Policy Research Working Paper 10762 Hotter Planet, Hotter Factories Uneven Impacts of Climate Change on Productivity Woubet Kassa Andinet Woldemichael Africa Region Office of the Chief Economist May 2024 Policy Research Working Paper 10762 Abstract This study documents the impacts of climate change on located, but also on other factors such as firm size, industry firm-level productivity by matching a globally comparable classification, income group, and region. Large firms, firms and standardized survey of nonagricultural firms covering in manufacturing, and those in low-income countries and 154 countries with climate data. The findings show that hotter climate zones tend to experience the biggest pro- the overall effects of rising temperatures on productiv- ductivity losses. The uneven impacts, with firms in already ity are negative but nonlinear and uneven across climate hotter regions and low-income countries experiencing zones. Firms in hotter zones experience steeper losses with steeper losses in productivity, suggest that climate change is increases in temperature. A 1 degree Celsius increase from reinforcing global income inequality. If the trends in global the typical wet-bulb temperature levels in the hottest cli- warming are not reversed over the coming decades, there is mate zone (25.7 degrees Celsius and above) results in a a heightened risk of widening inequality across countries. productivity decline of about 20.8 percent compared to The implications are especially dire for the poorest countries firms in the coldest climate zone. The effects vary not only in the hottest regions. based on the temperature zones within which firms are This paper is a product of the World Bank Office of the Chief Economist, Africa Region and the International Monetary Fund. 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 wkassa1@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 Hotter Planet, Hotter Factories Uneven Impacts of Climate Change on Productivity∗ Woubet Kassa† Andinet Woldemichael‡ Key Words: Climate Change, Firms, Labor Productivity, Temperature Changes JEL Codes: D24, J22, J24, J81, O14, O12, Q54 Q56, L60 ∗ The authors extend their sincere thanks to A. Patrick Behrer, Marco Marini, and Jim Tebrake for their thoughtful feedback. We also thank Cesar Calderon and Andrew Dabalen for their insights and comments. The study was enriched by feedback from attendees at various academic forums including the IMF STA Innovation Talk, the ASSA 2024 Annual Meeting of the American Economic Association, the Southern Economic Association, the Association of Environmental and Resource Economists and the Center for the Study of African Economies, University of Oxford. We are grateful to Somik V. Lall for his thorough review and valuable comments. Special thanks to Joshua Seth Wimpey for access and support with the WBES data. The views expressed herein are those of the author and should not be attributed to either the WBG, the IMF, its Executive Board, or its management. † The World Bank. Email: wkassa1@worldbank.org ‡ International Monetary Fund. Email: awoldemichael@imf.org 2 I. Introduction There is overwhelming scientific evidence that climate change poses possibly the biggest risk to the world and demands an urgent global response. Greenhouse gas emissions trap heat and lead to global warming, contributing to a rise in temperature (Stern, 2008). According to the Intergovernmental Panel on Cli- mate Change, a temperature rise of 1.0 degree Celsius (0 C ) above pre-industrial levels (1850–1900) has already materialized, and global warming of 1.50 C and 20 C will be exceeded during the 21st century unless deep reductions in carbon dioxide (CO2 ) and other greenhouse gas emissions occur in the coming decades (see Figure 1).(Masson-Delmotte et al., 2021)1 . Understanding the mechanisms through which climate change affects economies and estimating its impacts in various sectors are essential to implement mitigation and adaptation policies and actions at national and global scales. These mechanisms include climate finance and the implementation of the historic agreement at the 2022 United Nations Climate Change Conference to establish ”loss and damage” funds for vulnerable countries. Figure 1. : Global Surface Temperature Anomalies 2.0 2024 2023 1.5 2024 2023 Temperature anomaly (°C) 2020s 2010s 1.0 2000s 1990s 1980s 1970s 0.5 1960s 1950s Reference for preindustrial level (1850-1900) 1940s 0.0 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Source: Copernicus Climate Change Services (C3S) Data Store. This study examines the uneven effects of exposure to temperature anomalies on firm-level productivity.2 We use data from the World Bank Enterprise Sur- vey (WBES), which has more than 190,000 observations covering 154 countries 1 Some estimates put the increase in temperatures between 40 C , and 60 C (Houghton, 2004; World Bank, 2013) 2 The US National Aeronautics and Space Administration defines temperature anomaly as “the dif- ference in temperature from an average or baseline.” The baseline temperature is typically computed by averaging 30 or more years of temperature data. HOTTER PLANET, HOTTER FACTORIES 3 between 2006 and 2021. Using geomasked coordinates, we merge the WBES firm-level data with gridded historical monthly temperatures and relative humid- ity data between 1980 and 2021 obtained from the European Union’s Copernicus Climate Change Services (C3S). The gridded climate data have 0.250 × 0.250 hor- izontal resolution, which is a grid of approximately 27.5 × 27.5 kilometers (km). Climate change in this study refers to variations in climate patterns captured by temperature anomalies. We use both near-surface temperatures and relative humidity to calculate the wet-bulb temperature (WBT), which is used to calcu- late the heat index. The heat index is considered a good measure of heat stress conditions that can affect the human body.3 Our main source of identification is the exogenous variation in area-level deviation of the WBT from the long-term average.4 The area-level WBT is measured within a radius of 30 km around the establishment, and the area-level long-term average is the mean of the WBTs for each month between 1980 and 2021. We estimate a nonlinear regression model of log sales per worker—a good proxy for firm productivity—on WBT deviation, controlling for firm- and location-specific characteristics and country, subnational, region, and year fixed effects. We also perform heterogeneity analysis by firm size and industry classification, thus contributing to the rapidly growing literature that evaluates the impact of climate change on various aspects of the economy. The economic impacts of climate change, particularly rising temperature, on agri- culture and related sectors are now better understood and well documented in the enes and Green- literature (Mendelsohn et al., 1994; Schlenker et al., 2006; Deschˆ stone, 2007; Cline, 2007; Fisher et al., 2012; Carter et al., 2018; Arag´on et al., 2021). Most other studies provide estimates and predictions of aggregate impacts on output and economic growth. Studies on the micro-level impacts of climate change on firm productivity often focus on a few individual countries. This re- stricts our understanding of the possibly uneven and heterogeneous impacts of climate change across various climate zones and across countries in these zones. Early studies that estimated or predicted the macroeconomic impacts of changes in temperature on production, investment, health, and agriculture showed that in- creasing temperatures have large and uneven negative effects on economic growth and output, particularly in poorer countries (Dell et al., 2012; Burke et al., 2015). Acevedo et al. (2020) shows that the negative effect of temperature on aggre- gate output in countries with hot climates—mostly low-income countries—runs through reduced investment, depressed labor productivity, poorer human health, and lower agricultural and industrial output. Heal and Park (2013) find a strong association between temperature deviations from average and per capita income. 3 The US National Aeronautics and Space Administration considers the heat index as the “apparent temperature” or the temperature that the human body “feels”. 4 In the literature, deviation from the long-term average or reference year is typically referred to as a temperature anomaly. 4 They find that hotter years are associated with lower or higher output per capita ranging between 3 and 4 percent for hot or cold climate zones, respectively. In most cases, labor productivity may be the key link between climate shocks and economic outcomes at the macro level (Heal and Park, 2013). Yet, there are few studies on the direct links between labor productivity and temperature changes. Tol (2009) labels the labor productivity impacts of climate change as unknown un- knowns in a review of studies on the economic effects of climate change, noting the wide gap in the literature, although there has been significant progress since then. More recent studies have examined the impacts of rising temperatures on various aspects of labor productivity at the micro level (see Lai et al. (2023) for a review). However, many such studies provide average estimates of the impact of climate change or temperature deviations from the average, without accounting for the potentially uneven impacts across climate zones, regions, or countries. Neglect- ing the uneven impacts will have significant implications for understanding how climate change shapes the future distribution of economic outcomes across these categories. Using detailed production data from half a million manufacturing firms in China, Zou and Zhong (2022) find a relatively large negative impact of excess temperatures—a day with average temperature above 90 degrees Fahren- heit (0 F) is associated with a total factor productivity (TFP) loss of 0.56 percent, relative to a day with average temperature between 500 F and 600 F. A study of a census of manufacturing firms in India shows that annual plant output falls by about 2 percent per 10 C increase in temperature (Somanathan et al., 2021). In a U.S. study, Deryugina and Hsiang (2014) show that productivity on an individual day declines by 1.7 percent for each 10 C (1.80 F) increase in daily average temper- ature above 150 C (590 F). Many of these studies that examine the productivity impacts of temperature focus on a few individual countries, hence providing only limited variations in terms of both the impact of temperature changes on produc- tivity and variations across countries by income and geography. LoPalo (2023) addresses this challenge in an innovative study that extends the analysis to 46 countries, examining the effects of WBT on the productivity of Demographic and Health Survey interviewers. She finds that hot and humid temperatures significantly impact worker productivity. Data quality problems, such as missing responses and flags for poor data quality, become more frequent on hotter days and interviewers become less productive. The number of interviews completed per hour worked declines by 13.6 percent on the hottest days. However, interviews and data collection make up a unique context. LoPalo (2023) provides interesting evidence on the link between temperature and survey workers’ pro- ductivity from a broad set of countries and regions, allowing for heterogeneous impacts across countries. However, the findings cannot be generalized to enter- prises since data collection is a small share of economic activity, and the usually HOTTER PLANET, HOTTER FACTORIES 5 outdoor workplace setting of data collection is different from most production activities in non-agriculture sectors. Our study builds on and contributes to this literature. This paper makes two main contributions to the literature. First, it employs data from a global, standardized, and comparable survey of firms rather than an individual country. This presents opportunities to understand the potentially heterogeneous impacts of temperature change on firm productivity and allows the estimation of impacts across regions, climate zones, industries, and country income groups. To account for the potentially heterogeneous impacts of changes in temperature in colder and hotter climate zones, we estimate a nonlinear model using a binning approach in which we group together firms located in the same categories of temperatures. This may be the only study that estimates the im- pact of temperature change on productivity using a representative sample of firms in more than 150 countries, providing the most comprehensive study in the lit- erature. The large sample allows for additional heterogeneity analysis by firm characteristics, including firm size and industry classification, country income group, and region of the world. This analysis can be used in the determination and distribution of costs and investments associated with actions to mitigate and adapt to climate change in global negotiations. Second, this study uses high-resolution climate data capturing localized climate hazard impacts, which are the most relevant because the nature and extent of exposure and damage vary within a few thousand meters.5 The study com- bines gridded historical climate data with WBES firm-level data, allowing better identification of impact within a relatively highly geographically specified loca- tion. Thus, we can estimate important heterogeneity, controlling for both within- country and cross-country variations, going beyond cross-country to the level of subnational variations. In addition, the study contributes to the relatively scarce literature on the impacts of climate change on the non-agriculture sectors, es- pecially at the firm level. Despite the relatively rich literature on the impacts of climate change on agriculture, studies on the impacts on the non-agriculture sectors are relatively scarce. We document that the effects of rising temperatures are nonlinear and uneven across climate zones, where firms in hotter zones experience steeper losses in pro- ductivity with increases in temperature, compared to firms in relatively colder zones, which tend to register productivity gains. Specifically, a one unit (10 C ) increase in WBT deviation in the hottest WBT quantile 25.70 C and above results 5 The matched data on climate change and enterprises is another key contribution, since it can also be used by other researchers. 6 in decline in annual sales per worker by about 20.8 percent compared to firms in areas with the coldest WBT, where WBT ranges between 14.50 C and 23.70 C . We find positive impacts on productivity for firms located in the lower temperature zones, which reinforces the finding that the impacts of changes in temperature deviations are nonlinear across temperature zones. There is a change in the di- rection of impact or sign of the coefficient, suggesting a potential inflection point beyond which an increase in temperature has a detrimental impact on firm pro- ductivity, which is estimated to be 25.70 C in our sample. In addition, the effects vary not only based on the temperature categories within which firms are located, but also other factors, such as firm size, industry clas- sification, income group, and region. For example, large firms, firms in man- ufacturing, and those in low-income countries and hotter climate zones tend to experience the biggest productivity losses due to climate change. Given that many low-income countries are in hotter climate zones, the climate change impacts due to higher temperatures are further exacerbated by the limited capabilities to in- vest in adaptation. Poorer regions experience the highest losses due to climate change, especially in the hottest WBT categories. This adds to what we know about climate change reinforcing existing vulnerabilities in the regions of the world that are least capable of responding to the effects of climate change. By providing more granular evidence from 154 countries, the findings of this study have essential implications for current national and global policy debates on the costs of carbon, the distribution of gains and losses, and the distribution of re- sponsibilities and contributions to mitigate climate change. The remainder of the paper proceeds as follows. Section II briefly discusses the mechanisms through which the rise in temperature affects firm productivity. Sec- tion (III) describes the data and provides summary statistics of the outcome and control variables. Section (IV) lays out the empirical model and discusses our identification strategies. We present and discuss the key findings in Section (V), we conclude in Section (VI). II. Potential Mechanisms There are at least two channels through which exposure to heatwaves or high temperatures affects labor productivity—directly by impacting labor’s capacity to execute tasks and indirectly by impacting capital reallocation and causing dis- ruptions in the supply of key infrastructure, including power, and subsequent changes in energy prices due to climate change. In the first channel, exposure to temperatures above a certain threshold is associ- ated with an array of adverse impacts on human physiology, capacity to work, and HOTTER PLANET, HOTTER FACTORIES 7 cognitive performance. It poses a series of health risks, reducing labor productiv- ity at workplaces. Many of the adverse impacts, including reduced work capacity, heat stress, heat exhaustion, and dehydration, tend to materialize when the tem- perature exceeds the range of 250 C − 260 C (Kjellstrom et al., 2009a; Hsiang, 2010). About 30 percent of the global population currently lives in places where climatic conditions exceed the threshold for at least 20 days a year (Mora et al., 2017). Above this threshold, workers suffer heat stress 6 which is associated with reduced human performance and capacity. The workers must slow down to reduce their internal body heat and the risk of heat stroke. Elevated core temperature leads to physical fatigue, irritability, lethargy, impaired judgment, reduced vig- ilance, and loss of dexterity, coordination, and concentration (Kjellstrom et al., 2009a; ?; International Labour Organization, 2019). These adverse impacts of exposure to high temperatures could be worse in work environments in which the machinery also contributes to heat stress, particularly in a non-air-conditioned indoor workplace. Severe temperature changes could have catastrophic impacts. If body temperature rises above 380 C (”heat exhaustion”), physical and cognitive functions are impaired; above 40.60 C (”heat stroke”), the risks of organ damage, loss of consciousness, and death increase sharply (Klein et al., 2014). In addition, there are labor supply losses due to absenteeism (Somanathan et al., 2021), par- ticularly in the absence of indoor cooling technologies (Gupta and Somanathan, 2022), further reinforcing the productivity losses associated with higher temper- atures. However, these impacts are not indiscriminate as they depend on the ambient temperature, humidity, wind speed, and adoption of cooling technolo- gies. The second channel through which exposure to hotter temperatures affects pro- ductivity is the higher costs of adaptation in response to the adverse effects of higher temperatures. Businesses could be forced to redirect resources from other productive investments, such as purchases of new machinery and research and development, to investments in adaptation, such as purchasing climate control technologies. Severe heat could also lead to power outages, which introduce ad- ditional costs in the form of disruptions to business activity or investments in generators and other alternative sources of energy. In addition, extreme weather has a direct impact on the energy infrastructure itself as energy demand for cool- ing increases, overloading power grids and leading to outages. The rise in demand for power during hotter days could also contribute to rising energy prices, which in turn increases costs for businesses. In most instances, extended drought con- ditions adversely affect the level of hydropower generation, potentially resulting in power outages. All or one of these factors could force firms to invest in alter- native and potentially more expensive sources of power, such as solar or gasoline 6 Heat stress refers to the heat received in excess of that which the body can tolerate without suffering physiological impairment (Kjellstrom et al., 2009b; International Labour Organization, 2019). 8 backup generators. Finally, extreme temperature changes could compromise the effective functioning of critical infrastructure, communications, and transporta- tion systems, imposing additional disruptions to business activities. III. Data To estimate the impacts of temperature deviations from the long-term average on firm productivity, we match two sets of data. The first data set is global firm-level data from the WBES.7 This a nationally representative survey covering nonagricultural firms with five or more full-time permanent employees in low-, middle-, and high-income economies. The coverage is comprehensive, with more than 190,000 observations from 154 countries and spanning 2006–22 .8 The data are collected using a standardized questionnaire, allowing comparability across countries. The WBES collects information on several firm-level variables, includ- ing annual sales, number of workers, various firm-level characteristics, and self- reported obstacles to business, such as licensing and power outages. The WBES data include estimates of TFP for a subset of the sample for which detailed data on labor, capital, and material inputs are available. The WBES also contains confidential geomasked information on firms’ longitude and latitude coordinates that allows us to match the WBES data with area-level climate data. Geomasked coordinates are available for 143,047 observations. Map 1 shows the countries covered by the WBES for which geomasked locations of the firms are available, where the dots represent the specific locations of firms in the sample. The key outcome variable of interest is annual sales per worker, measured in 2009 US dollars, which is considered in the literature as a reasonable proxy for firm productivity. A limitation of this measure is that the price variation in sales may reflect both supply and demand factors including differences in market power, demand and quality (Cusolito and Maloney, 2018). Its attraction, however it its simplicity and direct interpretation compared to other relatively complicated productivity measures. We also use other measures of firm productivity, such as value added per worker and TFP. However, the latter two measures have a large number of missing observations, which we suspect are systematic. Another key firm performance indicator we examined, which is not reported here, is employ- ment growth, which also proxies for firm growth or lack thereof. We dropped observations from the sample with outliers in sales per worker, which represented a very small proportion of the total sample size.9 7 The WBES can be accessed at: http://www.enterprisesurveys.org We thank the Enterprise Analysis Unit of the Development Economics Global Indicators Department for making the data available. 8 Some firms appear more than once in countries with repeat surveys. 9 The criterion used to identify outliers is Outlier = | xi −x ˆ > 3|, where x ˆ is the median value and SD SD is the standard deviation. HOTTER PLANET, HOTTER FACTORIES 9 Figure 2. : The World Bank Enterprise Survey Coverage and Firms Location Source: Original map for this paper, based data from the World Bank Enterprise Survey Data. Note: Darkred dots represent the location of the firms. Table 1 presents summary statistics of the key variables, including the dependent variable, sales per worker, and selected control variables, including firm-level char- acteristics such as firm size, age, broad industry classification, ownership struc- ture, export status, and business obstacles reported by firms. Only about 15 percent of the firms are exporters, and 10 percent are considered foreign. There is a fairly even distribution of firm sizes: 47 percent are small (employing 5-19 workers), and the rest are medium-sized (20-99 workers) or large (100+ workers). About 55 percent of the firms are in the manufacturing sector, and the remaining 45 percent are in services. All regions are well represented in the sample: Sub- Saharan Africa accounts for 20 percent of the firms; East Asia and the Pacific and South Asia together, 20 percent; Europe and Central Asia, 31 percent; Latin America and the Caribbean, 20 percent; and the Middle East and North Africa, 10 percent. Further, we control for local socioeconomic factors that are likely to be correlated with firm-level productivity. These include population density from SADEC, road infrastructure density from Global Roads Inventory Project - GRIP - version 4, and pollution using ground-level fine particulate matter of 2.5 micrometers or smaller from NASA/Socioeconomic Data and Applications Center (SADEC) The second data set we use contains information on gridded historical tempera- tures and relative humidity from the European Union’s C3S Climate Data Store 10 Table 1—: Summary Statistics Variable Mean (SD) Annual average dry-bulb temperature (0 C) 29.11 (0.72) Annual average WBT (0 C) 24.62 (1.84) Annual average WBT (0 F) 75.28 (3.55) Deviation in average dry-bulb temperature (0 C) 0.04 (0.06) Deviation in average WBT (0 C) 0.62 (0.47) Deviation in average WBT (0 F) 33.10 (0.85) Sales per worker (US$, 2009) 71,730 (240,491) Ownership status Domestic (%) 90 Foreign (%) 10 Firm size: categorical Small(<20) (%) 47 Medium(20-99) (%) 34 Large(100 and over) (%) 19 Firm size: continuous 74.44 (179.01) Permanent workers (%) 95.30 (11.68) Temporary workers (%) 4.70 (11.68) Skilled workers (%) 71.16 (30.72) Unskilled workers (%) 28.84 (30.72) Firm age (year) 18.76 (15.72) Exporter status Non-exporter (%) 85 Exporter (%) 15 Road infrastructure (within 25km radius) Highways (km) 66.18 (114.58) Primary roads (km) 291.52 (365.83) Secondary roads (km) 325.94 (372.01) Tertiary roads (km) 426.39 (446.90) Population density 5,248.32 (6,889.85) PM2.5: diff between 1998 and 2019 -2.72 (13.15) Broad sector Manufacturing (%) 55 Services (%) 45 Region Africa (%) 20 East Asia and Pacific (%) 11 Europe and Central Asia (%) 31 Latin America and the Caribbean (%) 20 Middle East and North Africa (%) 10.0 South Asia (%) 9.2 Number of observations = 141,815 (Sabater, 2019).10 The C3S provides a comprehensive reanalysis dataset of var- 10 Historical temperatures and relative humidity can be obtained from the European Union’s C3S Climate Data Store. We downloaded the data on November 25, 2022. HOTTER PLANET, HOTTER FACTORIES 11 ious climate variables at high resolution and global scale. The data have global coverage that is gridded to a regular latitude-longitude grid of 0.250 × 0.250 or ≈ 27.5 × 27.5km grid spacing, covering 1980 to the present. We use the ERA5- Land data set of monthly temperatures and relative humidity measured within a range of 289 centimeters of soil depth to 2 meters above the surface level. Using these sets of climate data, we perform two computations. First, we compute the WBT11 in degrees Celsius to account for the effects of humidity on the human body when combined with high temperatures. It is a common practice in the literature to use WBT rather than dry-bulb or air temperature since the effect of changes in temperature varies at different levels of humidity (Kjellstrom et al., 2009c,a; Lemke and Kjellstrom, 2012; Adhvaryu et al., 2020; LoPalo, 2023). WBT provides a better measure for assessing the risks to the human body and health of both temperature and humidity, compared with using only air temperature. WBT is a nonlinear function of temperature and relative humidity, and it is often lower than (dry-bulb) temperature measures. We follow Chen and Chen (2022) and use the following formula to calculate the WBT: 1 W BT = T · tan−1 [0.152 · (rh + 8.314)( 2 ) ] + tan−1 (T + rh)− (1) 3 tan−1 (rh–1.676) + 0.004(rh)( 2 ) tan−1 (0.0231rh)–4.686 where W BT is the wet-bulb temperature (degree Celsius), T is the near-surface dry-bulb temperature (degree Celsius), and rh is relative humidity (percent). Second, for each geographic area j within a radius of 30km around the estab- lishment, we calculate the WBT deviations from the long-term average for each month of the year as follows: m=12 (W BTj,mt − W BTj,m ) (2) ∆W BTj,t = , 12 m=1 where ∆W BTj,t denotes the average deviation in WBT in location j and year t, W BTj,mt is WBT in location j for the month of m, and year t, W BTj,m is the long-term average (typical) WBT for the month of m in location j . A key contribution of this study is the estimation of the potentially non-linear effects of climate change on firm-level productivity. To achieve this, we group firms into four quartiles of WBTs: quartile 1: ≤ 23.70 C (the minimum WBT in 11 All temperature measures are in degrees Celsius unless otherwise specified. 12 our sample is 14.50 C), quartile 2: (23.70 C, 24.90 C ], quartile 3: (24.90 C, 25.70 C ], and quartile 4: ≥ 25.70 C (the maximum WBT in our sample is 28.220 C). The first quartile is the coldest climate zone, and the fourth quartile is the hottest with WBT above the threshold beyond which the human body starts to experi- ence heat stress. Various studies adopt a similar binning strategy to estimate the non-linear effects of climate change including Deschˆ enes and Greenstone (2007); Deryugina and Hsiang (2014); Somanathan et al. (2021) and (Chen and Chen, 2022). Identifying the impacts of changes in temperature is particularly impor- tant above certain key thresholds, which we attempt to capture in our categories. Identifying the impacts of changes in temperature is particularly important above certain key thresholds, which we attempt to capture in these WBT categories. Figure 3. : The difference in average monthly WBT between 1980 and 2021 Source: Original map for this paper, using the ERA5-Land data set from the Copernicus Climate Change Service Climate Data Store Figure 3 presents the difference in WBT between 1980 and 2021 for countries for which WBES data are available. The planet has been getting hotter in recent decades compared to the average WBT in 1980. The average annual deviation in WBT from the long-term average between 1980 and 2021 is 0.620 C in our sample. Although the rise in temperature over the past 41 years seems universal, there is considerable heterogeneity across geographic locations, with some areas expe- riencing WBT increases as high as 3.100 C. Countries such as India, Bangladesh, Myanmar, Thailand, and other countries in the Southeast Asia region experienced much higher increases in WBT compared to other regions of the world. We as- HOTTER PLANET, HOTTER FACTORIES 13 sess the extent to which such variation drives the observed variation in firm-level productivity. We match the WBT data with firms’ geomasked coordinates. Given the minimum resolution of ≈ 27.5 × 27.5km for the temperature and relative humidity data, we extract historical WBT values within a radius of 30km.12 The average an- nual dry-bulb temperature for our pooled observation is about 290 C (≈75.280 F) with the corresponding annual average WBT of 24.60 C (Table 1). Table 1 also shows the average WBT deviation from the long-term average, which is 0.620 C (≈33.10 F ). The deviations are positive in all locations showing that increae in WBT is universal. In addition, the extent of increase in WBT is heterogeneous across locations with some places experiencing a much higher increase of up to 3.10 C or about 15 percent hotter than the typical WBT. Figure 4 presents the distribution of WBT in degree Celsius (top panel) and percent increase (bottom panel). Figure 4. : Density of deviation in WBT Source: Original figure for this paper based on data from the World Bank Enterprise Surveys and the Copernicus Climate Change Service . Note: WBT = wet-bulb temperature. Figure 5 presents a binscatter plot of the WBT deviations from the long-term average and the logarithm of sales per worker. The WBT bins are indicated by the number of dots on the scatter. The grouping of temperature observations by bins suggests that observations within the same bin tend to experience similar 12 We use an arbitrary radius of 30 kilometers around the establishment, assuming a reasonable com- muting distance between the workplace and home. This takes into account the case that exposure to higher temperatures is not only at the firm’s premises but also en route to workplaces, residential areas, and other places. In a robustness exercise, we vary the radius to see if the estimates hold. 14 effects, while those across different categories or bins tend to experience distinctly different impacts as shown by the strong negative correlation. In the next section, we estimate formal regression models to quantify the negative correlation between WBT and firm productivity. Figure 5. : Correlation between Sales per Worker and WBT Deviation Source: Original figure for this paper, based on data from the World Bank Enterprise Surveys and the Copernicus Climate Change Service. Note: WBT deviations are the difference between WBT in month m and year t and the long-term average within a radius of 30km around the establishment. PPP = purchasing power parity; WBT = wet-bulb temperature. HOTTER PLANET, HOTTER FACTORIES 15 IV. Empirical Strategy We estimate a non-linear regression of the log of sales per worker and other mea- sures of firm performance on area-level deviations in temperatures. Our identifica- tion strategy exploits the variation in area-level temperatures near the establish- ment. In a robustness exercise, we narrow the 30km radius of our main mode, to ≈27.5km, which is the minimum grid size for both the temperatures and relative humidity data. The variation in area-level temperatures is plausibly exogenous as they are not the result of the firms’ activities. However, there are various sources of bias that we need to address, the main source arising from potentially omit- ted variables. In such cases, the deviations in area-level temperatures could be picking up variations in other observed and unobserved area-level factors, which are likely to be correlated with both temperatures and firm productivity. These include area-level sub-national patterns in economic activities, transportation ac- tivities, population growth, environmental degradation or restoration such as de- forestation or afforestation, and so forth. Some of these factors are time-varying and potentially correlated with temperatures. The variation in these factors could also differ by administrative levels in that some could vary at the city level, while others may vary at the sub-national, country, or continental level. We control for these factors using country, sub-national, and World Bank global region-level fixed effects. We also interact the geographic fixed effects with year-fixed effects. The basic linear regression model that estimates log sales per worker on deviation in WBT controlling for firm-level characteristics and sub-national, country, and regional factors is given by: (3) ln(yijkc,t ) = β ∆W BT jkc,t + γXijkc,t + θkc + ρt + θkc · ρt + ϵijkc,t , where yijkc,t denotes firm-level productivity measured by the log of sales per worker of firm i in area j of sub-national region k , country c, and year t. ∆W BT jk,t is the deviation in average annual near-surface WBT from the long-term av- erage within a radius of 30km around the establishment (that is, area j ) and sub-national region k , country c and year t. Xijkc,t is a vector of firm-level char- acteristics, θkc is the sub-national region fixed effect, ρt is year fixed effect, θkc · ρt is the interaction term for sub-national and year fixed effects, ϵijkc,t is the inde- pendent and identically distributed error term, and β and γ are vectors of the coefficients to be estimated. All regressions are weighted using WBES weights. The specification in Equation 3 assumes that the effects on productivity are lin- ear. However, depending on the climate zone, and WBT categories, temperature increases could have heterogeneous effects on firm productivity. Kolstad and 16 Moore (2020) note that since the marginal effect of warming varies as a function of climate, a linear response function is often inappropriate for modeling the effect of climate change. For instance, in colder climate zones, a temperature increase could be beneficial to a firm’s productivity, whereas in hotter climate zones a slight increase in WBT could be detrimental to human health, negatively affect- ing firm productivity. We capture such non-linearity by interacting the deviation in WBT from the long-term average across four WBT categories. This allows for estimating nonlinear temperature effects across the different WBT groupings. The specification with non-linear effects of temperatures on productivity can be written as: Q2 ln(yijkc,t ) = β ∆W BT W BT jkc,t + βQ2 ∆jkc,t × Zonejkc + Q3 β Q3 ∆W BT jkc,t × Zonejkc + Q4 (4) β Q4 ∆W BT jkc,t × Zonejkc + ZoneQ 2 Q3 Q4 jkc + Zonejkc + Zonejkc + γXijkc,t + θkc + ρt + θkc · ρt + ϵijkc,t , where ZoneQ 2 Q3 Q4 jkc , Zonejkc , and Zonejkc are dummy variables indicating the cli- mate zones based on the quartiles of WBT. All the other notation is defined as for Equation 3. The reference category that is omitted from the regression is the bottom quartile or the coldest WBT zone: ZoneQ 1 jkc . The coefficients of interest are βQ2 , βQ3 , βQ4 , which can be interpreted as the marginal effect of a unit devi- ation in WBT from the long-term average on firm productivity in the respective WBT category, compared to the reference group. The only assumptions required here are that the temperature shocks are exogenous to each firm and the impact on productivity of the temperature deviation from the long-term average remains constant within each category. There are two fundamental sources of variations that are important for interpreting the impacts and central for estimating the non-linear impact of climate change. These are the variation in the temperature anomalies across climate zones, which provides the shock to our empirical model, and the variation in the baseline temperatures or climate zones, which underpins the nonlinearity of the impact of climate change on productivity. V. Results and Discussion We examine the effect of temperature deviations from the long-term average on labor productivity by estimating a non-linear model of our productivity indicator on deviations of WBT from the long-term average, across four WBT categories HOTTER PLANET, HOTTER FACTORIES 17 Table 2—: Pooled OLS estimation on sales per worker: Whole Sample (1) (2) (3) (4) (5) (6) (7) (8) WBT Deviation×ZoneQ 2 (23.70 C,24.90 C ] 0.526∗∗∗ 0.520∗∗∗ 0.199∗∗∗ 0.210∗∗∗ 0.234∗∗∗ 0.231∗∗∗ 0.224∗∗∗ 0.224∗∗∗ (0.028) (0.028) (0.026) (0.026) (0.027) (0.027) (0.027) (0.027) WBT Deviation×ZoneQ 3 (24.90 C,25.70 C ] -0.136∗∗∗ 0.103∗∗∗ -0.075∗∗ -0.060∗ -0.073∗∗ -0.031 -0.040 -0.041 (0.036) (0.036) (0.036) (0.036) (0.037) (0.037) (0.037) (0.037) WBT Deviation×ZoneQ 4 (25.70 C,max) -1.330∗∗∗ -1.060∗∗∗ -0.205∗∗∗ -0.169∗∗∗ -0.225∗∗∗ -0.222∗∗∗ -0.228∗∗∗ -0.234∗∗∗ (0.059) (0.058) (0.059) (0.059) (0.059) (0.059) (0.059) (0.059) Sector dummy No Yes Yes Yes Yes Yes Yes Yes Country FE No No Yes Yes Yes Yes Yes Yes Year FE No No No Yes Yes Yes Yes Yes Country x year FE No No No No No Yes Yes Yes Population (25km radius) No No No No No Yes Yes Yes Road infrastructure (25km radius) No No No No No No Yes Yes Pollution (25km radius) No No No No No No No Yes Observations 88,876 86,467 86,467 86,467 86,467 86,467 86,467 86,467 R2 0.057 0.146 0.357 0.359 0.370 0.371 0.373 0.373 Note: Standard errors are in parentheses. All models control for firm age, ownership, export status, firm size (employment), proportion of permanent workers, proportion of skilled workers, 55 dummy variables indicating a narrow industry classification, and dummies indicating World Bank regions. All WBT categories are included separately in the model, in addition to the interaction terms. The reference category for climate zone is the bottom quartile in the distribution of WBT which is less than or equal to 23.70 C . The minimum WBT in our sample is 14.50 C. FE = fixed effect; km = kilometers; OLS = ordinary least squares; WBT = wet-bulb temperature. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1 representing different climate zones. Following Equation 413 the coefficients on the interaction term between the climate zones and the deviations in temperature capture the nonlinear impacts of temperatures that vary across the climate zones. The marginal effects are obtained by taking the partial derivatives of log sales per worker with respect to climate zone (ZoneQq ) and WBT, where q = {1, 2, 3, 4} represents the quartile. The coefficient estimates (βQq ) can then be interpreted as the impact on productivity of a unit (10 C ) deviation in WBT from the long- term average compared to firms in the coldest zone.14 Table 2 presents the main results. Table 8 in the appendix presents the complete results. Specification (1) presents the results for the baseline model with no controls or essential fixed effects, and hence they are less reliable. The specification ignores many factors, both observable and unobservable, that could bias our estimates, which could be captured by a set of controls and fixed effects. To address this, we estimate a series of models by including key controls that potentially explain 13 We also estimate the basic model specified in Equation 3 and including the squared value of WBT deviation. The results are shown in Table 7, in the appendix. 14 The marginal effects are obtained by taking the partial derivatives of log sales per worker with Qq ∂ (ln(yijkc,t )) ∂ (Zonejkc ) respect to climate zone (ZoneQq ) and WBT i.e., Qq · . ∂ (Zonejkc ) ∂ (∆W BT ) jkc,t 18 variations in productivity, including firm characteristics, such as ownership type, export participation, firm age and size, and industry classification. We include fixed effects to account for unobserved factors that could confound our estimates, including country, year, country and year, and sector fixed effects. Specifications (2) to (8) sequentially add additional fixed effects. We also control for other poten- tial factors that could explain variations in firm productivity and correlate with WBT, including population density to proxy and account for market demand, road infrastructure to account for the role of local infrastructure in productivity, and air pollution (fine particulate matter, pm2.5)15 . Specification (8) is the full specification, and provides the most reliable estimates. A. Nonlinear Impacts The results show that deviation in WBT has a strong uneven and non-linear impact on firm productivity across the different climate zones. The coefficients on the interaction term between the deviation in WBT in the relatively mild temperature category of ZoneQ 2 (23.70 C,24.90 C ] are consistently positive and strongly significant at 1 percent level. This suggests that an increase in temperature in colder climate zones may provide favorable conditions for raising productivity. Specifically, a unit increase in WBT from the long-term local average leads to an increase in productivity by as much as 20.9 percent relative to the reference group located in the colder climate zone of ZoneQ 1 ≤23.70 C .16 However, the coefficient changes from positive to negative for firms located in climate zones with WBT ranging between 24.90 C and 25.70 C ., although it is insignificant at this level. The change in the direction of the coefficient seems to be an inflection point beyond which an increase in temperature has a detrimental impact on firm productivity. The biomedical literature suggests that a WBT of 250 C is considered the thresh- old above which the human body starts to experience adverse impacts, including reduced work capacity (Kjellstrom et al., 2009c). The results seem to corroborate this. A unit increase in WBT in the hottest climate zone of 25.70 C and above results in a productivity decline of about 20.8 percent compared to firms in the coldest climate zone. The coefficient is statistically significant at the 1 percent confidence interval. The effect of an increase in temperature from the long-term average depends on the climate zone. The effect exhibits non-linearity with firms in hotter zones experiencing steeper losses in productivity in times of higher tem- peratures compared to firms in relatively colder zones. This finding is mostly consistent with earlier studies (Hsiang, 2010; Dell et al., 2012; Burke et al., 2015), although at different thresholds of temperatures and with much larger impacts. Although the literature on the effects of climate change on firm-level productivity 15 The complete table is presented in the appendix. 16 For the semi-log specification, we calculate the percent change as (exp(β ) − 1) ∗ 100. HOTTER PLANET, HOTTER FACTORIES 19 in the non-agriculture sector is still developing, relatively recent studies document estimates that are comparable to those from this study. Using data from half a million manufacturing firms in China, Zou and Zhong (2022) find that a day with a temperature above 900 F (320 C) is associated with a TFP loss of 0.56 percent, relative to a day with a temperature between 500 F (100 C) and 600 F (15.60 C). In a study of Indian manufacturing firms, Somanathan et al. (2021) find that plant output falls by about 2 percent per 10 C increase in temperature. In a US study, Deryugina and Hsiang (2014) show that the productivity of individual days de- clines by 1.7 percent for each 10 C (1.80 F) increase in daily average temperature above 150 C (590 F). Using data from personnel records, Cai et al. (2018) find that productivity at temperatures below 600 F (15.60 C) is about 11% less than that at 750 F and 790 F. The productivity losses associated with higher temperatures that we find in this study are relatively larger compared to these and other similar studies. The main reason is that our study identifies nonlinear impacts at the higher end of the temperature distribution, while impacts at the lower end could even be positive, as we have shown. Hence, the average effects that are often reported may hide these important uneven effects. In addition, the thresholds for the temperature categories that register negative and large impacts are much higher than the thresholds used in many of the earlier studies and are higher than the human body can tolerate. There are large variations across temperature zones and income groups and additional variations that yield values that are dif- ferent from those of other studies that focus on individual countries. Hence, the estimates we show here are reasonable estimates of the impacts on productivity of changes in temperature deviations from the long-run average. The relatively large coefficients must be interpreted in the context of a slow (tak- ing several decades) increase in temperatures, in that a unit (10 C ) increase in the long-term average is quite large. The average deviation for the pooled obser- vations in our sample is 0.620 C over four decades, between 1982 and 2022. The magnitude of these temperature anomalies varies by climate zones, and regions of the world, and the rate of increase has been faster in recent decades. Accord- ing to US National Aeronautics and Space Administration’s National Center for Environmental Information, for instance, in each decade since 1981, the annual dry-bulb temperatures have increased by 0.270 C in North America, 0.220 C in South America, 0.460 C in Europe, 0.280 C in Africa, 0.370 C in Asia, and 0.190 C in Oceania. Although not directly comparable to dry-bulb temperatures, the av- erage WBT deviations in our sample, which covers almost four decades, are closer to these continental averages. B. Heterogeneity Although they are not based on linear relationships, the estimates in the previous section are averages, so they mask a great deal of heterogeneity. The effects of increased temperatures on firm productivity could vary depending on various ob- 20 served and unobserved factors. In this section, we perform heterogeneity analysis by firm size, industry subsamples, income groupings, and world regions. For the heterogeneity analysis by firm size, we split the sample into three groups in line with the predominant literature: small enterprises (5-19 workers), medium-sized enterprises (20-99 workers), and large enterprises (100+ workers). For the hetero- geneity analysis by industry classification, we use the broad industry classification of manufacturing and services. To examine the heterogeneity by income, we use the 2023 World Bank country classification of income into low, middle, and high. Firm Size Table 3 presents the results of the estimation of the effect of the deviation in WBT on log sales per worker for each sub-sample by firm size. The results are mixed. Although we do not detect a statistically significant negative effect on medium- sized firms located in the two hottest climate zones, there are strong and negative impacts on both small and large firms in the hottest climate zone. For small firms, a unit increase in WBT from the long-term average in the ZoneQ 4 (25.70 C,max) group reduces productivity by 33.7 percent, compared to the reference group (coldest re- gion). The productivity loss for large firms in the same WBT zone is almost three times that for small firms. Large firms experience significantly large productivity losses in all the temperature categories, with the largest losses for firms in the hottest climate zone. The large and negative effects on the productivity of large firms are evident in the various specifications and suggest that the general impact that we observed earlier was driven largely by these negative effects on large firms. There are several plausible explanations, for why large firms are more prone to such effects compared to small or medium sized firms. First, large firms tend to be particularly concentrated in the manufacturing sector. About 69 percent of large firms are in manufacturing compared to 60 percent of medium-sized firms and 46 percent of small firms. Within the large manufacturing firms, a significant share is in metals and metal products; chemicals, plastics, and rubber; and textiles, garments, and leather. Among small firms, a large share is in retail, and accom- modations; food and beverages, hospitality, and other services. Table 9 in the appendix presents a tabulation of the various activities in each size classification. Some of these activities might not require major physical labor that could put the human body at risk of heat stress and exhaustion during hotter days. The main reason for the distinct effects we observe across firm sizes seems to arise from the differences in exposure to excess heat across various industries, with varying effects on workers depending on work intensity. We discuss this below in the context of the heterogeneity by sector. In addition to the different sets of activities, smaller firms could be resilient in HOTTER PLANET, HOTTER FACTORIES 21 adapting to changes in temperature for example by changing working hours since they often have greater flexibility in the duration and timing of their active work hours. Compared to large firms, smaller firms face less stringent safety and work- ing standard requirements, particularly for working conditions in hot climates. In contrast, large firms could face stringent requirements on working conditions, which implies that they require higher capital investments in climate adaptations including installing indoor climate controls, provision of air-conditioned trans- portation, and other facilities, all of which contribute to raising the costs of pro- duction and reducing productivity per worker. Although we observe a clearly distinct impact between small and large firms, the distinction is blurred with medium-sized firms, due to the mix of industries in which they are present. Table 3—: Estimation of Log Sales per Worker: By Firm Size Small Medium Large (<20) (20-99) (100+) WBT Deviation×ZoneQ 2 (23.70 C,24.90 C ] -0.032 -0.200∗∗∗ -0.557∗∗∗ (0.042) (0.046) (0.063) WBT Deviation×ZoneQ 3 (24.90 C,25.70 C ] -0.121 -0.090 -0.526∗∗∗ (0.074) (0.083) (0.118) WBT Deviation×ZoneQ 4 (25.70 C,max) -0.411∗∗∗ 0.175 -0.741∗∗∗ (0.107) (0.131) (0.163) Sector dummy Yes Yes Yes Country FE Yes Yes Yes Year FE Yes Yes Yes Country x year FE Yes Yes Yes Population (25km radius) Yes Yes Yes Road infrastructure (25km radius) Yes Yes Yes Pollution (25km radius) Yes Yes Yes Observations 39,515 30,193 16,759 R2 0.386 0.387 0.392 Standard errors are in parentheses. All models control for firm age, ownership, export status, firm size (employment), proportion of permanent workers, proportion of skilled workers, 55 dummy variables indicating a narrow industry classification, and dummies indicating World Bank regions. All WBT categories are included separately in the model, in addition to the interaction terms. The reference category for climate zone is the bottom quartile in the distribution of WBT which is less than or equal to 23.70 C . The minimum WBT in our sample is 14.50 C. FE = fixed effect; km = kilometers; OLS = ordinary least squares; WBT = wet-bulb temperature. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1 22 Table 4—: British Standards Institution: Categories of Work Intensity: By Sector Sector Intensity Category Agriculture, forestry and fishing 355 Moderate/high (4) Other industry 295 Moderate (3) Manufacturing 240 Light/moderate (2) Construction 355 Moderate/high (4) Wholesale and retail trade 240 Light/moderate (2) Information and communication 180 Light (1) Financial and insurance activities 180 Light (1) Public administration and defense 240 Light/moderate (2) Note: Adopted from Costa et al. (2016) using a classification of different job activities by work intensity used by the British and European standards on heat stress, BS EN 27243:1994, and based on the International Organization for Standardization 7243:1989 (see British Standards Institution 1994). Sector The second dimension of heterogeneity is the sector classification encompassing manufacturing and retail and services. Exposure to excess heat has varying ef- fects on workers, depending on work intensity, which varies with the nature of economic activities. Work intensity in the manufacturing sector is generally con- sidered light to moderate by the British and European standards for heat stress. Table 4 presents the level of work intensity of jobs by sector. The classification is based on the 1989 International Organization for Standards estimation of the effect of heat stress on working men (Costa et al., 2016). Work intensity in the agriculture, forestry, fishing and construction sectors is generally considered mod- erate to highly intensive. Manufacturing and wholesale and retail trade industries are considered to have light to moderate work intensity. We unpack the potential heterogeneity in the effects of exposure to excess heat by industry by estimat- ing the model for two broad industry classifications: manufacturing and services, where services include retail and other services. Manufacturing firms account for about 57 percent of the WBES data and comprise a variety of firms engaged in productive activities, from wood and furniture products to manufacturing ma- chinery, electronics, and communications equipment. Retail and services account for the remaining 43 percent of our sample. Firms in retail are those engaged in construction, retail, hotels, and hospitality services. Other services account for about 21 percent of the sample and include firms engaged in wholesale and other services. Table 5 presents the results by broad sector classification of manufacturing, and services (+retail). The effect varies by sector, with negative and statistically sig- nificant effects on the productivity of manufacturing firms located in both temper- ature categories where the WBT is above 24.9, while the results for services are not HOTTER PLANET, HOTTER FACTORIES 23 consistent. For manufacturing firms located in climatic zones ZoneQ 3 (24.90 C,25.70 C ] and ZoneQ 4 (25.70 C,max) , a 10 C increase in deviations from the long-term average is associated with productivity losses of 36 and 26 percent respectively, compared to the reference group. The impacts on the productivity of firms in the service sector are negative and significant in ZoneQ 2 (23.70 C,24.90 C ] and ZoneQ 4 (25.70 C,max) categories but not in the middle group, suggesting additional non-linearity impact as the temperature increases across the various zones. These effects in the manufactur- ing sector are larger than the effects we observe in the baseline model, suggesting that manufacturing firms explained a larger share of the productivity losses in our baseline result. These results suggest that the effects are heterogeneous and non-linear, varying by industry and temperature zones. Table 5—: Estimation of Log Sales per Worker: By Sector Manufacturing Services WBT Deviation×ZoneQ 2 (23.70 C,24.90 C ] -0.014 -0.185∗∗∗ (0.031) (0.046) WBT Deviation×ZoneQ 3 (24.90 C,25.70 C ] -0.446∗∗∗ 0.007 (0.059) (0.079) WBT Deviation×ZoneQ 4 (25.70 C,max) -0.301∗∗∗ -0.405∗∗∗ (0.093) (0.114) Narrow sector dummy Yes Yes Country FE Yes Yes Year FE Yes Yes Country x year FE Yes Yes Population (25km radius) Yes Yes Road infrastructure (25km radius) Yes Yes Pollution (25km radius) Yes Yes Observations 48,521 38,135 R2 0.392 0.375 Standard errors are in parentheses. All models control for firm age, ownership, export status, firm size (employment), proportion of permanent workers, proportion of skilled workers, 55 dummy variables indicating a narrow industry classification, and dummies indicating World Bank regions. All WBT categories are included separately in the model, in addition to the interaction terms. The reference category for climate zone is the bottom quartile in the distribution of WBT which is less than or equal to 23.70 C . The minimum WBT in our sample is 14.50 C. FE = fixed effect; km = kilometers; OLS = ordinary least squares; WBT = wet-bulb temperature. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1 24 Income Grouping A key departure of this study from earlier studies is its use of a globally compa- rable database of nonagricultural firms across climate zones, regions, and income groups. This presents an opportunity to estimate the potentially disparate im- pacts across not only temperature zones, but also income groups. A central ques- tion in the development discourse is how climate change affects the development trajectory of poorer countries and whether the effects are disproportionate across income groups. To check this, we estimate the model for subsamples based on income groups. We estimate nonlinear impacts for two sub-samples, one includ- ing only low-income economies and another including middle and high-income economies. Table 6 presents the estimation results, which indicate that hotter temperatures have a significant impact on firm productivity. In low-income countries, firms in the highest WBT category, ZoneQ 4 (25.70 C,max) or the hottest climate zone experience productivity losses of about 50.7 percent, the largest we find across all our specifications. The coefficient for this temperature category is not significant for the middle- and high-income sample, however, sug- gesting no impact, at the hottest climate zone. However, firms in the middle and high-income sample also experience productivity losses in the relatively colder cli- mate zones, ZoneQ 2 (23.70 C,24.90 C ] and ZoneQ 3 (24.90 C,25.70 C ] categories of about 17.5 and 19 percent respectively, compared to the reference group. Our results suggest a relatively large disproportionate impact of climate change as firms in poorer coun- tries in the hottest climate zones register the largest productivity losses (50.7 per- cent), while those in high-income countries do not experience productivity losses due to climate change, at least in the hotter WBT zone, ZoneQ 4 (25.70 C,max) . Caution is required in interpreting these results, because the distribution of countries by income in some cases may follow the geography of climate zones. Most low income countries may be in the hotter climate zones while many of the non-low income countries may be outside the hotter climate zones. Hence the observed variation in impact particularly across the WBT categories may reflect the distribution of countries in each WBT category, rather than a comparative representation of low and high income countries in the same climate zone. Overall, however, this find- ing complements existing studies that show the uneven impacts of climate change across income groups. For example, Dell et al. (2012) show that the largest neg- ative effects of temperature changes are observed in the lowest income quintile, and positive impacts of temperature changes have been observed in the richer countries. The implications of the uneven impact of temperature changes on in- comes and global inequality are significant. These findings suggest that climate change could redefine and put into question traditionally established frameworks in development including the concept of convergence. For the poorer countries in already hot climates, climate change poses significant risks to economic growth and the prospects of catching up to richer countries, even after controlling for HOTTER PLANET, HOTTER FACTORIES 25 Table 6—: Estimation of Log Sales per Worker: By Income Group Low income Middle and high income WBT Deviation×ZoneQ 2 (23.70 C,24.90 C ] 0.091∗ -0.192∗∗∗ (0.054) (0.040) WBT Deviation×ZoneQ 3 (24.90 C,25.70 C ] -0.114 -0.211∗∗∗ (0.115) (0.066) WBT Deviation×ZoneQ 4 (25.70 C,max) -0.707∗∗∗ 0.038 (0.083) (0.133) Sector dummy Yes Yes Country FE Yes Yes Year FE Yes Yes Country x year FE Yes Yes Population (25km radius) Yes Yes Road infrastructure (25km radius) Yes Yes Pollution (25km radius) Yes Yes Observations 43,375 42,931 R2 0.195 0.300 Standard errors are in parentheses. All models control for firm age, ownership, export status, firm size (employment), proportion of permanent workers, proportion of skilled workers, 55 dummy variables indicating a narrow industry classification, and dummies indicating World Bank regions. All WBT categories are included separately in the model, in addition to the interaction terms. The reference category for climate zone is the bottom quartile in the distribution of WBT which is less than or equal to 23.70 C . The minimum WBT in our sample is 14.50 C. FE = fixed effect; km = kilometers; OLS = ordinary least squares; WBT = wet-bulb temperature. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1 other factors. Summary of Heterogeneity To sum up, our analysis of heterogeneity suggests that the results vary signifi- cantly by firm size, sector classification, and income group. The uneven impacts of climate change that we observe across temperature zones are not uniform across these categories. Large firms, firms in manufacturing, and those in low-income countries and hotter climate zones tend to experience the biggest productivity losses due to climate change. Given that many low-income countries are in hotter climate zones, the impacts of climate change due to higher temperatures are fur- ther exacerbated by limited capabilities to invest in adaptation. Firms in poorer regions experience the highest losses due to climate change, especially in the 26 hottest WBT categories. This reinforces what we know about climate change re- inforcing existing vulnerabilities in the regions of the world that are least capable of responding to its effects. HOTTER PLANET, HOTTER FACTORIES 27 VI. Conclusion Using matched climate and firm data covering 154 countries, we documented that the effects of rising temperature are non-linear and uneven across climate zones. Firms in hotter zones experience steeper losses in productivity with increases in temperature compared to firms in relatively colder zones, which tend to register productivity gains. A unit (10 C ) increase in WBT in the hottest climate zone of 25.70 C and above results in a productivity decline of about 20.8% compared to firms in the coldest climate zone. There is a change in the direction of im- pact or sign of the main coefficient, indicating a potential inflection point beyond which an increase in temperature has a detrimental impact on firm productivity, which was estimated to be 25.70 C in our sample. In addition, the effects vary not only based on the temperature categories within which firms are located, but also on other factors such as firm size, industry classification, income group and region. For example, large firms, firms in manufacturing and those in low-income countries and hotter climate zones tend to experience the biggest productivity losses due to climate change. Given that many low income countries are in hotter climate zones, the climate change impacts due to higher temperature are further exacerbated by limited capabilities to invest in adaptation. Firms in poorer re- gions experience the highest productivity losses due to climate change especially those in the hottest temperature categories. This reinforces what we know about climate change reinforcing existing vulnerabilities in the regions of the world that are least capable of responding and adapting to climate change. This study builds on and makes important contributions to the emerging lit- erature. First, using a globally comparable and standardized survey of firms rather than an individual country presents opportunities for better understand- ing the potentially heterogeneous and uneven impacts of temperature changes on firm outcomes. Most earlier studies that examined climate change impacts using micro-data relied on individual country observations due to limitations of data comparability across countries and regions especially at the micro level. The survey data also allows for estimation of impacts across regions, climate zones, industries, and incomes. This structure enables estimation of the impact at the global level. Comparison of these impacts is essential given the global nature of climate change and the potentially distinct types and magnitudes of the impacts. This may be the only study that estimates the impact of temperature changes on productivity at the widest geographical coverage using a representative sam- ple of firms in more than 150 countries, providing the most comprehensive study in the literature. In addition, despite the relatively rich literature on the im- pacts of climate change on agriculture, studies on the non-agriculture sector are relatively scarce. This study contributes to understanding the impact on the non- agriculture sector, using firm-level data. 28 By providing more granular evidence from 154 countries, the findings of this study have essential implications for current national and global policy discourse on the costs of carbon, the distribution of gains and losses and the distribution of re- sponsibilities and contributions to mitigate climate change. The findings reveal uneven impacts, where firms in already hotter regions, in the tropics and in low income countries experience steeper losses in productivity, while there may be gains in colder zones. This suggests that climate change is reinforcing global in- come inequalities. The reduction in global inequality during the 1980’s and 1990’s (Sala-i Martin, 2006) is at risk. If the trends in global warming are not reversed over the coming decades, there is a heightened risk of widening inequalities across countries. The implications are especially dire for the poorest countries in the hottest regions. Climate change has also put into question the notion of conver- gence, in which the fast growth of low- and middle-income countries enables them to catch up to high-income economies. For low-income countries, growth will be even a more daunting challenge due to climate change. The results from the study have important implications for quantifying the im- pacts of climate change on productivity and economic growth, evaluating the costs and benefits associated with climate change, and determining the required contributions and allocation of funds for adaptation and climate action. The find- ings can be useful contributions to international climate negotiations and related policy dialogues. One caveat of the study is that we did not directly account for firm-level investments in adaptation that would have lessened the impact of increases in temperature on productivity. We used proxies that would only indi- rectly address this. This is an important area for future studies both in collecting more information related to investments in adaptation and in properly accounting for such investments. HOTTER PLANET, HOTTER FACTORIES 29 References Acevedo, S., Mrkaic, M., Novta, N., Pugacheva, E., and Topalova, P. (2020). The effects of weather shocks on economic activity: what are the channels of impact? Journal of Macroeconomics, 65:103207. Adhvaryu, A., Kala, N., and Nyshadham, A. (2020). The light and the heat: Productivity co-benefits of energy-saving technology. Review of Economics and Statistics, 102(4):779–792. on, F. M., Oteiza, F., and Rud, J. P. (2021). Climate change and agriculture: Arag´ Subsistence farmers’ response to extreme heat. American Economic Journal: Economic Policy, 13(1):1–35. Burke, M., Hsiang, S. M., and Miguel, E. (2015). Global non-linear effect of temperature on economic production. Nature, 527(7577):235–239. Cai, X., Lu, Y., and Wang, J. (2018). The impact of temperature on manufactur- ing worker productivity: evidence from personnel data. Journal of Comparative Economics, 46(4):889–905. Carter, C., Cui, X., Ghanem, D., and M´ erel, P. (2018). Identifying the eco- nomic impacts of climate change on agriculture. Annual Review of Resource Economics, 10:361–380. Chen, H.-Y. and Chen, C.-C. (2022). An empirical equation for wet-bulb temper- ature using air temperature and relative humidity. Atmosphere, 13(11):1765. Cline, W. R. (2007). Global warming and agriculture: Impact estimates by coun- try. Peterson Institute. Costa, H., Floater, G., Hooyberghs, H., Verbeke, S., De Ridder, K., et al. (2016). Climate change, heat stress and labour productivity: A cost methodology for city economies. Centre For Climate Change Economics And Policy London, UK. Cusolito, A. P. and Maloney, W. F. (2018). Productivity revisited: Shifting paradigms in analysis and policy. World Bank Publications. Dell, M., Jones, B. F., and Olken, B. A. (2012). Temperature shocks and economic growth: Evidence from the last half century. American Economic Journal: Macroeconomics, 4(3):66–95. Deryugina, T. and Hsiang, S. M. (2014). Does the environment still matter? daily temperature and income in the united states. Technical report, National Bureau of Economic Research. enes, O. and Greenstone, M. (2007). The economic impacts of climate Deschˆ change: evidence from agricultural output and random fluctuations in weather. American economic review, 97(1):354–385. 30 Fisher, A. C., Hanemann, W. M., Roberts, M. J., and Schlenker, W. (2012). The economic impacts of climate change: evidence from agricultural output and random fluctuations in weather: comment. American Economic Review, 102(7):3749–3760. Gupta, R. and Somanathan, E. (2022). The impact of temperature on worker absenteeism in the indian manufacturing sector. Heal, G. and Park, J. (2013). Feeling the heat: Temperature, physiology & the wealth of nations. Technical report, National Bureau of Economic Research. Houghton, J. (2004). Global warming: the complete briefing. Cambridge university press. Hsiang, S. M. (2010). Temperatures and cyclones strongly associated with eco- nomic production in the caribbean and central america. Proceedings of the National Academy of sciences, 107(35):15367–15372. International Labour Organization (2019). Working on a warmer planet: the impact of heat stress on labour productivity and decent work. Kjellstrom, T., Gabrysch, S., Lemke, B., and Dear, K. (2009a). The ‘hothaps’ programme for assessing climate change impacts on occupational health and productivity: an invitation to carry out field studies. Global health action, 2(1):2082. Kjellstrom, T., Holmer, I., and Lemke, B. (2009b). Workplace heat stress, health and productivity–an increasing challenge for low and middle-income countries during climate change. Global health action, 2(1):2047. Kjellstrom, T., Kovats, R. S., Lloyd, S. J., Holt, T., and Tol, R. S. (2009c). The direct impact of climate change on regional labor productivity. Archives of Environmental & Occupational Health, 64(4):217–227. Klein, R. J., Midglev, G., Preston, B., Alam, M., Berkhout, F., Dow, K., and Shaw, M. (2014). Climate change 2014: impacts, adaptation, and vulnerability. IPCC fifth assessment report, Stockholm, Sweden. Kolstad, C. D. and Moore, F. C. (2020). Estimating the economic impacts of cli- mate change using weather observations. Review of Environmental Economics and Policy. Lai, W., Qiu, Y., Tang, Q., Xi, C., and Zhang, P. (2023). The effects of temper- ature on labor productivity. Annual Review of Resource Economics, 15. Lemke, B. and Kjellstrom, T. (2012). Calculating workplace wbgt from meteoro- logical data: a tool for climate change assessment. Industrial Health, 50(4):267– 278. HOTTER PLANET, HOTTER FACTORIES 31 LoPalo, M. (2023). Temperature, worker productivity, and adaptation: evidence from survey data production. American Economic Journal: Applied Economics, 15(1):192–229. ean, C., Berger, S., Masson-Delmotte, V., Zhai, P., Pirani, A., Connors, S. L., P´ Caud, N., Chen, Y., Goldfarb, L., Gomis, M., et al. (2021). Climate change 2021: the physical science basis. Contribution of working group I to the sixth assessment report of the intergovernmental panel on climate change, 2. Mendelsohn, R., Nordhaus, W. D., and Shaw, D. (1994). The impact of global warming on agriculture: a ricardian analysis. The American economic review, pages 753–771. Mora, C., Dousset, B., Caldwell, I. R., Powell, F. E., Geronimo, R. C., Bielecki, C. R., Counsell, C. W., Dietrich, B. S., Johnston, E. T., Louis, L. V., et al. (2017). Global risk of deadly heat. Nature climate change, 7(7):501–506. Sabater, J. M. (2019). Era5-land monthly averaged data from 1981 to present. Copernicus Climate Change Service (C3S) Climate Data Store (CDS), 10. Sala-i Martin, X. (2006). The world distribution of income: falling poverty and. . . convergence, period. The Quarterly Journal of Economics, 121(2):351–397. Schlenker, W., Hanemann, W. M., and Fisher, A. C. (2006). The impact of global warming on us agriculture: an econometric analysis of optimal growing conditions. Review of Economics and statistics, 88(1):113–125. Somanathan, E., Somanathan, R., Sudarshan, A., and Tewari, M. (2021). The impact of temperature on productivity and labor supply: Evidence from indian manufacturing. Journal of Political Economy, 129(6):1797–1827. Stern, N. (2008). The economics of climate change. American Economic Review, 98(2):1–37. Tol, R. S. J. (2009). The economic effects of climate change. Journal of economic perspectives, 23(2):29–51. World Bank (2013). Turn down the heat: Climate extremes, regional impacts, and the case for resilience. Zou, H. and Zhong, M.-R. (2022). Factor reallocation and cost pass-through under the carbon emission trading policy: Evidence from chinese metal industrial chain. Journal of Environmental Management, 313:114924. 32 Appendix A: Additional Tables and Figures HOTTER PLANET, HOTTER FACTORIES 33 Table 7—: Simple pooled OLS estimation on sales per worker: Whole Sample (1) (2) (3) (4) (5) (6) WBT Deviation 0.115∗∗∗ 0.083∗∗∗ 0.210∗∗∗ 0.158∗∗∗ 0.049 0.148∗∗∗ (0.012) (0.029) (0.031) (0.031) (0.035) (0.035) WBT Deviation squared 0.018 0.012 0.039∗∗ 0.087∗∗∗ 0.057∗∗∗ (0.016) (0.016) (0.016) (0.016) (0.016) Firm size: medium(20-99) 0.090∗∗∗ 0.127∗∗∗ 0.119∗∗∗ 0.114∗∗∗ 0.112∗∗∗ (0.012) (0.011) (0.011) (0.011) (0.011) Firm size: large(100 and over) -0.047∗∗ 0.076∗∗∗ 0.065∗∗∗ 0.054∗∗∗ 0.057∗∗∗ (0.022) (0.019) (0.019) (0.019) (0.019) Ownership: Foreign 0.340∗∗∗ 0.265∗∗∗ 0.258∗∗∗ 0.256∗∗∗ 0.234∗∗∗ (0.025) (0.021) (0.021) (0.021) (0.021) Exporter 0.601∗∗∗ 0.349∗∗∗ 0.349∗∗∗ 0.358∗∗∗ 0.354∗∗∗ (0.017) (0.015) (0.015) (0.015) (0.015) Firm Age 0.011∗∗∗ 0.003∗∗∗ 0.004∗∗∗ 0.004∗∗∗ 0.004∗∗∗ (0.0004) (0.0003) (0.0003) (0.0003) (0.0003) Permanent workers (%) -0.010∗∗∗ -0.004∗∗∗ -0.004∗∗∗ -0.004∗∗∗ -0.004∗∗∗ (0.0005) (0.0004) (0.0004) (0.0004) (0.0004) Skilled workers (%) -0.003∗∗∗ -0.0004 -0.0003 -0.0005∗ -0.0005∗ (0.0003) (0.0003) (0.0003) (0.0003) (0.0003) Dummy: Broad Services -0.387 0.811 0.906 0.829 0.795 (0.670) (0.581) (0.581) (0.582) (0.580) log(Population) 0.029∗∗∗ (0.004) log(Highway in km) 0.037∗∗∗ (0.003) pm2 -0.0004 (0.001) Sector Dummy No Yes Yes Yes Yes Yes Country FE No No Yes Yes Yes Yes Year FE No No No Yes Yes Yes Country X Year FE No No No No No Yes N 88,876 86,467 86,467 86,467 86,467 86,467 R2 0.057 0.146 0.357 0.359 0.370 0.373 Note: Standard errors are in parentheses. All models control for 55 dummy variables indicating a narrow industry classification, and dummies indicating World Bank regions. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1 34 Table 8—: Pooled OLS estimation on sales per worker: Whole Sample (1) (2) (3) (4) (5) (6) (7) (8) WBT Deviation 0.071∗∗∗ -0.063∗∗∗ 0.188∗∗∗ 0.178∗∗∗ 0.168∗∗∗ 0.167∗∗∗ 0.190∗∗∗ 0.189∗∗∗ (0.021) (0.022) (0.021) (0.021) (0.022) (0.022) (0.022) (0.022) ZoneQ 2 (23.70 C,24.90 C ] 0.205∗∗∗ 0.052∗∗ -0.111∗∗∗ -0.113∗∗∗ -0.112∗∗∗ -0.117∗∗∗ -0.113∗∗∗ -0.115∗∗∗ (0.024) (0.024) (0.024) (0.024) (0.024) (0.024) (0.024) (0.024) ZoneQ 3 (24.90 C,25.70 C ] 0.714∗∗∗ 0.369∗∗∗ 0.098∗∗∗ 0.094∗∗∗ 0.109∗∗∗ 0.055∗∗ 0.055∗∗ 0.053∗∗ (0.022) (0.023) (0.023) (0.023) (0.023) (0.024) (0.024) (0.024) ZoneQ 4 (25.70 C,max) 0.383∗∗∗ 0.044 0.096∗∗∗ 0.084∗∗∗ 0.104∗∗∗ 0.070∗∗ 0.061∗∗ 0.059∗ (0.028) (0.029) (0.030) (0.030) (0.031) (0.031) (0.031) (0.031) WBT Deviation×ZoneQ 2 (23.70 C,24.90 C ] 0.526∗∗∗ 0.520∗∗∗ 0.199∗∗∗ 0.210∗∗∗ 0.234∗∗∗ 0.231∗∗∗ 0.224∗∗∗ 0.224∗∗∗ (0.028) (0.028) (0.026) (0.026) (0.027) (0.027) (0.027) (0.027) WBT Deviation×ZoneQ 3 (24.90 C,25.70 C ] -0.136∗∗∗ 0.103∗∗∗ -0.075∗∗ -0.060∗ -0.073∗∗ -0.031 -0.040 -0.041 (0.036) (0.036) (0.036) (0.036) (0.037) (0.037) (0.037) (0.037) WBT Deviation×ZoneQ 4 (25.70 C,max) -1.330∗∗∗ -1.060∗∗∗ -0.205∗∗∗ -0.169∗∗∗ -0.225∗∗∗ -0.222∗∗∗ -0.228∗∗∗ -0.234∗∗∗ (0.059) (0.058) (0.059) (0.059) (0.059) (0.059) (0.059) (0.059) Firm size: medium(20-99) 0.132∗∗∗ 0.130∗∗∗ 0.121∗∗∗ 0.117∗∗∗ 0.116∗∗∗ 0.114∗∗∗ 0.114∗∗∗ (0.012) (0.011) (0.011) (0.011) (0.011) (0.011) (0.011) Firm size: large(100+) -0.016 0.073∗∗∗ 0.061∗∗∗ 0.051∗∗∗ 0.056∗∗∗ 0.052∗∗∗ 0.052∗∗∗ (0.021) (0.019) (0.019) (0.019) (0.019) (0.019) (0.019) Foreign ownership 0.351∗∗∗ 0.266∗∗∗ 0.259∗∗∗ 0.257∗∗∗ 0.236∗∗∗ 0.236∗∗∗ 0.236∗∗∗ (0.024) (0.021) (0.021) (0.021) (0.021) (0.021) (0.021) Exporter 0.567∗∗∗ 0.346∗∗∗ 0.347∗∗∗ 0.356∗∗∗ 0.356∗∗∗ 0.352∗∗∗ 0.352∗∗∗ (0.017) (0.015) (0.015) (0.015) (0.015) (0.015) (0.015) Firm age 0.010∗∗∗ 0.003∗∗∗ 0.004∗∗∗ 0.004∗∗∗ 0.004∗∗∗ 0.004∗∗∗ 0.004∗∗∗ (0.0004) (0.0003) (0.0003) (0.0003) (0.0003) (0.0003) (0.0003) Permanent workers (%) -0.011∗∗∗ -0.004∗∗∗ -0.004∗∗∗ -0.004∗∗∗ -0.004∗∗∗ -0.004∗∗∗ -0.004∗∗∗ (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) Skilled workers (%) -0.003∗∗∗ -0.0004 -0.0003 -0.001∗∗ -0.001∗ -0.001∗∗ -0.001∗∗ (0.0003) (0.0003) (0.0003) (0.0003) (0.0003) (0.0003) (0.0003) log(Population) 0.052∗∗∗ 0.026∗∗∗ 0.026∗∗∗ (0.003) (0.004) (0.004) log(Highway) 0.037∗∗∗ 0.037∗∗∗ (0.003) (0.003) Pollution (pm2 dev.) -0.001 (0.001) Sector Dummy No Yes Yes Yes Yes Yes Yes Yes Country FE No No Yes Yes Yes Yes Yes Yes Year FE No No No Yes Yes Yes Yes Yes Country X Year FE No No No No No Yes Yes Yes N 88,876 86,467 86,467 86,467 86,467 86,467 86,467 86,467 R2 0.057 0.146 0.357 0.359 0.370 0.371 0.373 0.373 Note: Standard errors are in parentheses. All models control for 55 dummy variables indicating a narrow industry classification, and dummies indicating World Bank regions. The reference category for climate zone is the bottom quartile in the distribution of WBT which is less than or equal to 23.70 C . The minimum WBT in our sample is 14.50 C. FE = fixed effect; km = kilometers; OLS = ordinary least squares; WBT = wet-bulb temperature. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1 HOTTER PLANET, HOTTER FACTORIES 35 Table 9—: Cross-tabulation of firm size, narrow industry classification, percent Small(< 20) Medium (20 − 99) Large(100+) (Percent) (Percent) (Percent) Accommodation 52.21 40.71 7.08 Basic Metals & Metal Products 28.69 38.82 32.49 Basic Metals/Fabricated Metals/Machinery & Equip. 39.37 41.12 19.51 Chemicals & Chemical Products 30.58 40.34 29.08 Chemicals, NonMetallic Mineral, Plastics & Rubber 39.24 47.47 13.29 Chemicals, Plastics & Rubber 31.90 37.61 30.49 Construction 40.51 40.18 19.31 Electronics 26.01 31.66 42.33 Electronics & Communications Equip. 23.25 41.84 34.91 Fabricated Metal Products 39.06 41.12 19.82 Food 39.95 36.22 23.84 Food and beverage service activities 68.00 28.00 4.00 Food/Leather/Wood/Tobacco/Rubber Products 42.86 34.69 22.45 Furniture 54.89 33.42 11.70 Garments 41.98 30.41 27.61 Hospitality & Tourism 56.52 35.49 7.98 Hotels & Restaurants 34.03 46.75 19.21 IT & IT Services 45.15 34.64 20.22 Leather Products 48.41 33.82 17.77 Machinery & Equipment 33.79 41.93 24.28 Machinery & Equipment & Electronics 25.62 26.60 47.78 Machinery & Equipment, Electronics & Vehicles 26.02 32.92 41.07 Manufacturing 42.03 35.82 22.15 Manufacturing Panel 64.17 28.35 7.48 Metals, Machinery, Computer & Electronics 64.10 25.64 10.26 Minerals, Metals, Machinery & Equipment 45.56 36.67 17.78 Mining Related Manufacturing 44.44 28.89 26.67 Motor Vehicles 18.79 46.92 34.28 Motor Vehicles & Transport Equip. 43.55 27.42 29.03 NonMetallic Mineral Products 43.32 37.28 19.40 Other Manufacturing 40.25 36.26 23.49 Other Services 52.42 32.30 15.28 Other Services Panel 59.65 34.65 5.70 Petroleum products, Plastics & Rubber 32.12 38.18 29.70 Pharmaceuticals & Medical Products 9.09 31.82 59.09 Printing & Publishing 58.23 30.38 11.39 Rest of Universe 47.35 35.60 17.05 Retail 64.02 25.65 10.33 Retail & IT 99.24 0.76 0 Retail Panel 85.90 12.82 1.28 Rubber & Plastics Products 31.66 40.89 27.45 Services 56.55 32.19 11.26 Services of Motor Vehicles 34.55 42.59 22.86 Services of Motor Vehicles/Wholesale/Retail 69.46 24.90 5.64 Textiles 28.71 38.00 33.29 Textiles & Garments 39.92 33.62 26.46 Textiles, Garments & Leather 60.34 35.34 4.31 Textiles, Garments, Leather & Paper 40.00 42.50 17.50 Transport 55.31 30.73 13.97 Transport, Storage, & Communications 47.84 37.65 14.51 Wholesale 65.24 28.30 6.46 Wholesale & Retail 58.45 32.22 9.33 Wholesale of Agri Inputs & Equipment 37.14 37.14 25.71 Wood Products 60.16 30.08 9.76 Wood Products & Furniture 40.74 39.51 19.75 Wood products, Furniture, Paper & Publishing 45.32 27.49 27.19 No. of obs. 84,620 60,965 34,482