Policy Research Working Paper 8834 The Effects of Pollution and Business Environment on Firm Productivity in Africa Maria E. Soppelsa Nancy Lozano-Gracia L. Colin Xu Social, Urban, Rural and Resilience Global Practice April 2019 Policy Research Working Paper 8834 Abstract This paper explores the links between city competitiveness results. First, the negative association between air pollution and air pollution and the business environment. Because and firm performance can be seen at lower than expected competitive cities not only attract more productive firms, levels of pollution. Second, the effects of capacity agglom- but also facilitate their business, the paper look at firm eration on labor productivity growth are stronger compared performance as a proxy for city competitiveness. It focuses to other regions. These findings suggest that cities in this on African firms, because this region is developing fast and region should address pollution issues soon, as they con- experiencing increasing pollution levels and the effects of tinue to grow fast and pollution levels are becoming an agglomeration economies. The analysis finds two interesting increasing concern. This paper is a product of the Social, Urban, Rural and Resilience Global Practice. 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 nlozano@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 The Effects of Pollution and Business Environment on Firm Productivity in Africa1 Maria E. Soppelsa, Nancy Lozano-Gracia and L. Colin Xu JEL: Q53, O1, L5, D9 1 This work is part of the Pollution Management and Environmental Health program (PMEH) supported under TF0A3415. The PMEH program focuses on providing technical assistance to these countries, facilitating pollution management, promoting public awareness, and helping them reduce the negative effects of pollution on health. The authors are grateful to Somik V. Lall and Richard Damania for their comments. 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. 1. Introduction2 Between 2000 and today, African economies have strengthened their position in the global economy, showing average annual growth rates of 5.9 percent between 2000 and 2008 (World Economic Forum 2009), and 5.2 percent in the last 10 years. 3 Furthermore, not only per capita GDP is growing, but also total factor productivity shows an average positive growth rates of just above 1% for the first time since the 1970s (Rodrik 2016). 4 At the same time, air pollution has climbed to dangerous levels, increasing annual deaths by 36 percent between 1990 and 2013 (Roy 2016). While some literature exists on the determinants of firm productivity for some African countries, to the best of our knowledge there is no comprehensive study that compares such determinants across regions, and none that links productivity at the firm level with air pollution measures for a sample of firms that spans all continents and regions. What are the key determinants of firm productivity in Africa and how do they compare to the rest of the world? Are there any links between firm productivity and air pollution? To shed light on these questions, we choose to focus on exploring the links between two sets of variables that may affect firm productivity: a group of business environment characteristics and air pollution. The business environment has been identified as one of the key factors that contribute to development (Stern 2002, World Bank 2005, World Bank 2010). It provides the framework where firms interact, trade, and compete. It includes not only the basic 2 This paper aims to explore how air pollution affects city competitiveness, particularly for African cities. Although different alternatives to measure city competitiveness have been explored by the literature (see for example Porter (1990), Faberberg (1996), Boltho (1996), World Bankl (2016)), we choose to look at firm productivity. Our decision is motivated by two arguments. First, firms that are more productive tend to locate in an environment where they can reach their maximum profit. They want to hire the best, most productive workers. They also want to have access to a good set of amenities and services. More productive firms will then cluster in more competitive cities. Second, cities that offer a good quality of life will be better positioned to attract firms and qualified workers. Since good quality of life and the standard of living are determined by the productivity of an economy, then productivity defines competitiveness (Porter, 2000, p.19). Moreover, he also points out that cities cannot become more productive and competitive unless firms become more productive. High city GDP, employment, and productivity (the three measures of competitiveness highlighted by the World Bank Report, 2016) are actually a consequence of how well firms do. This close relationship between city performance and firm performance hints that looking at firm productivity is a good proxy for measuring city competitiveness. 3 Based on national GDP growth from the Oxford Economics Dataset. 4 Even though there is some heterogeneity across the continent, the growth in total factor productivity ranges between 0.5% for the East Region and 2% for the Southern Region (with the exception of the North Region that shows a negative total factor productivity growth rate of -0.5%). 2 legal structure, but also other city characteristics that can affect firms’ performance, such as human capital or agglomeration economies.5 The same firm in a different environment will probably experience different challenges that can affect its productivity levels and outcomes. Air pollution has also been considered an important factor affecting workers and firms through two main channels. On the one hand, pollution levels can affect workers’ health and productivity. Recent studies on the effects of air pollution on worker productivity have quantified the negative effects on different production sectors (Graff-Zivin and Neidell 2012, T. Chang, J. S. Graff-Zivin, et al. 2014, T. Chang, J. Graff-Zivin, et al. 2016, Adhvaryu, Kala and Nyshadham 2014). On the other hand, environmental regulations can impact firms’ costs and influence their location. For example, studies such as Henderson (1974) and Kahn and Mansur, (2013) have shown that manufacturing firms cluster in areas where environmental regulations are not too strict. In developing countries, cities often limit such regulations, offering “pollution havens” that can attract major factories and contribute to development (Birdsall and Wheeler 1993, Cole 2004, Eskeland and Harrison 2003), while exacerbating the negative consequences on the environment. We build upon related work by Harrison, Lin and Xu (2014), and Reyes, Robert and Xu (2017). Harrison et al. (2014) use the World Bank Enterprise Survey data to provide evidence that key factors for improving African firm performance are political competition, infrastructure, and access to finance. Reyes et al. (2017) offered a more comprehensive list of the business environment factors and add measures of BE elements such as taxes, labor regulation, entry and exit, along with the agglomeration environment. They conduct a global analysis while controlling for country fixed effects, offering better control of country heterogeneity. They also explore how the effects of BE variables are different depending on firms’ characteristics (such as age, size) and country level of development. The literature on the effects of pollution in Africa is less developed. Although the topic has gained attention in recent years, the scarcity of pollution data for the region had limited long- 5 There are many studies on the effects of agglomeration economies on long-term growth. See for example Glaeser and Gottlieb (2009); Jones and Romer (2010); Clarke, Li and Xu (2016); Li, Long and Xu (2017)). 3 term analysis (Petkova et al., 2013).6 However, some recent examples that look at air pollution in Africa exist. Roy (2016) focuses on the health effects of air pollution and some economic costs, emphasizing the simultaneous challenges the continent faces: economic development and pollution control. Kirchberger (2016) looks at the spatial links between pollution (and other disamenities) and living standards in Sub-Saharan Africa, while Bishop (2017) develops a broader approach by concentrating not only on pollution but also on the health and economic effects of climate change in general. However, to the best of our knowledge, there is no study that focuses on the links between pollution and productivity in Africa. The contribution of our paper is twofold. First, it extends the literature, particularly the work of Reyes et al. (2017), to focus on Africa, looking at how BE variables affect firm productivity. We follow the same approach by allowing the BE factors to affect firm growth differently for African firms than for those in the rest of the world, and then offer quantitative assessment of key factors for improving African firm growth. We offer diagnoses of hindering business environment (BE) factors for firm performance in Africa with a focus on issues related to agglomeration (or urban factors). Second, it contributes to the pollution literature by focusing on air pollution as another possible determinant of firm productivity and exploring whether its relationship with African firm performance is different, compared to the rest of the world. Although the link between pollution and productivity is well established (see Kahn et al. (2017) for a survey on this topic), this paper focuses on Africa and how air pollution is correlated to worker and firm productivity in that region. Following Reyes et. al (2017), we use the World Bank Enterprise Survey data set. We distinguish three different classifications for business environment (BE): Basic BE, Refined BE, and Agglomeration Environment. Basic BE is likely essential for facilitating firm growth. It consists of Basic Protection (corruption, basic property rights protection, and basic physical safety), infrastructure (power supply and modern telecom), human capital, and access to finance (such as bank finance and trade credit). Refined BE covers other BE elements that have been frequently mentioned in the literature, but perhaps less frequently mentioned than Basic BE. It includes topics such as entry and exit, labor regulations, tax burdens. The Agglomeration Environment is 6 Petkova et al. (2013) provide an overview of available studies. 4 a new category that has not been emphasized in the literature, but recent papers find it important for explaining firm performance across the world (Clarke, Li and Xu 2017). It incorporates whether or not a firm is located in a large city, the extent to which a firm is surrounded by “capable firms”, measured as the share of firms whose number of employees exceeds 50 (capacity agglomeration), and whether the firm competes with firms in the informal sector (informal competition). When looking at air pollution, we focus on measures of particulate matter. Even though many pollutants contribute to contaminate the air, particulate matter is the one that affects human health the most. In particular, small particles, those with a diameter of 2.5 microns or less, are the most dangerous because they can travel in the blood stream and sit in the lungs, producing long lasting health consequences. 7 We use remote sensing data (van Donkelaar, 2016) and assign pollution measures to each particular firm. Finally, in order to measure firm performance, we use the same two variables as Reyes et al. (2017): employment growth and labor productivity growth. Employment growth corresponds to the one-year employment change, and labor productivity is measured as sales over the number of permanent employees. Our findings can be summarized in two main points. Frist, among the business environment variables, capacity agglomeration is the only variable that shows a different effect for African firms. Capacity agglomeration (the share of firms with at least 50 employees) has a stronger positive effect in this region, increasing labor productivity growth. Second, concentration levels of PM 2.5 also have a distinct effect for African firms. The relationship between pollution and firm performance goes through three stages: negative, positive, and negative again (cubic relationship between PM 2.5 and labor productivity growth). Although, as expected, we find a range of pollution levels where the correlation with productivity is positive, it is interesting to see that pollution effects turn negative at considerably low levels of PM 2.5 (approximately 30 /3). 7Deaths attributable to ambient PM2.5 increased from 3.5 million in 1990 to 4.2 million in 2015. Ambient PM2.5 was the fifth-ranking mortality risk factor in 2015 (Cohen, et al. 2017). 5 The rest of the paper is structured as follows. Section 2 describes the data and methodology. Section 3 presents the analysis of how the business environment affects African firms, Section 4 does the same for pollution, and Section 5 briefly focuses on the effects of improving these key variables. Section 6 concludes and hints to some possible course of action for African cities. 2. Data and Methodology Because this paper builds on Reyes, Roberts, and Xu (2017), the main data source we use is also the World Bank Enterprise Survey Dataset. This survey, conducted by the World Bank, takes place in 709 cities, in 128 countries, and collects information on the business climate to understand how this affects firms’ performance. 8 This data set includes information on firm characteristics (such as size, age, number of employees), firm performance (such as employment growth rate, labor productivity, total factor productivity), and business environment variables (such as obstacles to grow, overdraft facility). A list of key variables and sources is presented in Table 1. Summary statistics are shown in Table 2. It is worth noting that WBES data are a representative sample of the non-agricultural economic activity at the country level. Although the data are geographically stratified, this is done on a country-by-country basis, which results in no common stratification for all firms. Moreover, because the population of interest is limited to registered firms with more than 5 employees, most of these firms are clustered around the main urban centers. This could introduce some bias because more productive or fast-growing firms may be attracted to big cities. However, when estimating the model for different subsamples depending on city size, results do not show significant differences. 9 Furthermore, even though firms’ geographic distribution is representative for the country, representativeness at the city level is not automatically assured. Because this paper focuses on the correlation between pollution and firm performance -at the firm level, representativeness at 8 For more details on WBES, see http:\\www.enterprisesurveys.org. 9 Results appear to be slightly stronger for a subsample of firms located in cities with a population between 250,000 and 1 million people. 6 the city level is not a concern. We choose to focus on firm productivity and employment, exploiting the variability of the data at the firm level. Even though we include a variable on capacity agglomeration, we do not intend to draw conclusions that involve the interpretation of firm agglomeration and sorting variables at the city level. The limitations of the data will become more pressing if the analysis wants to be extended at the city level. Table 1. Variable definitions and Sources Variables Definition and source One-year employment growth rate. First calculate 3-year growth rate as the change in permanent employees in 3 years divided by initial number of permanent employees. Then L-grow convert it to one-year growth rate (g1). The L-growth = 2*g1/(2+g1), which is the Haltiwanger measure of labor growth, bound by -2 and 2 to avoid the typical extreme outlier problem associated with the un-transformed g1. The logarithm of labor productivity (LP). LP is measured as sales over the number of permanent lnLP employees. Winsorized at tail 2 percent to avoid the outlier issue. The annualized (Haltiwanger) LP growth rate. The procedure is the same with L-growth except LP-growth that the basic building block is LP instead of L. Total factor productivity, estimated as the residual from industry-specific production function with log value added as the dependent variable and log capital and log labor as the TFP independent variables. K is replacement cost of land and machine. L is the number of permanent employees plus 0.5 times the number of temporary employees. Foreign The share of foreign ownership of the firm. OwnBiggest The ownership share of the largest owner. L0 20-100 The firm’s number of employees three years ago was 20-100 (more than 100). (100+) Age 6-10 (10+) The firm’s age is between 6 and 10 years (or 10 or more years). The firm is located in a city of a million residents or the capital. This variable has 20.5% of missing, and to avoid loss of sample, we impute the missing with World Bank income group BigCity dummies, regional dummies (LAC, MNA, SAR, ECA, EAP, Africa), service industry dummy, size dummies (10-20 employees, 20-60 employees, 60+ employees), age dummies (firm age between 6 and 10 years old; 10 years plus). Small The firm is a small enterprise, that is, employing fewer than 50 employees. Young The firm’s age is younger than 10 years. Rich The firm is located in a country that is below the medium GDP per capita in the sample. Exporter The firm is an exporter. High-k Log of the industry average capital-labor ratio is above the medium of the industries. lnPop1 Log(once-lagged population level) lnGDPPC1 Log(once-lagged GDP per capita in real US dollars). lnPopDen1 Log(once-lagged population density) City-level average of the firm’s answer on whether “XX’ constitutes an obstacle, ranking from XX Obstaclec 0 to 4 (with 4 being more severe obstacles). XX is one of the following: corruption, skilled labor 7 availability, labor regulation, tax rate burdens, land acquisition. The subscript c for these variables indicate that it is based on city-level average rather than a firm’s answer. Overdraft City share of firms with overdraft facility. Based on WBES calculation. Facilityc City average of the proportion of total annual sales of goods and services that are paid for after Trade Creditc delivery. Based on WBES calculation. City share of firms who say that they directly compete with informal firms. It is a measure of Inf the importance of the informal sector and its competition with the formal sector. Based on Competititonc WBES calculation. Outagec City share of firms that experienced a power outage in the survey year. Based on WBES. Webc City share of firms that answer that they use websites to conduct business. Security Costc City average of the share of a firm’s sales paid for security. Based on WBES. Capacity The share of firms whose number of employees exceeding 50, as a proxy of capacity Agglomerationc agglomeration. Computed based on sample firms. Table 2. Summary Statistics N mean sd p10 p25 p50 p75 p90 L-growth 97669 0.041 0.143 -0.084 0.000 0.007 0.096 0.179 L-growth for high-growth firms 5,412 0.370 0.197 0.170 0.230 0.321 0.454 0.627 L-growth for non-high-growth firms 92,257 0.022 0.113 -0.096 0.000 0.000 0.074 0.135 LP-growth 70068 -0.026 0.299 -0.270 -0.126 -0.040 0.051 0.199 LP-growth for high-growth firms 3,582 0.655 0.452 0.190 0.305 0.570 0.815 1.361 LP-growth for non-high-growth firms 66,486 -0.063 0.239 -0.279 -0.133 -0.047 0.032 0.130 TFP 33288 -0.133 1.351 -1.856 -1.032 -0.137 0.709 1.549 lnLP 85803 9.534 1.554 7.431 8.496 9.632 10.635 11.517 Export share 106384 0.100 0.253 0.000 0.000 0.000 0.000 0.400 Foreign share 106284 0.075 0.246 0.000 0.000 0.000 0.000 0.030 OwnLargest 103061 0.795 0.264 0.400 0.510 1.000 1.000 1.000 L0 20-100 97930 0.300 0.458 0.000 0.000 0.000 1.000 1.000 L0 100+ 97930 0.153 0.360 0.000 0.000 0.000 0.000 1.000 Age 6-10 106338 0.221 0.415 0.000 0.000 0.000 0.000 1.000 age 10+ 106338 0.635 0.481 0.000 0.000 1.000 1.000 1.000 BigCity 107777 0.481 0.450 0.000 0.000 0.474 1.000 1.000 Small 107777 0.714 0.452 0.000 0.000 1.000 1.000 1.000 young 106338 0.365 0.481 0.000 0.000 0.000 1.000 1.000 rich 102222 0.488 0.500 0.000 0.000 0.000 1.000 1.000 cityBig 85700 0.476 0.499 0.000 0.000 0.000 1.000 1.000 service 107777 0.446 0.497 0.000 0.000 0.000 1.000 1.000 exporter 106384 0.219 0.413 0.000 0.000 0.000 0.000 1.000 High-k 107,716 0.486 0.500 0.000 0.000 0.000 1.000 1.000 Country level lnPop1 122 15.764 2.001 12.779 14.802 15.999 17.111 18.260 lnGDPPC1 121 7.583 1.264 5.920 6.596 7.666 8.578 9.277 8 lnPopDen1 122 4.084 1.260 2.418 3.446 4.240 4.867 5.660 City level variables Corruption obstacle 638 0.489 0.276 0.092 0.279 0.493 0.697 0.878 Overdraft facility 636 0.377 0.263 0.070 0.146 0.325 0.585 0.771 Trade credit 638 0.427 0.198 0.178 0.272 0.414 0.574 0.706 Inf competition 633 0.499 0.208 0.246 0.338 0.478 0.659 0.783 Skilled labor obstacle 642 0.408 0.223 0.125 0.224 0.388 0.567 0.712 Labor regulation obstacle 638 0.282 0.200 0.058 0.140 0.243 0.402 0.554 Tax rate obstacle 642 0.582 0.224 0.267 0.446 0.594 0.757 0.870 Land access obstacle 638 0.334 0.212 0.100 0.177 0.284 0.469 0.664 Outage 638 0.575 0.269 0.220 0.340 0.578 0.832 0.933 Web Intensity 642 0.440 0.239 0.114 0.241 0.448 0.618 0.766 Security Cost 642 0.018 0.018 0.004 0.008 0.014 0.022 0.037 Capacity agglomeration 648 0.268 0.146 0.077 0.161 0.260 0.354 0.456 In order to include pollution measures in our analysis, we need access to the GPS location of each of the firms of the survey. Because this information is confidential, we only have access to a masked location for each firm, randomly assigned within a radius of 2km of the true firm location. Although not all firms in the WBES data set have GPS locations, comparing the distribution of the key variables between the whole sample and the sample of geo-referenced firms shows no significant differences (see Annex I for details). We have a sample of 56,977 geo- referenced firms. Originally, we have air pollution data for two pollutants, nitrogen dioxide (NO2) from NASA OMI satellite and PM 2.5, from NASA MODIS and MISR Aerosol Optical Depth. Both data sets are in raster format, with a resolution of 0.1x0.1 degrees, approximately 11km by 11km at the equator. Each pixel contains the annual mean of the corresponding pollutant. For nitrogen dioxide, the data span between 1997 and 2011, while for PM 2.5, the period is 1998 to 2015. Because different models may differ in how they estimate the PM 2.5 surface concentration levels, we complement NASA data with PM 2.5 raster data (with the same resolution) estimated by van Donkelaar et al. (2016). Their work uses geographically weighted regression to calibrate ground-based PM 2.5 concentration levels using satellite data and ground level data were available. We overlay WBES masked firm locations with pollution data rasters and extract NO2 and PM 2.5 levels for each location. WBES data include a cross section of firms surveyed between 2006 9 and 2015. Only two countries have three waves of the survey (Bulgaria and the Democratic Republic of Congo), approximately half of the countries have two waves (20% of them are African countries), and the rest has only one wave. Although this is not a panel, firms were asked information on sales and employment for the year of the survey and the three previous years, which allowed the construction of annualized growth rates. To assign pollution data to each firm, we use the year of the survey wave and assign the pollution level that corresponds to that same year. For example, the Republic of Yemen has two waves, 2010 and 2013. Then for firms surveyed during the wave 2010, we assign 2010 pollution mean values, and for firms surveyed in 2013, we assign 2013 pollution mean values. In the end, because firms that have GPS location were mostly surveyed after 2010, our sample is reduced to 80 countries with only one wave of the survey (with the exception of the aforementioned example of Yemen that keeps the two waves). For our estimation, we only use PM 2.5 data. NO2 data are only available until 2011 and geo- reference information is available starting at 2010, which results in most of the geo-referenced firms with no available NO2 data. In the end, the sample for the second half of our analysis consists of 56,841 firms, distributed in 412 cities, in 79 countries. From these firms, 17.4% are in Africa, 6.2% in East Asia Pacific, 12.3% in Europe and Central Asia, 14.3% in Latin America, 12.6% in the Middle East and North Africa, 21% in South Asia, and 16.2% in high-income countries. PM 2.5 concentration levels vary widely across our sample (Figure 1). Only 30% of the firms are located in cities where PM 2.5 levels are considered safe by the WHO, i.e., PM 2.5 annual mean below 10 /3 . Half of the firms are still exposed to low levels, around 13 /3 , and approximately 10% of the firms are in cities with PM 2.5 concentration levels above 50 /3. Moreover, the distribution is not homogenous across regions (Table 3). High-income countries show low levels of PM 2.5, while regions like South Asia or East Asia are among the most polluted ones, with an average around 45 and maximum levels of 111 and 94 respectively. 10 Table 3. PM 2.5 Summary Statistics by Region Obs Mean Std. Dev. Min Max Region Africa 9,889 13.68 8.06 1 35 East Asia Pacific 3,509 45.70 22.51 2 94 East Europe and Central Asia 7,004 14.64 4.97 5 30 Latin America 8,148 6.13 3.25 1 23 Middle East and North Africa 7,157 10.08 2.99 4 21 South Asia 11,968 45.89 18.38 6 111 High income OECD 3,517 12.58 4.80 0 23 High-income non-OECD 5,649 12.50 4.60 2 24 For our analysis, we follow the same approach as in Reyes et al. (2017). Focusing on dynamic firm performance as captured by firm employment growth rate and firm labor productivity growth rate, the authors link firm growth with basic firm controls (i.e., firm size, age, foreign ownership), industry, year dummies, and of course, the BE variables. A key advantage of their specification is that they control for country fixed effects, which essentially hold constant all country-specific characteristics such as political institutions, culture, geography, size. Their identification of the impacts of the various dimensions of the BE thus comes from comparing firm performance across city-industry cells within countries. Moreover, since they capture the BE 11 using city-industry average of firm-level response on aspects of the business environment (such as access to bank loans, trade credit, spending on security as a share of sales, specific BE aspect considered an obstacle), they have thousands of unique values of the BE variables, and thus effectively circumvent the serious multicollinearity issue embedded in cross-country studies of the BE effects. To avoid outlier issues when using growth rates, we use the same approach as in Reyes et al. (2017). 10 We also cluster the standard errors at the city level. We focus on two main concerns: how does the business environment affect African firms, and how air pollution is correlated with firm productivity. To answer the first question, we allow the BE effects to differ for African firms and firms in the rest of the world, and see what implications emerged from the investigation. To aid interpretation, we show that there is no strong evidence of sorting along city size in our sample. For the second question, we introduce PM 2.5 measures of air pollution as another determinant of firm productivity and look at the effects on employment and labor productivity. When following this strategy for the whole world, we expect to find lower employment growth and labor productivity growth rates for firms with higher PM 2.5 concentration levels. We then focus only on a subsample of African firms. Because there is not enough variability of pollution data at the country level, we do not control for country fixed effects in this last specification. We estimate an equation of the form: = + + + + + where , , , , index firms, countries, cities, industries, and year, respectively. is a vector of firm level controls, such as, age, size, type of ownership, is a vector of underlying 10 From Reyes et a. (2017): “… we follow Davis and Haltiwanger (1999) by calculating mid-point growth rates by dividing the change in employment (or labor productivity) between the survey year and three years earlier by the simple average of employment (or labor productivity) in the beginning and ending years. This bounds the resulting growth rate between – 2 and +2, thereby significantly reducing the influence of outliers”. 12 BE effects at the city-level, and is a vector with the corresponding pollution measure. We also control by country and year dummies. The problem when trying to correctly identify the effects of pollution on labor productivity and employment, is that pollution is usually accompanied by other variables that can bias the estimates. For example, pollution may be accompanied by economic growth that can also impact firm performance. Another possible bias may come from the fact that workers and firms may respond to pollution by adjusting their behavior, and through that mitigate the effect that pollution would otherwise have on their productivity levels. They may adopt some measures that could include the use of individual masks, or using air filters, or even decisions to relocate. The data at hand limit our ability to control for all these variables and restrict the conclusions we can draw from this analysis. Hence, we do not claim that the effects of pollution on labor productivity and employment are well identified, but instead, this work makes an effort to effectively track the evolution of these two connected variables and see how their correlation differs for Africa with respect to other regions. Moreover, as mentioned by Graff-Zivin and Nediell (2012), estimating the effect of pollution on worker productivity could be challenging due to two main reasons. First, measures of output per worker usually do not isolate worker productivity from other inputs. Second, exposure to pollution is often endogenous. For example, there maybe be sorting of high productivity-high income workers in less polluted cities (Banzhaf and Walsh, 2008). Moreover, even if exposure to pollution is exogenous, some workers may decide on how much exposure they get (by changing shifts, for example) making exposure still endogenous (Neidell, 2009). With these restrictions in mind, we use labor productivity as a measure of firm performance and look at the correlation between pollution and labor productivity. Measuring individual worker productivity and estimating the marginal effects of pollution exceeds the scope of this paper. 3. Allowing Africa-Specific Business Environment Effects We look at how BE effects are different for African cities. We report the specification in which the Africa dummy is interacted with our key business environment (BE) variables. To report a parsimonious specification and to avoid multicollinearity, we first interact the Africa dummy 13 with all BE variables that are statistically significant in at least one of the two outcome regressions. We then drop those African-interaction terms, and re-run. We continue this process so that we keep only the African interaction terms that are significant in at least one of the two outcome regressions. In the end, only Capacity Agglomeration has a significant interaction term with the Africa dummy. Table 4 only presents the results of this iteration. Table 4. How African Countries Differ in Effects of the Business Environment L-growth LP-growth coef se coef se Security Cost -0.076 0.049 -0.112 0.108 Corruption obstacle -0.015* 0.006 0.019 0.013 Overdraft facility 0.014* 0.006 0.034* 0.014 Trade credit -0.008 0.008 0.036* 0.016 Skilled labor obstacle 0.004 0.006 -0.011 0.013 Labor regulation obstacle 0.002 0.007 0.003 0.015 Tax rate obstacle -0.007 0.007 0.008 0.015 Land access obstacle -0.006 0.007 -0.002 0.014 Outage -0.001 0.007 -0.003 0.013 Web Intensity 0.018** 0.006 0.036** 0.013 In Big City 0.006* 0.003 -0.001 0.005 Capacity agglomeration 0.042* 0.016 -0.001 0.027 Capacity agglomeration * Africa 0.032 0.035 0.171* 0.076 Inf competition -0.012* 0.005 0.006 0.011 Other controls Yes Yes Industry dummies Yes Yes Observations 57,455 44,279 Adj. R squared 0.131 0.286 *, ** represent statistical significance at the 5 and 1 percent levels. Heteroskedasticity-corrected standard errors clustered at the city level. Other controls include year dummies, missing dummy for BE variables, and base controls Capacity agglomeration had much more positive association with labor productivity growth only in African cities. One standard deviation increase (0.12) is associated with a labor productivity growth increase of almost 2 percentage points, a considerably large effect. In other regions, the coefficient is not statistically significant, showing there is no association between capacity agglomeration and labor productivity growth. This is suggestive evidence that capacity agglomeration is a key issue for Africa. Moreover, compared to other regions, Africa shows the lowest level of capacity agglomeration (Table 5). 14 Table 5. Regional summary statistics AFR EAP ECA LAC MNA SAR High Income Mean Sd Mean Sd Mean Sd Mean Sd Mean Sd Mean Sd Mean Sd L-grow 0.041 0.172 0.054 0.093 0.022 0.124 0.027 0.140 0.016 0.148 0.029 0.083 0.027 0.140 LP-grow -0.059 0.596 -0.042 0.148 -0.014 0.269 - 0.234 -0.126 0.248 -0.026 0.149 -0.032 0.249 0.024 Foreign Share 0.086 0.253 0.036 0.169 0.052 0.197 0.100 0.283 0.054 0.207 0.008 0.077 0.066 0.236 Own Largest 0.873 0.226 0.849 0.223 0.812 0.252 0.718 0.274 0.752 0.285 0.781 0.266 0.794 0.265 Exporter 0.190 0.392 0.197 0.398 0.235 0.424 0.297 0.457 0.234 0.424 0.164 0.370 0.246 0.431 Once-lagged GDP (log) 6.501 0.710 7.975 0.217 8.050 0.900 8.251 0.593 7.550 0.585 6.921 0.326 9.170 0.567 In Big City 0.584 0.493 0.856 0.334 0.189 0.303 0.705 0.456 0.454 0.498 0.436 0.496 0.297 0.444 Corruption Obstacle 0.446 0.285 0.140 0.199 0.441 0.247 0.644 0.190 0.800 0.236 0.444 0.310 0.395 0.174 Overdraft Facility 0.271 0.212 0.282 0.286 0.369 0.266 0.579 0.233 0.323 0.256 0.522 0.262 0.416 0.312 Trade Credit 0.244 0.149 0.521 0.244 0.364 0.189 0.562 0.186 0.384 0.191 0.501 0.214 0.538 0.174 Inf. Competition 0.653 0.212 0.499 0.269 0.416 0.198 0.690 0.156 0.440 0.195 0.408 0.223 0.367 0.216 Skilled Labor Obstacle 0.311 0.193 0.233 0.218 0.322 0.193 0.644 0.144 0.320 0.203 0.316 0.223 0.457 0.209 Labor Regulation Obstacle 0.259 0.179 0.112 0.161 0.161 0.134 0.532 0.186 0.287 0.161 0.293 0.202 0.299 0.215 Tax Rate Obstacle 0.599 0.230 0.295 0.280 0.525 0.211 0.641 0.199 0.548 0.178 0.540 0.240 0.682 0.217 Land Access Obstacle 0.460 0.209 0.188 0.204 0.209 0.158 0.297 0.119 0.323 0.217 0.340 0.255 0.224 0.157 Outage 0.824 0.183 0.506 0.315 0.442 0.214 0.543 0.196 0.669 0.320 0.674 0.282 0.317 0.177 Web Intensity 0.314 0.213 0.589 0.320 0.553 0.227 0.607 0.186 0.486 0.239 0.475 0.255 0.679 0.197 Security Cost 0.032 0.027 0.009 0.011 0.018 0.019 0.018 0.014 0.012 0.015 0.012 0.010 0.012 0.012 Capacity Agglomeration 0.183 0.109 0.489 0.203 0.263 0.099 0.369 0.104 0.261 0.132 0.367 0.128 0.266 0.091 PM 2.5 13.682 8.064 45.697 22.510 14.636 4.974 6.128 3.249 10.085 2.986 45.894 18.379 12.528 4.675 As an additional exercise, we look at how BE effects differ by city size of a firm’s location and explore whether there are sorting effects of firms across cities. Sorting effects can be present for different reasons as specific characteristics of cities may attract certain types of firms. For example, big cities could provide better infrastructure and access to finance. They also usually provide better amenities that can attract more technologically-advantaged firms and access to better education that could impact workers’ qualifications. To investigate how BE effects differ by city size of a firm’s location, we interact the dummy variable of whether the firm is located in a city of more than a million residents with the BE variables. 11 To reduce multicollinearity, we drop those interaction terms that are statistically 11 The data allow us to potentially delve into richer ways of how city size affects firm growth. In particular, we know whether the firm is located in a small city (i.e., less than 50,000, between 50,000-250,000, 250,000 to one million, capital city). In empirical explorations, we found that being affiliated with medium cities (i.e., 250,000 to a million) or small cities (i.e., 50,000 to 250,000) did not have statistically significantly different effects from being affiliated with little cities (i.e., fewer than 50,000 people). We thus only keep the dummy of being affiliated with big cities. 15 insignificant in both employment growth (L-growth) and labor productivity growth (LP-growth) regressions (Table 6). Table 6. How the BE Effects Differ by City Size Lgrow LPgrow coef se coef se In Big City 0.017* 0.007 0.019 0.015 Corruption Obstacles -0.015* 0.006 0.023 0.014 Overdraft Facility 0.015* 0.006 0.031* 0.014 Trade Credit -0.008 0.009 0.045** 0.016 Inf Competition -0.012* 0.006 0.002 0.012 Skilled Labor Obstacle 0.000 0.006 -0.001 0.013 Labor Regulation Obstacle 0.002 0.009 0.021 0.017 Labor Regulation Obstacles * In Big City -0.001 0.009 -0.047** 0.017 Tax Rate Obstacle -0.005 0.007 0.004 0.016 Land Access Obstacles -0.005 0.007 -0.006 0.015 Outage 0.003 0.007 -0.009 0.014 Web Intensity 0.019** 0.006 0.029* 0.013 Security Costs -0.105* 0.053 -0.049 0.115 Capacity Agglomeration 0.047* 0.019 0.032 0.028 Capacity Agglomeration * In Big City -0.034 0.019 -0.023 0.035 Other Controls as in Reyes et al. (2017) Yes Yes Observations 52,614 40,507 Adjusted R squared 0.123 0.287 Interestingly, most BE variables do not have different effects in big versus non-big cities, which may point at the fact that many business environment constraints are still due to national level restrictions, such as tax or trade regulations. However, labor inflexibility seems to have more harmful association with labor productivity growth in big cities. Only in big cities, increasing Labor Regulation Obstacles by one standard deviation (0.2) is associated with a drop in labor productivity growth by 0.5 percentage points. This makes sense since in large cities the benefits of a flexible labor market should be larger due to better opportunities for labor-employer matching, and labor inflexibility is likely to cause more harm in terms of productivity growth. In contrast, the benefits of capacity agglomeration, while present in both big and non-big cities, are more pronounced in non-big firms. Increasing capacity agglomeration by one standard deviation 16 (0.15) is associated with higher employment growth by 0.7 percentage points in non-big cities, and by 0.2 percentage points in big cities. The stand-alone Big City effect is also pronounced: its premium for employment growth is 1.7 percentage points. To see if there is sorting of firms in big cities, we use the affiliation of high-tech industries by firms as a proxy of firm quality or technology. We want to see if firms located in big cities are more likely to be in high-tech industries. High-tech industries are those in metals and machinery, electronics, chemicals and pharmaceuticals, non-metallic and plastic materials, auto and auto components. The rest of the manufacturing firms are then classified as low-tech manufacturing, such as those in textiles, leather, garments, food, wood and furniture, and other manufacturing.12 After controlling for country dummies, the coefficient that relates the high-tech dummy with the In Big City dummy clearly shows that there is no statistically significant correlation between the high-tech status and In Big City (Table 7). Table 7. A Check on Sorting of Firms and Cities High-Tech dummy High-Tech dummy coef se coef se In Big City 0.040 0.033 0.004 0.011 Country Dummies no no yes yes N 45,839 45,839 r2_a 0.002 0.155 Standard errors are clustered at the city level. 4. Pollution Effects on Firm Productivity We now move our focus to the second question: is firm performance affected by air pollution? Following a similar strategy, we estimate the effect of PM 2.5 on employment growth and labor productivity growth. As mentioned earlier, because location data are missing for some of the firms, we restrict our sample to only those firms for which we have location and PM 2.5 concentration values. Additionally, data on PM 2.5 from van Donkelaar have better coverage than PM 2.5 obtained from NASA directly, so we opt to use the first one. We explore different model specifications and present the results of the model that better fits the data. Globally, pollution seems to be associated with employment growth positively and 12 This exercise is only do-able for manufacturing firms, and we thus only use the manufacturing firms. 17 linearly, while for labor productivity growth a quadratic specification fits the data better. Intuitively, on the one hand, we expect labor productivity to grow at the beginning, until a certain point, beyond which higher levels of pollution will start leading to declining values of labor productivity. On the other hand, the positive linear relationship between pollution and employment growth may indicate that firms will always try to hire more employees, even while pollution levels are rising. Since we find that productivity decreases while pollution increases (after a certain point) this positive relationship between pollution and employment growth is not surprising. Firms need to hire more employees to achieve similar production levels. When focusing on African firms, however, we find that a cubic specification of pollution fits better, for both dependent variables. When looking at the effect of pollution on firm performance around the world, we find that results differ for employment growth versus labor productivity growth (Table 8). Because PM 2.5 and employment growth are linearly correlated, we only see constant effects. As expected, the effect is negative (-0.0002). An increase of PM 2.5 in 10 units, is associated with a reduction in employment growth of 0.2%. This effect appears to be constant over time, which implies considerably big effects for firms in cities experiencing high levels of pollution. For example, in China and India, firms in cities like Chengdu or Bihar are exposed to pollution levels above 90 /3. For these firms, employment growth could be approximately 1.4 percentage points compared to what it could be if pollution levels were below the WHO guideline (10 /3). Table 8. Pollution Effect on Firm Performance (Global) L-growth LP-growth coef se coef se Security Cost -0.165* 0.073 -0.027 0.141 Corruption obstacle -0.028** 0.008 0.019 0.017 Overdraft facility 0.013* 0.007 0.033× 0.017 Trade credit -0.010 0.009 0.049** 0.018 Skilled labor obstacle 0.001 0.007 -0.005 0.016 Labor regulation obstacle 0.013 0.008 -0.003 0.018 Tax rate obstacle -0.004 0.008 0.008 0.018 Land access obstacle -0.010 0.008 0.005 0.021 Outage 0.007 0.007 -0.019 0.015 Web Intensity 0.019** 0.007 0.034* 0.014 In Big City 0.004 0.003 -0.003 0.006 18 Capacity agglomeration 0.023 0.018 -0.004 0.030 Informal competition -0.013* 0.007 0.001 0.013 2.5 -0.0002 × 0.0001 0.003** 0.001 2.52 -0.00002** 0.00001 Other controls Yes Yes Industry dummies Yes Yes Observations 35,736 26,910 Adj. R squared 0.111 0.289 × , *, ** represent statistical significance at the 10, 5, and 1 percent levels. Heteroskedasticity-corrected standard errors clustered at the city level. Other controls include year dummies, missing dummy for BE variables, and base controls. The relationship between PM 2.5 and labor productivity, on the contrary, appears to be non- linear. The effect is positive (0.003) at low levels of pollution, although decreasing as pollution levels rise (-0.00002). The intuition behind this result is that lower levels of this pollutant may identify an area in the earlier stages of development. There, the negative effects of pollution on productivity are offset by the positive effects of growth. As the level of pollution rises, the negative effects are more notable and start to show a negative correlation with labor productivity.13 The next step is to analyze whether this correlation between air pollution and firm performance is different (or stronger) for African firms. To do this, we limit our sample to only African firms, and run our analysis. A cubic specification helps explain the relationship between pollution and both dependent variables better. We show these results in Table 9. Table 9. Pollution Effects on African Firms L-growth LP-growth coef se coef se Security Cost -0.163 0.130 -0.166 0.313 Corruption obstacle -0.043* 0.018 -0.178* 0.068 Overdraft facility -0.015 0.018 0.104× 0.053 Trade credit -0.026 0.028 0.160× 0.092 Skilled labor obstacle 0.042 × 0.022 0.038 0.083 Labor regulation obstacle 0.036 0.024 -0.176** 0.061 Tax rate obstacle 0.001 0.018 0.112 0.078 13 The inverted U-shape relationship between pollution and growth is widely known as the Environmental Kuznets Curve (Grossman and Krugman, 1991). This states that countries in their earlier face of development will increase pollution levels until they reach a maximum where they learn to manage pollution and still be able to keep developing. Developing nations are then on the positive segment of the Environmental Kuznets Curve, while more developed nations are placed at the end, on the negative segment. 19 Land access obstacle -0.028 0.020 0.069 0.067 Outage 0.082** 0.019 -0.117 0.083 Web Intensity 0.025 0.019 0.070 0.069 In Big City 0.010 0.008 -0.023 0.028 Capacity agglomeration 0.052 0.052 0.398* 0.155 Inf competition -0.075** 0.017 0.036 0.063 2.5 -0.009× 0.005 -0.083** 0.015 2.52 0.001* 0.0003 0.005** 0.001 2.5 3 -0.00001* 0.00001 -0.00007** 0.00002 Other controls Yes Yes Industry dummies Yes Yes Observations 4,563 2,742 Adj. R squared 0.167 0.484 × , *, ** represent statistical significance at the 10, 5, and 1 percent levels. Heteroskedasticity-corrected standard errors clustered at the city level. Other controls include year dummies, missing dummy for BE variables, and base controls. These results shed some light on how pollution correlates with productivity for African firms differently. Since the relationship appears to be cubic, the effect of pollution on employment growth and labor productivity growth differ depending on the level of pollution. At low levels of pollution, the effect on both dependent variables is negative (-0.009 and -0.083, respectively). Because the quadratic term is positive, these negative effects will have less impact as pollution levels rise, turning into positive effects. The negative cubic coefficients show that after a certain point, pollution will again have a negative effect on employment and labor productivity growth. The effects on both dependent variables are similar. In terms of employment growth, African firms experience negative effects of pollution at extremely low levels of PM 2.5, approximately until 8 /3. At these levels, African firms may not be experiencing enough of the positive benefits from growth needed to offset the negative effects of pollution. From there, the effects of pollution increase for higher levels of PM 2.5, until approximately 26 /3 , where the effects turn negative again. For labor productivity growth, the effects are negative for PM 2.5 levels below 13 /3. From that level until approximately 32 /3, the effect of pollution is positive, again, as a consequence of the positive effects of growth offsetting the negative pollution effects. After that, the effect of pollution on labor productivity growth is negative again. Pollution then has negative effects on employment growth sooner than on labor productivity growth. 20 As a robustness check, we run the same regression using a spline specification for the pollution variable. Splines allow for more flexibility in the relationship between the variables being considered -in this case, pollution and firm performance, and reduce model misspecification. We follow Harrel (2010) and choose four knots at specific quantiles of the independent variable. 14 While overall results are consistent with our previous findings and our final conclusions still hold, the significance of the coefficient of the first spline is slightly reduced. For a full characterization of these robustness checks see Annex II. 5. Gains from Improving the Business Environment and Pollution To shed light on what African countries can do to facilitate firm growth, we conduct some back of the envelope calculation of potential gains from improving key BE variables relevant to Africa and managing PM 2.5 levels. For the BE variables, we only report those “policy instruments” that proved to be statistically significant for either employment growth or labor productivity growth for the African countries. We conduct the following thought experiment. Using the Africa-specific coefficients of BE variables, if a key BE element improves by one standard deviation (i.e., increase by one standard deviation for “good” BE elements and reduce by one standard deviation for “bad” BE elements), what percentage of the standard deviation of the outcome in Africa can be expected to change (assuming causality). The results are in Table 10. Table 10. Potential Growth Effects of Improving Key Policy Instruments for African Firm Here, we only report those “policy instruments” with the following characteristics. First, the instrument must be statistically significant for the specific growth result for the African countries. Second, the instrument must be something that the government might be able to do something. % African SD in the outcome explained by one Africa-SD improvement in the instrument Improve by one Africa- ∗ , African coeff. SD , For L-Growth: , = . 14 Harrell (2010) argues that for most datasets choosing 4 knots is “an adequate fit of the model and is a good compromise between flexibility and loss of precision caused by overfitting a small sample”. These four knots correspond to 5%, 35%, 65%, and 95% quantiles, and take the following PM 2.5 values: 4, 10, 17, 31. 21 Corruption Obstacles -0.015 -0.3 2.9% Overdraft Facility 0.014 0.23 2.7% Informal Competition -0.012 -0.21 1.6% Web Intensity 0.018 0.17 1.9% Capacity agglomeration 0.074 0.12 5.6% For LP-Growth , = . Overdraft Facility 0.034 0.23 1.9% Trade Credit 0.036 0.14 1.2% Web Intensity 0.036 0.17 1.5% Capacity agglomeration 0.170 0.12 4.9% Note. The coefficients are for the African regions. That is, they are based on Table 4. African Standard Deviations correspond to the whole sample. This table reports the associated change (in percent of one standard deviation of the outcome in Africa) when the key variable changes by one standard deviation (in Africa). The most important BE element is clearly capacity agglomeration. Increasing it by one African standard deviation is associated with an increase in African employment growth and labor productivity growth standard deviations by 5.6 and 4.9 percent, respectively. This is the largest positive association we observe in all BE variables. Relatedly, improving Informal Competition by one standard deviation is associated with increasing 1.6 percent of labor growth standard deviations. As a group, improving Basic BE is clearly important. First, improving corruption obstacles by one African standard deviation is associated with higher employment growth by slightly less than 3 percent of one employment growth standard deviation. Second, improving modern infrastructure (i.e., Web Intensity) by one African standard deviation is associated with increasing employment growth and labor productivity growth standard deviations by 1.9% and 1.5%, respectively. The effects are thus also potentially large. Finally, access to finance appears to be an important area to work on. Improving access to bank finance by one African standard deviation is associated with increasing 2.7% of one employment growth standard deviation, and 1.9% of one labor productivity growth standard deviation. Similarly, improving trade credit by one African standard deviation is associated with increasing 1.2% of one labor productivity growth standard deviation. 22 Table 11. Potential Growth Effects of Improving PM 2.5 Levels for African Firm Here, we only report those “policy instruments” with the following characteristics. First, the instrument must be statistically significant for the specific growth result for the African countries. Second, the instrument must be something that the government might be able to do something. % African SD in the outcome explained by one Africa-SD improvement in the instrument Improve by one ∗ , African coeff. Africa-SD , For L-Growth: , = . 2.5 (at x=mean) 0.0032 8.064 14.89% 2.5 (at x=40) -0.0199 -8.064 92.99% For LP-Growth: , = . 2.5 (at x=mean) 0.0041 8.064 5.48% 2.5 (at x=40) -0.0479 -8.064 64.78% Note. The coefficients are for the African regions. That is, they are based on Table 9. They have been calculated at two different values for x, with the following equation: 1 + 22 + 33 2 , where 1 is the coefficient for 2.5, 2 the coefficient for 2.52 , and 3 the coefficient for 2.53 . This table reports the associated change (in percent of one standard deviation of the outcome in Africa) when pollution changes by one standard deviation (in Africa). The impact is even more shocking when looking at pollution levels (Table 11). After relatively low PM 2.5 concentration levels, the effects turn negative, with a huge impact on both, employment growth and labor productivity growth. For example, at 40 /3 , a reduction in one standard deviation of PM 2.5, approximately 8 units, is associated with an increase in employment growth in 93% of its standard deviation (0.172), for African firms. This huge effect translates in an increase of 16.03% in employment growth rate. At the mean level, the annual employment growth rate in Africa will go from 4.1% to 20.1%. For labor productivity, the same exercise yields an increase in labor productivity growth of 38.63 percentage points, which at the mean translates in a positive growth rate of 32.8%. 6. Conclusions In this paper, we shed some light on how air pollution and business environment affect city competitiveness through firm performance, particularly for African firms. We measure firm performance with two variables, employment growth and labor productivity growth. We use 23 firm-level data from the World Bank Enterprise Survey and combine it with PM 2.5 data estimated by van Donkelaar et al. (2016) (based on NASA MODIS and MIRS satellite data). We perform the analysis at the global level and then focus on how these effects are different for African firms. When compared to the rest of the world, African firms experience different effects in terms of pollution. While a global analysis shows that the association between pollution and labor productivity is positive in the beginning and negative as pollution rises (quadratic), African firms experience a different story. PM 2.5 levels are negatively associated with labor productivity for these firms at low levels of pollution. Between approximately 8 /3 and 25 /3, the effects turn positive, showing that as pollution and growth rise, the positive effect from growth offsets the negative effects of pollution. After a not so high level of pollution (an annual average equal to 25 /3) its effects on employment and labor productivity growth are negative again. The fact that African cities are growing fast but PM 2.5 levels have not yet reached the high levels experienced in places like China and India, poses an opportunity to plan ahead and avoid the mistake of not accounting for the negative effects associated with fast and unplanned growth. PM 2.5 has a negative effect on firm performance at levels that are not as high as the ones that these countries are currently experiencing. Pollution management is crucial to tackle the problem before it gets more serious. African cities could benefit from the examples we see now in cities that have experienced fast growth. Looking at business environment and the effects on firm performance, we find that the only variable that affects African firms differently is capacity agglomeration. The effect of capacity agglomeration on labor productivity growth is stronger for firms in this region. An increase of 0.12 percentage points in capacity agglomeration results in an increase of almost 2 percentage points in labor productivity growth. There is no evidence of sorting of firms in big cities and most BE environment variables appear to have similar effects, regardless of the city size. This suggests that the stronger effect of capacity agglomeration for African firms is not clouded by sorting of firms or city size. These results suggest that African firms face a few different challenges compared to firms in other regions. Moreover, addressing these problems could be key in pushing economic growth 24 for the region, as many African countries are in early stages of development. Cities in Africa are urbanizing fast, trying to lift the population out of poverty. Managing pollution and improving the business environment could help their firms improve their performance and will certainly help African cities become more competitive. References Adhvaryu, Achyuta, Namrata Kala, and Anant Nyshadham. 2014. "Management and shocks to worker productivity: evidence from air pollution exposure in an Indian garment factory." Unpublished Working Paper, University of Michigan. Begg, Iain. 1999. "Cities and Competitiveness." Urban Studies 36 (5-6): 795-809. Birdsall, Nancy, and David Wheeler. 1993. "Trade Policy and INdustrial Pollution in Latin America: where are the pollution havens?" Journal of Environament & Development 2 (1): 137-149. Bishop, Russell. 2017. "Confronting Climate Change: Africa’s leadership on an increasingly urgent issue." In Foresight Africa: Top priorities for the continent in 2017. Brookings Institution. Boltho, Andrea. 1996. "The Assessment: International Competitiveness." Oxford Review of Economic Policy 12 (3): 1-16. Chang, Tom, Joshua Graff-Zivin, Tal Gross, and Matthew Neidell. 2016. "The effect of pollution on worker productivity: evidence from call-center workers in China 2016." National Bureau of Economic Research (No. w22328). Chang, Tom, Joshua S. Graff-Zivin, Tal Gross, and Matthew J. Neidell. 2014. "Particulate pollution and the productivity of pear packers." National Bureau of Economic Research (No. w19944. ). Clarke, George, Yue Li, and Lixin Colin Xu. 2016. "Business Environment, Economic Agglomeration, and Job Creation around the World." Applied Economics 48 (33): 3088- 3103. Clarke, George, Yue Li, and Lixin Colin Xu. 2017. "Business Environment, Economic Agglomeration, and Job Creation around the World." Applied Economics 48 (33): 3088- 3103. 25 Cohen, Aaron J., Michael Brauer, Richard Burnett, and et. al. 2017. "Estimates and 25-year trends of the global burden of disease attributable to ambient air pollution: an analysis of data from the Global Burden of Diseases Study 2015." The Lancet, Volume 389, Issue 10082 1907-1918. Cole, Matthew A. 2004. "Trade, the Pollution Haven Hypothesis and the Environmental Kuznets Curve." Ecologial Economics 48 (1): 71-81. Davis, S. J., and J. Haltiwanger. 1999. "On the driving force between cyclical movements in employment and job reallocation." American Economic Review 89: 1234-1258. Eskeland, Gunnar S., and Ann E. Harrison. 2003. "Moving to Greener Pastures? Multinationals and the Pollution Haven Hypothesis." Journal of Development Economics 70 (1): 1-23. Fagerberg, Jan. 1996. "Technology and Competitiveness." Oxford Review of Economic Policy 12 (3): 39-51. Glaeser, Edward L., and J.D Gottlieb. 2009. "The Wealth of Cities: Agglomeration Economies and Spatial Equilibrium in the United States." Journal of Economic Literature 47 (4): 983-1028. Glaeser, Edward L. 1998. "Are Cities Dying?" Journal of Economic Perspectives 12 (2): 139-160. Graff-Zivin, Joshua, and Matthew Neidell. 2012. "The impact of pollution on worker productivity." The American economic review 102 (7): 3652-3673. Harrell, Frank E. (2010) "Regression Modeling Strategies: with applications to linear models, logistic regression, and survival analysis". Springer-Verlag New York, Inc., New York. Harrison, Ann, Justin Y. Lin, and Lixin Colin Xu. 2014. "Explaining Africa's (Dis)advantage." World Development 63: 59-77. Henderson, J. Vernon. 1974. "The sizes and types of cities." The American Economic Review 640- 656. Jones, Charles, and Paul M. Romer. 2010. "The New Kaldor Facts: Ideas, Institutions, Population, and Human Capital." American Economic Journal: Macroeconomics 2 (1): 224-245. Kahn, Matthew E, Nancy Lozano-Gracia, and Maria E Soppelsa. 2017. "Does Pollution Hinder Developing Countries Urban Competitiveness?" Working Paper. Kahn, Matthew E., and Erin T. Mansur. 2013. "Do local energy prices and regulation affect the geographic concentration of employment?" Journal of Public Economics 101: 105-114. 26 Kirchberger, Martina. 2016. "Pollution, Crime, and Mistrust across Space: Evidence from Sub- Saharan Africa." Working Paper. Li, Wei, Xiaoning C. Long, and Lixin Colin Xu. 2017. "Regulation, Agglomeration, and the Reversal of Fortune between China and India." mimeo, World Bank. Petkova, Elizaveta P., Jack W Darby, Nicole H Volavka-Close, and Patrick L Kinney. 2013. "Particulate matter pollution in African cities." Air Quality, Atmosphere & Health 6 (3): 603-614. Porter, Michael E. 2000. "Location, competition, and economic development: Local clusters in a global economy." Economic development quarterly 14 (1): 15-34. Porter, Michael. 1990. "The Competitive Advantage of Nations." Harvard Business Review. Reyes, Jose Daniel, Mark Roberts, and L. Colin Xu. 2017. "The Heterogeneous Growth Effects of the Business Environment. Firm-Level Evidence for a Global Sample of Cities." Policy Research Working Paper, 8114. Rodrik, Dani. 2016. "An African Growth Miracle?" Journal of African Economies 1-18. Roy, Rana. 2016. "The cost of air pollution in Africa." OECD Development Centre Working Ppaers (OECD Publishing) (333). Stern, Nicholas. 2002. A Strategy for Development. Washington, DC: The World Bank. Tolley, George S. 1974. "The Welfare Economics of City Bigness." Journal of Urban Economics 1 (3): 325-345. van Donkelaar, A., R.V Martin, M. Brauer, and et al. 2016. "Global Estimates of Fine Particulate Matter using a Combined Geophysical-Statistical Method with Information from Satellites, Models, and Monitors." Environ. Sci. Technol. World Bank. 2005. "A Better Investment Climate for Everyone." World Development Report, World Bank. World Bank. 2016. "Competitive Cities for Jobs and Growth. What, Who, and How." World Bank. 2010. "Doing Business." World Economic Forum. n.d. Accessed 2018. https://www.weforum.org/agenda/2016/10/air- pollution-the-true-cost-in-numbers/. 27 World Economic Forum. 2009. "The Africa Competitiveness Report 2009." World Bank, Washington DC. 28 Annex I We compare the distribution for the dependent variables and the relevant independent variables for the two samples: complete sample and geo-referenced firms. Dependent Variable: Independent Variables: 29 30 31 32 Annex II As a robustness check, we re-run the regression for African firms using splines, finding the following results. Table A1. Pollution Effects on African Firms – Model Specification using splines L-growth LP-growth coef se coef se Security Cost -0.157 (0.134) -0.167 (0.331) Corruption obstacle -0.0402× (0.0181) -0.173× (0.0677) Overdraft facility -0.0131 (0.0177) 0.113× (0.0533) Trade credit -0.0254 (0.0280) 0.161 (0.0939) Skilled labor obstacle 0.0436 (0.0222) 0.0444 (0.0853) Labor regulation obstacle 0.0336 (0.0237) -0.179* (0.0625) Tax rate obstacle 0.00170 (0.0171) 0.113 (0.0775) Land access obstacle -0.0292 (0.0198) 0.0700 (0.0663) Outage 0.0807** (0.0193) -0.111 (0.0822) Web Intensity 0.0241 (0.0186) 0.0706 (0.0695) In Big City 0.00988 (0.00764) -0.0251 (0.0278) Capacity agglomeration 0.0611 (0.0508) 0.416× (0.157) Inf competition -0.0766** (0.0170) 0.0301 (0.0643) 2.5 ( 4 10) -0.00286 (0.00217) -0.0384** (0.00787) 2.5 ( 10 17) 0.0295× (0.0122) 0.166** (0.0434) 2.5 ( 17 31) -0.0686 × (0.0274) -0.312* (0.101) Other controls Yes Yes Industry dummies Yes Yes Observations 4,563 2,742 Adj. R squared 0.168 0.483 × , *, ** represent statistical significance at the 10, 5, and 1 percent levels. Heteroskedasticity-corrected standard errors clustered at the city level. Other controls include year dummies, missing dummy for BE variables, and base controls. 33