD) 661 3790 70818 0 1818182 Employment 2093 26 66 1 1001 Sales per worker (000 USD) 661 100 1399 0 35651 Age 1657 20 18 1 203 Monthly oil price (USD) 2281 68 26 34 106 Quarterly electricity price (USD 2281 0 0 0 0 Energy inefficient 2208 0 0 0 1 Comoros Sales (000 USD) 0 . . . . Employment 592 11 56 2 801 Sales per worker (000 USD) 0 . . . . Age 0 . . . . Monthly oil price (USD) 600 81 0 81 81 Quarterly electricity price (USD 0 . . . . Energy inefficient 0 . . . . Croatia Sales (000 USD) 77 7813 26759 0 208176 Employment 458 52 162 1 2001 Sales per worker (000 USD) 77 74 143 0 817 Age 0 . . . . Monthly oil price (USD) 470 100 22 44 113 Quarterly electricity price (USD 0 . . . . Energy inefficient 470 0 0 0 1 Ghana Sales (000 USD) 0 . . . . Employment 9139 8 30 1 1201 Sales per worker (000 USD) 0 . . . . Age 4193 16 12 1 119 Monthly oil price (USD) 9418 58 17 34 75 Quarterly electricity price (USD 9418 0 0 0 0 Energy inefficient 9394 1 0 0 1 Greece Sales (000 USD) 0 . . . . Employment 1123 20 168 1 5001 Sales per worker (000 USD) 0 . . . . Age 0 . . . . 27 Monthly oil price (USD) 1123 111 4 106 113 Quarterly electricity price (USD 0 . . . . Energy inefficient 1116 0 0 0 1 India Sales (000 USD) 5321 3888 89489 0 5152672 Employment 8152 57 192 2 8601 Sales per worker (000 USD) 5321 87 1918 0 109631 Age 8152 21 14 1 184 Monthly oil price (USD) 8152 78 29 34 113 Quarterly electricity price (USD 8152 0 0 0 0 Energy inefficient 8152 0 0 0 1 Kenya Sales (000 USD) 3303 321 1144 0 11453 Employment 6061 51 487 1 25001 Sales per worker (000 USD) 3224 21 117 0 2818 Age 6813 18 15 1 173 Monthly oil price (USD) 6925 61 21 34 90 Quarterly electricity price (USD 6925 0 0 0 0 Energy inefficient 5779 0 0 0 1 Kyrgyzstan Sales (000 USD) 62 44 129 0 738 Employment 1410 22 108 1 3001 Sales per worker (000 USD) 56 5 14 0 82 Age 822 11 8 1 32 Monthly oil price (USD) 1598 93 28 44 113 Quarterly electricity price (USD 1598 0 0 0 0 Energy inefficient 1382 0 0 0 1 Malawi Sales (000 USD) 2310 25 235 0 8544 Employment 2875 9 37 1 740 Sales per worker (000 USD) 2310 3 12 0 380 Age 2863 13 10 1 121 Monthly oil price (USD) 2875 77 22 43 107 Quarterly electricity price (USD 2875 0 0 0 0 Energy inefficient 2875 1 0 0 1 28 Malaysia Sales (000 USD) 1325 0 0 0 0 Employment 1697 164 606 2 13001 Sales per worker (000 USD) 1325 0 0 0 0 Age 0 . . . . Monthly oil price (USD) 1697 101 15 43 106 Quarterly electricity price (USD 1697 0 0 0 0 Energy inefficient 0 . . . . Nepal Sales (000 USD) 2830 313 3423 0 131789 Employment 4306 12 55 1 1501 Sales per worker (000 USD) 2810 17 112 0 3889 Age 4439 16 10 3 55 Monthly oil price (USD) 4439 75 30 34 113 Quarterly electricity price (USD 4439 0 0 0 0 Energy inefficient 4077 0 0 0 1 Pakistan Sales (000 USD) 1924 89 435 0 7810 Employment 2426 27 262 1 8001 Sales per worker (000 USD) 1923 14 65 0 1464 Age 2702 26 3 22 44 Monthly oil price (USD) 2702 88 25 34 113 Quarterly electricity price (USD 2702 0 0 0 0 Energy inefficient 2702 0 0 0 1 Paraguay Sales (000 USD) 535 17387899 47489599 0 2.86E+08 Employment 815 23 94 1 1366 Sales per worker (000 USD) 530 5136440 18771421 0 1.86E+08 Age 838 21 16 3 118 Monthly oil price (USD) 838 84 28 44 113 Quarterly electricity price (USD 838 0 0 0 0 Energy inefficient 769 1 0 0 1 Philippines Sales (000 USD) 0 . . . . Employment 3379 2140156 49827371 2 2.00E+09 Sales per worker (000 USD) 0 . . . . 29 Age 3430 10 14 1 173 Monthly oil price (USD) 3839 106 0 106 106 Quarterly electricity price (USD 0 . . . . Energy inefficient 0 . . . . Poland Sales (000 USD) 575 4432 7443 0 67358 Employment 1532 43 55 2 301 Sales per worker (000 USD) 575 103 245 0 3333 Age 1520 24 11 1 93 Monthly oil price (USD) 1534 56 17 34 77 Quarterly electricity price (USD 1534 0 0 0 0 Energy inefficient 1529 0 0 0 1 Romania Sales (000 USD) 326 6.60E+11 8.887E+12 0 1.44E+14 Employment 2081 28 50 1 551 Sales per worker (000 USD) 326 1.01E+11 1.417E+12 0 2.40E+13 Age 1702 18 9 1 101 Monthly oil price (USD) 2082 69 29 34 106 Quarterly electricity price (USD 2082 0 0 0 0 Energy inefficient 1966 0 0 0 1 Senegal Sales (000 USD) 382 85 356 0 4613 Employment 713 19 61 1 1001 Sales per worker (000 USD) 379 11 70 0 1253 Age 642 18 9 3 54 Monthly oil price (USD) 729 61 20 34 81 Quarterly electricity price (USD 729 0 0 0 0 Energy inefficient 0 . . . . Tajikistan Sales (000 USD) 159 243 1175 0 13157 Employment 1523 16 18 1 181 Sales per worker (000 USD) 146 13 52 0 572 Age 1040 12 9 1 62 Monthly oil price (USD) 1679 90 30 44 113 Quarterly electricity price (USD 0 . . . . Energy inefficient 1475 0 0 0 1 30 Tunisia Sales (000 USD) 250 30388 264704 0 3131673 Employment 3022 79 495 1 22001 Sales per worker (000 USD) 250 525 4940 0 55923 Age 3044 17 12 1 124 Monthly oil price (USD) 3070 66 21 34 81 Quarterly electricity price (USD 3070 0 0 0 0 Energy inefficient 0 . . . . Uzbekistan Sales (000 USD) 108 838 3621 0 30162 Employment 1374 22 107 1 3101 Sales per worker (000 USD) 96 62 272 0 1809 Age 729 9 8 1 77 Monthly oil price (USD) 1482 95 28 44 113 Quarterly electricity price (USD 1482 0 0 0 0 Energy inefficient 1376 0 0 0 1 Vietnam Sales (000 USD) 231 0 0 0 1 Employment 243 112 211 1 1399 Sales per worker (000 USD) 231 0 0 0 0 Age 0 . . . . Monthly oil price (USD) 243 66 26 34 106 Quarterly electricity price (USD 0 . . . . Energy inefficient 0 . . . . Table A3. Energy Intense Classification using Battiston et al. (2022) CPRS classification methodology Agriculture Direct CO2 emissions from fossil fuel Not energy intense but reductions via afforestation and low carbon farming; low substitutability as for transport Manufacturing Intensive use of energy according to EU Energy intense classification Carbon Leakage; mostly direct CO2 emissions (fuel mix); no fuel substitutability Construction Mostly direct CO2 emissions (fuel mix - Energy intense heating); no fuel substitutability Retail and wholesale Low CO2 emissions Not energy intense 31 Transportation Mostly direct CO2 emissions (fuel mix); Energy intense no fuel substitutability, but this is changing Accommodation low CO2 emissions Not energy intense Food services low CO2 emissions Not energy intense Information and communication low CO2 emissions Not energy intense Financial services or real state low CO2 emissions Not energy intense Education low CO2 emissions Not energy intense Health low CO2 emissions Not energy intense Other services low CO2 emissions Not energy intense 32 Table A4: Instruments’ correlation based on Equation (4) (1) (2) (3) Energy Energy Energy inefficient inefficient inefficient Energy inefficient IV-1 (size-subsector-location) 1.615*** (0.0951) Energy inefficient IV-2 (size-subsector) 0.791** (0.361) Energy inefficient IV-3 (size-sector) 2.261*** (0.623) Constant -1.384*** -1.790*** -2.208*** (0.144) (0.196) (0.242) Observations 13,706 13,706 13,706 Country FE YES YES YES Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Weighted by country sample size Controls for size, age, wave, subsector, and fuel and electricity average subsidies of the previous 5 years Controls for size[1], age, wave, subsector r, and fuel and electricity average subsidies of the previous 5 years [1] Constructed using employment in columns (1) and (2), and sales in column (3) 33 Appendix B: Robustness checks Table B1: Summary estimation results based on Equation (3) with triple interaction using an instrumental variable for energy inefficiency (1) (2) (3) (4) (5) (6) (7) (8) (9) Sales per Sales per Sales per Employment Employment Employment Sales (log) Sales (log) Sales (log) worker (log) worker (log) worker (log) (log) (log) (log) log Price elect*Energy intense*Energy inefficient IV1 -0.401 -0.839 -0.702*** (0.601) (0.612) (0.229) log Price elect*Energy intense*Energy inefficient IV2 0.135 -0.420 -0.666** (1.461) (1.484) (0.321) log Price elect*Energy intense*Energy inefficient IV3 -1.367 -1.965 -0.801** (1.568) (1.592) (0.329) log Price elect*Energy inefficient IV1 -0.833 -1.008* -0.835*** (0.548) (0.559) (0.192) log Price elect*Energy inefficient IV2 0.307 -0.0870 -0.596*** (1.109) (1.123) (0.219) log Price elect*Energy inefficient IV3 1.935 1.680 -0.483** (1.214) (1.230) (0.232) log Price elect*Energy intense 1.368*** 1.190** 1.690*** 1.421*** 1.283** 1.809*** -0.141 0.260** 0.306** (0.322) (0.516) (0.554) (0.327) (0.522) (0.559) (0.118) (0.121) (0.121) log Price electricity -2.385*** -3.092*** -3.693*** -2.353*** -3.068*** -3.726*** 0.217 -0.0739 -0.0294 (0.507) (0.494) (0.515) (0.517) (0.501) (0.522) (0.166) (0.102) (0.0998) Constant -48.26*** -51.50*** -51.39*** -43.55*** -47.00*** -46.93*** 10.10*** 8.543*** 8.076*** (2.628) (2.303) (2.346) (2.656) (2.327) (2.370) (0.787) (0.480) (0.417) Observations 19,237 19,237 19,237 19,237 19,237 19,237 19,237 19,237 19,237 R-squared 0.233 0.232 0.232 0.335 0.334 0.334 0.571 0.803 0.848 Country FE YES YES YES YES YES YES YES YES YES Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1 Weighted by country sample size; Controls for size[1], age, wave, and subsector [1] Sales if dependent variable is employment 34 Table B2: Summary estimation results based on Equation (3) with different specifications regarding FEs, clustering, weights, and energy price Country-Year FE Clustered by country-sector Not weighted by country size Not weighted excl. India 3-month MA energy price (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) Sales per Sales per Sales per Sales per Sales per Employm Employm Employm Employm Employm worker Sales (log) worker Sales (log) worker Sales (log) worker Sales (log) worker Sales (log) ent (log) ent (log) ent (log) ent (log) ent (log) (log) (log) (log) (log) (log) Price electricity, log 0.199 -0.248 -0.915** -2.711** -2.842** -0.538** -2.290*** -2.336*** -0.190 -0.741 -0.859* -0.0628 3.796*** 3.828*** -0.608*** (1.699) (1.734) (0.390) (1.152) (1.175) (0.231) (0.481) (0.488) (0.143) (0.483) (0.490) (0.146) (0.624) (0.630) (0.188) log Price elect*Energy 0.308 0.362 -0.109 1.662** 1.611** -0.316 0.762*** 0.807*** -0.109 0.399 0.465* -0.169* 1.649*** 1.625*** -0.276** intense (0.299) (0.305) (0.119) (0.755) (0.735) (0.219) (0.271) (0.274) (0.0876) (0.278) (0.280) (0.0876) (0.341) (0.345) (0.123) log Price elect*Energy -1.976*** -2.188*** -0.981*** 0.155 0.176 0.151 -0.727 -0.811 -0.380** -1.927*** -1.867*** -0.425*** -0.718 -1.012* -0.970*** inefficient (0.574) (0.586) (0.211) (0.426) (0.432) (0.227) (0.488) (0.495) (0.155) (0.497) (0.504) (0.155) (0.587) (0.597) (0.211) log El. Price*Energy 0.566 0.126 -0.678*** -0.644 -0.729 -0.516* 0.675 0.404 -0.443*** 1.074* 0.792 -0.463*** -0.512 -0.741 -0.441* intense*Energy inefficient (0.580) (0.594) (0.229) (0.470) (0.466) (0.305) (0.582) (0.588) (0.159) (0.583) (0.588) (0.159) (0.629) (0.640) (0.231) Constant -0.103 1.907 -1.077 -30.01*** -26.29*** 6.583*** -36.97*** -32.74*** 7.530*** -13.86*** -10.23*** 6.389*** -6.419*** -3.465 1.993*** (5.092) (5.187) (1.310) (7.498) (7.595) (1.018) (2.267) (2.294) (0.672) (2.306) (2.337) (0.703) (2.261) (2.292) (0.675) Observations 19,237 19,237 19,237 19,237 19,237 19,237 19,237 19,237 19,237 13,916 13,916 13,916 19,237 19,237 19,237 R-squared 0.306 0.396 0.580 0.215 0.321 0.490 0.212 0.310 0.552 0.215 0.321 0.466 0.225 0.330 0.567 Country FE YES YES YES YES YES YES YES YES YES YES YES YES YES YES YES Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Weighted by country sample size Controls for size[1], age, wave, and subsector [1] Sales if dependent variable is employment 35