Market Concentration, Trade Exposure, and Productivity in Developing Countries: Evidence from Mexico* Carlos Rodríguez-Castelán† Luis-Felipe López-Calva‡ Oscar Barriga-Cabanillas§ April 23, 2019 Abstract Middle income countries face an elusive quest for productivity growth as they struggle to catch-up with the income level of developed economies. Although the economic literature has identified product market competition as one of the main determinants of productivity growth, most studies have focused on cross- country analyses and have defined market concentration at the national and industry level. This paper takes advantage of an uncommon database –for developing country standards—containing 20-years of establishment census data in Mexico to estimate the effects of local industry concentration in the manufacturing sector on productivity. To correct for potential endogeneity between concentration and productivity we implement a credible instrumental variable approach to isolate the effect of the exogenous component of market concentration trends on local productivity. Our main results show negative and statistically significant relationship between local industry concentration and firm-level productivity. A decline in one standard deviation in local industry concentration implies an economically meaningful increase of 0.014 standard deviations in TFPR. Local industry concentration also has heterogeneous effects on productivity for different types of industries, while its impact varies by different levels of exposure to international markets. Interestingly, our results show that the effect of being more exposed to the international markets compensates, and in most cases, reverses the negative effects concentration has on productivity. These results are robust to specifications using panel data, controlling for a proxy of markups and employing alternative measures of local industry concentration. Based on these results, it is important that government policies aimed at fostering productivity target those industries and regions more affected by the lack of local competition and less exposure to international markets. Keywords: Productivity, Market Concentration, Instrumental Variables JEL codes: D24, D43, F12, F14, L23 * The authors want to thank Gladys Lopez Acevedo and Mariana Vijil who provided substantial comments to this version of the paper. Also, the authors would like to thank discussants and participants at the various conferences and workshops, in particular LACEA 2017, the 5th Congress on Mexico organized by the Universidad Iberoamericana and the Jobs for Development conference in Bogota. We also thank Natalia Volkow and Liliana Martinez from the Microdata Center of INEGI for their support with the data use in compliance with the confidentiality requirements set by the Mexican Laws. The findings, interpretations, and conclusions in this paper are entirely those of the authors and do not necessarily represent the views of the organizations to which the authors are affiliated, not the World Bank Group, its Executive Directors, or the countries they represent. † World Bank. E-mail: crodriguezc@worldbank.org (corresponding author). ‡ UNDP. E-mail: luis.lopez-calva@undp.org § UC Davis. E-mail: obarriga@ucdavis.edu 1. Introduction The promise of income convergence between developing and developed countries has been lost in what has been denominated the middle-income trap (Garrett, 2004). Despite substantial investment rates, improvements in human capital, and the demographic dividend, most middle-income countries have not been able to catch-up developed economies. Most studies have concluded that the main factor behind the lack of sustained income growth for developing countries has been stagnant productivity growth (OECD, 2014). The economic literature has identified a multiplicity of factors associated with patterns of low productivity growth which include the weak business environment, factor misallocation, distortive fiscal policy, insufficient human capital and concentrated market structures. Among these and other determinants, using enterprise survey data, Cusolito and Maloney (2018) show that product market competition is by far the largest contributor to variations in revenue total factor productivity (TFPR) dispersion. On this regard, most studies that have examined the links between competition and productivity have focused on cross-country analysis of aggregate level of productivity growth or have defined market concentration at the national or industry levels. Only with a few exceptions for developed economies and specific industries (Nickell, 1996; Schmitz, 2005; Dunne, Klimek and Schmitz, 2008; De Loecker, 2011), this research question has received little attention on developing countries, particularly due to the lack of adequate data to study this phenomenon rigorously. Thus, due to the nature of the available data, this paper provides an important contribution to this literature, particularly in the context of developing countries. The paper explores the links between market concentration and productivity, delving into the heterogeneity of the effect of concentration due to trade exposure. The case of Mexico is particularly interesting, given that the country implemented large market reforms aimed to open to foreign trade and liberalize the economy over the past 30 years. It is a country that has diversified its economy and which has experienced a significant human capital upgrade, but that, after years of implementing these ambitious set of reforms, still lags in TFP growth when compared to other advanced economies (Lopez Cordova and Rebolledo, 2016; Levy, 2018, World Bank, 2018). The Mexican case is striking. If productivity had kept pace with other developing economies, relative income per capita would have been 24 per cent higher in 2008 vs. 1960 (see Busso et al., 2012). In 1990, according to WDI data, Mexican had higher income per capita levels than Korea, Chile, and Uruguay, while those three countries were well above Mexico by 2017, all of them in the group of high-income countries today. Besides low TFP growth, despite being one of the countries in the world with the largest number of trade agreements, Mexico also lags among other OECD members in product market competition indicators (Conway, Janod, and Nicoletti, 2005). This makes Mexico an emblematic case of study to understand the effects of competition on productivity in a context of trade liberalization. 1 The analysis takes advantage of 20 years of establishment census data to understand to what extent local industry concentration explains the stagnation in productivity growth in the manufacturing sector in Mexico. Local industry concentration is approximated using a Herfindahl-Hirschman Index (HHI) at the 3-digit sector level for fifty-four metropolitan areas. Although a preferred measure of productivity would be TFPQ, the analysis uses TFPR given the lack of price data in the establishment census data. We estimate a linear model at the firm level using five rounds of the Economic Census in Mexico (1994, 1999, 2004, 2009, 2014). To address potential endogeneity concerns of the relationship between local market competition and firm-level productivity, our strategy is based on an instrumental variable approach strategy following a Bartik (1991) procedure. Our identification assumption relies in the fact that changes in 3-digit industry concentration at a national level impact firm-level productivity only through its effect in the variation in local market concentration in the same industry. Our instrument produces a counterfactual industry concentration index in a metropolitan region by applying national changes on sectoral concentration for each industry, using as base year each metropolitan area-industry concentration index from the 1994 Economic Census. By design, our instrument cannot be influenced by selection into metropolitan areas; rather, it isolates the component of change in the local concentration that is driven by national trends, such as changes in the national policies or industry specific incentives. Our main results show negative and statistically significant relationship between local industry concentration and firm-level productivity. Estimates from our preferred model show that a decrease in one standard deviation in local industry concentration implies an increase of 0.014 standard deviations in TFPR. These results are also robust to other measures of concentration such as the share of the total revenues by the largest five firms within a metropolitan area. Local industry concentration has heterogeneous effects for different types of industries and, very important in the Mexican context, also vary by different levels of exposure to international markets. On this later point, our results show that the negative effects of higher market concentration on productivity are neutralized for firms in local areas more exposed to international markets; implying that the relevant competition for those firms may not be local but international. It is important to mention that, since our definition of firm-level TFPR is constructed from the residuals of the revenue function, firms with a high residual are considered to have a higher level of productivity. However, since TFPR can be separated into TFPQ and changes in firm-level prices, and since prices are typically a function of markups and marginal costs, it may be possible that even after including sector and metropolitan area fixed effects to control for local demand and other local factors that may affect market power, this TFPR residual still captures the mark-up level a firm has.5 If that is 5 Differently to other studies examining the relationship between productivity and competition due to trade liberalization which also consider changes in marginal costs due to potential reduction in inputs relevant for the production function of firms, this should not be an issue in this study since marginal costs of firms should not be affected substantively due to changes in local industry concentration. 2 the case, a firm with a high residual, given the level of the control variables, can report a higher productivity due, in part, to its advantage in price, and not solely for its productivity. This would in turn imply that, in the presence of firm-level market power, our results are a lower-bound of the potential effects of changes in local industry concentration. To explore the extent to which this may be an issue, we conduct two robustness tests. First, we produce a productivity measure through a robust estimation of the production function (Olley and Pakes, 1996) using a panel of firms for the 2009 and 2014 rounds of the Economic Census. This allow us to control for endogeneity of inputs, correct for selection (exit) and deal with unobserved (quasi-) permanent differences across firms. Secondly, we estimate our preferred model using TFPR from the five rounds of the census, but we introduce as a control a measure of firm-level markup defined as revenues over total costs. For both cases we show that our main results are robust to the use of the Olley and Pakes correction of the productivity measure, as well as to the inclusion of a proxy for firm-level markups. This paper contributes significantly to the literature in several ways. First, it takes advantage of microdata from several rounds of establishment census to expands the limited literature on the long- term dynamics between firm-level productivity and market concentration in a developing country setting, where this type of data is rarely available. Secondly, differently to most of previous studies which examine the relationship between competition and productivity driven by changes in external competition through trade liberalization, our study focuses the productivity impacts of variations in local industry concentration. Third, our results have significant policy implications in terms of the focus of policies aiming at improving productivity. Given the heterogeneity of impacts of local industry concentration on productivity by sector and by level of local exposure to international markets, it is important that government policies and investment are targeted to sub-sectors and regions less exposed to trade and more affected by to lack of local competition. The rest of the paper is organized as follows. Section 2 presents a brief literature review on the links between industry concentration and productivity. Section 3 describes the main sources of data and presents key stylized facts. Section 4 discusses the empirical strategy. Section 5 presents the main results, including heterogeneous effects and robustness checks. And Section 6 concludes. 2. Literature review The literature on endogenous growth shows that in the standard model of endogenous technological change there is a rent dissipation effect (Romer, 1986; and Aghion and Howitt, 1992). This result implies that increments on product market competition in the intermediate producer’s level reduces expected future profits from innovations and the rate of technical change. However, Aghion et al. (2001) extend this basic framework showing that a positive relationship between market competition and growth might still exist in the case of an oligopolistic firm that uses innovation to temporarily shield from competition; in this manner, incentives to innovate remain present and become stronger 3 the closer a firm is to the technological frontier.6 Thus, a positive relationship between product market competition and growth is not an unambiguous implication of theoretical work. Most studies examining the relationship between competition and productivity focus on the aggregate level of productivity growth but do not explore the effect on individual firms. With a few exceptions for developed economies (Galindo and Schiantarelli, 2005; Schmitz, 2005; Dunne, Klimek and Schmitz, 2008; De Loecker, 2011), this research questions has received little attention on developing countries where lack of adequate data appears to be the main limitation for the existence of such studies. In studies that rely on firm-level data, Nickell (1996) finds that competition, measured by increased numbers of competitors or by lower levels of rents, is associated with higher total factor productivity growth. The author measures competition using a Lerner Index, or price cost margin averaged across firms within the industry. This measure has several advantages over indicators such as market share or the Herfindahl concentration index as those measures rely more directly on precise definitions of geographic and product markets. This is important, since in the case of the U.K. firms operate in international markets, causing market concentration measures based only on U.K. data to be misleading. Similarly, Aghion et al. (2005) replicates Nickell (1996) Lerner Index and combines it with policy changes7 to address the endogeneity between competition and innovation. Their results show a positive effect of product market competition on productivity growth, particularly at low levels of competition. Analyzing the effect competition on productivity growth, Aghion et al. (2008) combines three different datasets in order to compare product market competition in South African manufacturing firms and sectors to their corresponding sectors worldwide. The author find that markups are significantly higher in South African industries than in their worldwide counterparts, causing that average profitability margins are twice as large; these larger markups translate into a lower productivity growth rates. Back of the envelop calculations reveal that a reduction of 10 percent in markups would lead to an increase in productivity growth between 2 and 2.5 percent per year. In a related study, Ospina and Schiffbauer (2006) captures competition pressure directly from firm managers self- reported assessment using the World Bank Enterprise Survey8. Estimating markups as sales over operating costs, the authors find that firms with markups 20 percent higher than the average have 1.2 percent lower TFP levels and 8 percent lower labor productivity. 6 It follows, in their model, that an increase in competition involves an innovation-tradeoff: it reduces monopoly rents, but enhances the incentive to innovate in order to escape competition 7 Thatcher era privatizations, the EU Single Market Programme,12 and the Monopoly and Merger Commission investigations that resulted in structural or behavioral remedies being imposed on the industry 8 The question explicitly asks “How important is pressure from domestic competitors on key decisions about your business with respect to reducing the production costs of existing products or services?”. The answer to this question go from 1 to 4 where 1 is low importance and 4 high importance. 4 When international competition on local markets is included, results from the literature show that high levels of local concentration might not end up causing a negative effect on productivity. Two mechanisms have been proposed to explain this. The first is self-selection: Only the more productive firms engage in export activities and can compete internationally. The second mechanism, called the ‘learning-by-exporting’ hypothesis states that firms that enter the export markets gain access to technical expertise from their buyers, which non-exporters do not have; this allows them to improve their efficiency level. While the self-selection hypothesis has been relatively easy to prove empirically (Clerides, et al., 1998; Bernard and Jensen, 1999; Van Biesebroeck, 2006; and, Alvarez and Lopez, 2004), the evidence on the learning hypothesis has been less clear-cut since the detailed information required to isolate changes that occur within firms as it starts to exporting are hard to come by. One exception is the identification strategy used by Loecker (2007) who takes advantage of the massive entrance of Slovenian firms into the export markets between 1994 to 2000. The author is able to identify the instantaneous and future productivity gains upon export entry. Using matching sampling techniques to control for self-selection into export markets, and Olley and Pakes (1996) procedure to estimate productivity, his results reveal that new export firms become more productive, with a productivity gap that widens over time and that is higher for firms exporting towards higher income regions. Similarly, in a more controlled setting, Atkin, Khandelwal, and Osman (2016) find similar results by conducting a randomized experiment to generate exogenous variation in the access to foreign markets for rug producers in Egypt. The authors find large improvements in quality as well as an increase in profits between 16-26 percent. 3. Data and stylized facts on market concentration and economic productivity in Mexico Economic Census Data Our analysis uses detailed establishment level data from Mexico’s Economic Census (MEC) collected every five years by the National Statistical Office (INEGI). The census measures economic activity taking place in private establishments with a fixed location. The MEC collects information on firms’ sales, value added, number of workers, types of contractual arrangements, labor remunerations, and value of fixed capital. As explained, one of the objectives of the paper is to learn about the joint effect of international competition and high levels of concentration. For this reason, although the database covers all non-agricultural activity, we focus on the manufacturing sector since it is the sectors that were affected in a greater degree by the exports promotion strategy. We use five rounds of the census taken from the years 1994, 1999, 2004, 2009, and 2014. Additionally, MEC allows to construct a panel of firms for the data of 2009 and 2014, but only for that period. 5 In 2014, the Census collected information from around 4 million private establishments; of those, around 11 percent belonged to the manufacturing sector. At the 3-digits level, the Census divides the manufacturing sector in 21 different types of industries.9 Our final sample uses 20 different economics sectors after excluding the firms in the Petroleum and Coal Products industry; we exclude this sector given its need for large and localized capital invests by very few firms. In our preferred sample, we use a total of 229,865 establishments in 2014; a detailed list of the number of manufacturing establishments per sector, at the 3-digit, can be found in Annex 1. Estimation of the production function and TFPR trends We consider a Cobb-Douglas production functions to estimate productivity at the firm level.10 For each firm , we use as inputs the logarithm of the total capital, labor, and the total cost of inputs. In our main specification, we include fixed effect for year (), metropolitan area (), and sector (): ln(, ) = 0 + 1 , + 2 , + 3 , + + + + , (1) 11 We estimate the TFP as the residual from the previous equation: = exp( ̂, ) . The previous equation is estimated for both the logarithm of value added and the total sales. As has been thoroughly explored in the literature, there is a double causality between input selection and unobserved productivity variables: more productive firms choose higher quality inputs and combine them in a more efficient way than less efficient firms, moreover, this productivity determines the probability of exit of a firm in following periods. This endogeneity can still be present after including fixed effects. When possible, we solve for this endogeneity problem by implementing an Olley and Pakes (1996) correction. This transformation only requires a monotonic relationship between a firm-level decision variable and the unobserved firm-level state variable “productivity”. Unfortunately, the Mexican Economic Census only allows us to construct a panel of firms for the censuses of 2009 and 2014. When using all the census rounds, we include year, metropolitan area, and sector fixed effects to account as best as possible for the relation between input choice and productivity, as well as unobserved market conditions that affect the level of productivity of the firm Table 1 shows the evolution of TFP by sector at the 3-digit level within the manufacturing industry using the 1994 Economic Census as base-year. Only a few sectors present a reduction on their 9 The latest round of the Census follows the NAICS classification system. We converted the codes used in previous rounds into this classification. 10 Results using a Translog production function will be implemented on a next iteration of the paper, but we do not expect to see any changes in the paper’s main conclusion. 11 We eliminate the results on the top and bottom deciles of the distribution since their predicted TFP is too high. The results do not change after using the variance of the estimated TFP as a weight. 6 aggregate productivity when calculated at the national level: Wood products, paper manufacturing, nonmetallic mineral products manufacturing, and computer and electronics; nevertheless, during the overall period the evolution of the Mexican manufactory sector productivity is stagnant. Table 1. Evolution of TFP by industry in the manufacturing sector in Mexico Evolution of TFP by sector Descriptive 1994 base year Stats 1999 2004 2009 2014 Avg. S.D Food Manufacturing 0.985 0.980 0.982 1.046 1.092 1.468 Beverage and Tobacco Product Manufacturing 1.000 0.958 0.983 1.074 1.152 1.289 Textile Mills 1.009 1.085 0.995 1.049 1.158 1.604 Textile Products Mills 0.921 1.046 0.989 1.053 1.166 1.247 Apparel Manufacturing 0.982 0.985 0.991 1.054 1.202 1.122 Leather and allied manufacturing 0.996 0.958 0.888 1.043 1.190 1.216 Wood products 0.965 0.964 0.934 0.961 1.092 0.816 Paper manufacturing 1.006 0.986 0.964 0.950 1.118 1.102 Printing and related support activities 0.998 1.001 0.980 1.002 1.086 0.611 Chemical manufacturing 0.913 0.909 0.897 0.921 1.115 1.923 Plastic and rubber products 1.018 1.020 1.005 1.078 1.105 1.890 Nonmetallic mineral products manufacturing 1.014 1.011 0.977 1.008 1.098 0.661 Primary metal manufacturing 0.982 0.981 0.946 1.003 1.095 0.642 Fabricated metal products 1.002 1.001 0.976 1.040 1.084 0.707 Machinery manufacturing 0.964 0.973 0.949 1.036 1.092 0.992 Computer and electronics 1.089 1.043 1.042 1.067 1.101 0.635 Electrical equipment, appliance, and components 0.998 1.000 0.984 1.010 1.093 0.634 Transport equipment 1.023 1.019 1.040 1.279 1.171 4.218 Furniture and related products 0.991 0.974 0.954 1.005 1.091 0.693 Miscellaneous manufacturing 0.931 0.940 0.898 0.934 1.114 1.063 Note: The TFP at the firm level was calculated using equation (1) Source: Authors calculations bases on Economic Census There is also substantial gap in productivity levels the top and bottom performers across industries and across metropolitan areas in Mexico. For some sectors, this gap appears to be very widening overtime and present across industries over a period of twenty years, see Figure 1 (also see Levy, 2018). Figure 1. Evolution TFP in selected industries, 1994-2014 7 Source: Authors calculations based on Mexican Manufacturing Census. Note: Selected sectors only. Figures shows the percentiles of the TFP for selected sectors. Dynamics of industry concentration To measure local industry concentration, we use the fifty-four Mexican Metropolitan Areas defined by the National Council of Population (CONAPO) in 2000.1213 We construct several concentration indexes at the metropolitan area level; these include HHI using the total sales, as well the percentage of total production generated by the 5 largest firms.14 When industries concentration levels are calculated at a national level, evidence suggests that industry concentration in Mexico is low; however, high levels of variation are still present. Supporting evidence for this is presented in Table 2, where concentration is calculated using the HHI index at the 3-digit sector level. This table also shows how even as concentration levels are low, some sectors observe substantial changes in relative terms. A first glimpse to the concentration levels across regions in Mexico also reveals wide differences across industries and regions. Figure 2 shows the HHI index for selected industries.15 12 During the period of study additional metropolitan areas were created. We use only the ones that were defined in 2000; these areas did not suffer any modification. 13 Officially, the CONAPO definition of metropolitan areas in the year 2000 included 55 areas. However, due to data limitations and the size of the metropolitan area we included it inside the Puebla Tlaxcala region. Annex 2 shows a list of all the metropolitan areas. These areas are defined as a group of “two or more municipalities in which a city with a population of at least 50,000 is located, this urban area extends over the limit of the municipality that originally contained the core city incorporating, physically or under its area of direct influence other adjacent predominantly urban municipalities, all of which either have a high degree of social and economic integration. The definition also includes municipalities that given its characteristics are relevant for the development of urban planning policies. Additionally, a metropolitan area is defined for those municipalities with a million or more inhabitants, as well as those cities with 250 thousand inhabitants that share urban processes with cities in the United States.” Annex 3 presents the total number of manufacturing companies by metropolitan area over the 1994 and 2009 period. 14 The data in Mexico does not allow us to calculate a mark-up Lerner index since we do not have price for inputs. 15 The results for all the industries can be found in an online Annex. 8 Table 2. Evolution of HHI in Mexico (National) Evolution of HHI index 1994 1999 2004 2009 2014 Food Manufacturing 0.005 0.004 0.005 0.005 0.006 Beverage and Tobacco Product Manufacturing 0.025 0.020 0.022 0.031 0.032 Textile Mills 0.004 0.006 0.009 0.011 0.013 Textile Products Mills 0.070 0.012 0.019 0.019 0.013 Apparel Manufacturing 0.003 0.002 0.004 0.005 0.007 Leather and allied manufacturing 0.004 0.004 0.006 0.006 0.010 Wood products 0.039 0.009 0.007 0.009 0.007 Paper manufacturing 0.015 0.013 0.015 0.012 0.013 Printing and related support activities 0.006 0.005 0.004 0.005 0.007 Chemical manufacturing 0.015 0.012 0.022 0.030 0.019 Plastic and rubber products 0.009 0.010 0.006 0.005 0.004 Nonmetallic mineral products manufacturing 0.011 0.010 0.011 0.009 0.009 Primary metal manufacturing 0.035 0.043 0.052 0.063 0.068 Fabricated metal products 0.006 0.013 0.011 0.014 0.004 Machinery manufacturing 0.026 0.018 0.020 0.022 0.024 Computer and electronics 0.025 0.169 0.056 0.020 0.013 Electrical equipment, appliance and components 0.013 0.016 0.016 0.016 0.022 Transport equipment 0.078 0.053 0.050 0.036 0.029 Furniture and related products 0.004 0.003 0.004 0.005 0.006 Miscellaneous manufacturing 0.008 0.008 0.008 0.010 0.010 Note: HHI index uses the total value of production of each manufacturing establishment recoded in the Census. Source: Authors calculations bases on Economic Census Figure 2 Evolution of the HH index in selected industries, 1994-2014 Source: Authors calculations based on Mexican Manufacturing Census. Note: Selected sectors only. 9 When data is analyzed at the metropolitan area level there is a wide range of changes across industries during the period 1994-2014. Figure 3 below shows how for each sector the concentration at the metropolitan area level evolved over time. In 11 out of 20 sectors in total, the median HHI increased.. Figure 3. Relative change HHI index by industry, 1994-2014 Source: Author calculations using Mexican Manufacturing Census Note: The HHI index was created using the total value of production of each manufacturing establishment recoded in the Census. Outliers with changes higher than 10 were not included. Trade and external exposure data For the years of analysis 2004, 2009, and 2014 we introduce a control for the level of external exposure to international markets at the Metropolitan Area. We use data from the Mexican Atlas of Economic Complexity to test how exposure to international markets eliminates the negative effects of market concentration on productivity. This instrument is interesting since it allows us to know economic patterns and the productive ecosystems at a relatively small geographical location. The Atlas of Economic Complexity is the best approximation available to local level trade patterns. It estimates the amount of exports and imports at a 3-digit sector level that are demanded or generated from each metropolitan area the that location. It uses as raw data the customs registries that cover all transactions independently of their destination and then assigns it to municipalities. 10 Most of the exports in Mexico are produced inside the Metropolitan Areas, (see Table 3). This is expected to happen since, by construction, metropolitan areas concentrate most of the formal economic activity. We use this to our advantage by relying on the fact that there is little external competition taking place in the remaining municipalities of the country. Table 3. Share of exports produced inside the Metropolitan Areas, 2004-2014 2004 2009 2014 Food Manufacturing 97.10% 97.47% 97.58% Beverage and Tobacco Product Manufacturing 99.92% 99.92% 99.80% Textile Mills 80.28% 90.16% 91.07% Textile Products Mills 97.17% 96.62% 97.45% Apparel Manufacturing 87.61% 87.41% 86.54% Leather and allied manufacturing 97.42% 95.73% 90.58% Wood products 98.07% 98.25% 98.85% Paper manufacturing 95.63% 95.34% 96.13% Printing and related support activities 93.12% 98.43% 99.08% Chemical manufacturing 95.76% 98.56% 98.67% Plastic and rubber products 97.25% 94.51% 92.63% Nonmetallic mineral products manufacturing 88.38% 93.50% 92.57% Primary metal manufacturing 97.19% 90.24% 92.92% Fabricated metal products 89.87% 92.19% 93.05% Machinery manufacturing 95.21% 96.47% 96.20% Computer and electronics 94.41% 95.05% 94.54% Electrical equipment, appliance and components 95.11% 93.30% 94.88% Transport equipment 100.00% 99.81% 96.99% Furniture and related products 98.40% 92.78% 95.67% Miscellaneous manufacturing 97.41% 97.77% 90.09% Note: Author Calculations using The Mexican Economic Atlas 4. Empirical Strategy To study the effects of competition on economic productivity in Mexico we regress the level of 3- digit sector concentration (approximated with HHI) for every metropolitan area against the level of productivity of each firm located in the relevant metropolitan area. The model estimated includes as controls the share of each sector at the 3-digit level and the change in the number of firms at the same level of disaggregation. To account for unobservable time invariant characteristics that affect each sector annually, we also include year and sector specific fixed effects. We estimate the effect of market concentration on productivity in the manufacturing sector. Equation (1) describes the model we estimate. The variable ,, denotes the concentration level for industry , in the metropolitan region , in a given year . Fixed effects are included to account for sector and metropolitan area-specific shocks; however, they limit the number of controls that can be used in the regression. As controls, we include the sector share of total production value in each region, as well 11 as the percentual change on the total number of firms on each region/sector. These controls are included to account for sector importance on economic region’s overall activity. 3 TFP,, = 0 + 1 ,, + ∑ ,, + + + + , (2) =2 It is clear that the inclusion of controls and fixed effects in equation (2) is not enough to establish the causal relationship between concentration and productivity. The issue is that local market concentration for each industry may not be exogenous, and thus it may be correlated with other characteristics that affect the concentration level of the industry in a given metropolitan area. For example, initial levels of concentration in a metropolitan area may affect the local government incentives and the local resources that are demanded by the same sector. To circumvent this concern about potential reverse causality from the endogenous sorting of firms across metropolitan areas we implement a Bartik instrument strategy. Our instrument predicts the concentration level in a metropolitan region by applying national changes on sectoral concentration for each industry, using as base year the 1994 concentration levels, arguably exogenous for the rest of the analysis. By keeping constant the concentration level across metropolitan areas at the level of 1994, we foreclose the possibility that local specific policies/resources determine future changes in local levels of concentration.16 By design, our instrument cannot be influenced by selection into metropolitan areas; rather, it isolates the component of change in the local concentration that is driven by national trends, such as changes in the national trade policies or industry specific incentives. Moreover, out instrumental variable procedure isolates the industry-specific characteristics and the market structures that determine a “natural” concentration level. In this way, the total change in concentration in a geographic area j can be divided into an “exogenous component” recreated using the instrument and deviations from this predicted change. In this manner, our instrument allows us to only use the exogenous source of variation (see for instance Boustan et al., 2008). By design, our instrument isolates the component of change in the local market that is driven by national trends, such as changes in the competitive environment and in labor market. Formally, our instrument is defined for each sector on a specific metropolitan area, at a period T, as: _,, = ,,=0 × , Where ,,=0 is the concentration level for sector j at the metropolitan area level for the first Economic Census available, while , represents the growth rate for the same sector j at the national level between period = 0 and . There is a wide variation in the level of concentration across metropolitan areas. When constructing our instrument, however, the level of the resulting 16 Alternatives estimation strategies can include focusing only in the impact of the entrance of China to the WTO. This strategy relies on measuring the exposure of Mexico to external competition coming from China, and using a Bartik approach to control for the exposure of Mexican sectors to other international competitors. 12 shares differs from the observed data. This is in line with having an instrument that does not simply follows the trends observed in the actual data.17 We find a strong and positive relationship between the two measures suggesting that much of the change in local industry concentration from 1994 to 2014 was driven by national trends rather than local factors. Figure 4 shows the results of the first-stage regression between actual and predicted HHI index coefficients for selected sectors. Further tests show that that trends in changes in productivity are not correlated with initial concentration levels, a key identification assumption. Figure 4. First-stage regression: Relationship between actual and conterfactual HHI at the metropolitan area level in selected industries, 1994–2014 17This is indicative that that areas that initially had a lot of concentration for a specific sector in the initial year did not drive the national change in the indicator since other areas with initial smaller concentration became proportionally more concentrated. 13 5. Main results: Market concentration lowers productivity Our main results show a negative, statistically significant, and economically meaningful relationship between concentration and economic productivity. As Table 4 shows, these result hold, and become more pronounced following the IV strategy, confirming that in the absence of instrumental variables we would underestimate the true effect of local industry concentration on productivity.18 All the results hold when other measures of concentration are used. Table 4. The effects of market concentration on economic productivity Manufacturing sector in Mexico, 1994-2014 (1) (2) (3) (4) Productivity HHI -0.074*** -0.069*** -0.075*** -0.097*** (0.003) (0.003) (0.004) (0.005) Constant 1.069*** 1.070*** 1.059*** 1.052*** (0.001) (0.001) (0.001) (0.002) Partial R-squared NA NA NA 0.4938 Kleinbergen-Paap statistic NA NA NA 684,055 Controls No Yes Yes Yes FE year No No Yes Yes FE Sector No No Yes Yes IV No No No Yes Observations 867,126 687,356 687,356 687,224 R-squared 0.001 0.001 0.001 0.002 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Heterogeneous effects by industry An source of heterogeneity, that is hidden in the regressions shown in Table 4, is how differences in technologies in each industry affect the impact concentration has on productivity. To study this effect, we estimated Equation (2) at the sector level. Figure 5 shows a negative impact of concentration on productivity in 10 out of 20 sectors. Furthermore, after implementing a Holm correction to account for Family Wise Error Rate we find no evidence that false rejection due to Multiple Hypothesis Testing. 18 In exploratory work, we also find a non-linear effect of concentration on productivity when a quadratic term on the effect of the HHI is included. This non-linearity is indicative that as the concentration increases total productivity falls but at a lower rate, reflecting a competition/innovation tradeoff. This tradeoff is caused by oligopoly rents on innovation, that enhance the incentive to innovate to escape competition, but as the pressure of competition disappears, the incentives to innovate also dwindle, as developed in Romer, 1986; and Aghion and Howitt, 1992) 14 Figure 5. Effects of market concentration on economic productivity, at the 3-digit category Note: Statistically significant results shown on grey Source: Authors calculations bases on Economic Census The economic significance of these coefficients varies by sector. Our results imply that a change of one standard deviation in the concentration index in the Leather Manufacturing sector implies a change of -0.071 standard deviations in the productivity level. Consider the stagnant levels of productivity in Mexico in the last 20 years, the results of the regressions show that concentration was an important contributor for this phenomenon. Table 5 shows the economic impacts for all those sectors that present a statistically significant result. Exploring the characteristics that make that specific sectors respond differently to concentration is a topic that requires further research. Possible explanations for this include sector-specific capital to labor ratios that favor concentration19 and competition from close substitutes. We focus on the former by exploring how exposure to international markets changes the incentives that keep a concentrated industry competitive. 19We do not include the Petroleum and Coal Products industry in the results of this paper given the low number of establishments, and its geographical concentration, in the sample. This sector is an extreme of a situation where market characteristics force situation with very few companies. The low number of establishments seriously damps our capacity to make significant conclusions about the sector. 15 Table 5. Economic significance at the sector level Sector Economic Significance Food Manufacturing -0.016 Apparel Manufacturing -0.048 Leather and allied manufacturing -0.071 Wood products -0.014 Paper manufacturing -0.025 Chemical manufacturing -0.009 Fabricated metal products -0.028 Machinery manufacturing -0.040 Transport equipment -0.018 Miscellaneous manufacturing -0.022 Note: Authors calculations based on the Economic Census Preferred specification used: IV Heterogeneous effects by exposure to international markets Next, we explore whether the effects of concentration on productivity are somewhat neutralized when the relevant market of manufacturing firms is international, an analysis that is also new at this level of empirical rigor for developing country contexts. For this, we take advantage of data from the Atlas of Economic Complexity to measure sector-specific exposure to international markets in each metropolitan area. Our preferred measure of exposure to international markets is the share of total exports, of each sector, that are produced in each metropolitan area. This variable has an important level of variation that is useful when we estimate how exposure to international markets acts to counter act the negative effects of local concentration. As can be seen in Table 6, even when the average share is similar across sectors, there is a high level of dispersion between metropolitan areas. There are several measures of exposure that we can construct and use in this study. We use the share of exports for each sector produced in each metropolitan area as it reflects how competitive the manufactures in each metropolitan is when compared to the other regions in Mexico. Possible reasons for differential levels of productivity may include how close the region is to the border with the United States or to important ports, levels of human capital, and the effects of agglomeration. A second measure that is usually proposed is the import share; we decided not to use it as our preferred measured because it is difficult to account what percentage of the imports that enter to a metropolitan area are used into products that will be later consumed in the internal or international markets.20 20 Estimating the model of equation (2) using the import share does not change qualitatively our results. 16 Table 6. Average share of exports from each Metropolitan Area Mean Median sd max min Food Manufacturing 1.80% 0.28% 0.044 32% 0% Beverage and Tobacco Product Manufacturing 1.85% 0.00% 0.073 64% 0% Textile Mills 2.29% 0.42% 0.041 21% 0% Textile Products Mills 1.80% 0.13% 0.042 26% 0% Apparel Manufacturing 1.61% 0.19% 0.036 20% 0% Leather and allied manufacturing 1.88% 0.12% 0.037 22% 0% Wood products 1.82% 0.05% 0.080 65% 0% Paper manufacturing 1.83% 0.18% 0.040 22% 0% Printing and related support activities 1.79% 0.22% 0.064 49% 0% Chemical manufacturing 1.88% 0.23% 0.054 38% 0% Plastic and rubber products 1.79% 0.13% 0.033 16% 0% Nonmetallic mineral products manufacturing 1.69% 0.15% 0.043 35% 0% Primary metal manufacturing 2.21% 0.32% 0.040 23% 0% Fabricated metal products 1.70% 0.10% 0.033 17% 0% Machinery manufacturing 1.89% 0.13% 0.034 22% 0% Computer and electronics 2.93% 0.10% 0.064 31% 0% Electrical equipment, appliance and components 2.28% 0.04% 0.045 21% 0% Transport equipment 1.98% 0.04% 0.049 30% 0% Furniture and related products 1.77% 0.07% 0.058 44% 0% Miscellaneous manufacturing 1.76% 0.03% 0.055 38% 0% We estimate the model described in equation (2) at the firm level. We use the percentage of the total exports of each sector, that are produced in each metropolitan area , at time . To study how exposure to international markets changes the negative effect of concentration on productivity, we interact our variable for exposure with the concentration that we used in the previous sections of the paper. TFP, = 0 + 1 , + 2 ℎ, + 3 ℎ, ∗ , + ∑ , + + + + , (2) Table 7 shows that concentration has a negative impact on productivity, but as the share of exports goes up, this effect is neutralized. For instance, results in the column suggest that at certain level of local competition, more exposure to external markets will cancel the negative effect of market concentration on productivity. Given the magnitude of the interaction coefficient, one would expect that high exposure to external markets would completely cancel, and even revert, the negative effect of lack of competition at the local level. 17 Table 7. The joint effect of concentration and exposure to the external market on economic productivity in the manufacturing sector in Mexico, 1994-2014 (1) (2) (3) HHI -0.089*** -0.085*** -0.086*** (0.004) (0.004) (0.004) Share exports 0.002 0.003 (0.003) (0.004) Share exports * HHI 0.831*** 0.880*** (0.089) (0.094) Constant 1.070*** 1.071*** 1.073*** (0.001) (0.001) (0.001) Controls No Yes Yes FE No No Yes Observations 557,163 441,478 441,478 R-squared 0.001 0.001 0.001 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 When we expand the estimation to obtain sector-specific results, we find that out of twenty sectors nine present an interaction, between exposure to international markets and concentration, that is positive and statistically significant. Two sectors have a coefficient for the interaction term that is negative, nevertheless, both the concentration and exposure to international market coefficients have the expected sign. Finally, three sectors have a positive and statistically significant coefficient on the concentration term (See Table 8).21 21These sectors are: Textile Mills, Furniture products, and Electrical equipment. When we do not account for the role of exposure to the international markets the first two sectors present a positive, but not significant, coefficient for the concentration variable. 18 Table 8. Sector specific effects of concentration and exposure to international markets on productivity. Share exports * R- HHI Share exports Obs. HHI squared Food Manufacturing -0.102*** (0.008) 0.141*** (0.019) -2.885*** (0.493) 158,321 0.003 Beverage and Tobacco Product Manufacturing -0.003 (0.019) -0.006 (0.086) -0.128 (0.356) 17,144 0.005 Textile Mills 0.125* (0.071) 0.309* (0.176) -1.164 (1.310) 1,914 0.028 Textile Products Mills 0.014 (0.052) -0.087 (0.218) 1.151 (1.093) 5,284 0.004 Apparel Manufacturing -0.108*** (0.030) 0.640*** (0.082) -2.511** (1.037) 24,594 0.009 Leather and allied manufacturing -0.199*** (0.049) -0.159 (0.120) 1.767 (1.110) 13,749 0.008 Wood products -0.083*** (0.027) 0.029 (0.122) 0.875 (1.428) 18,845 0.001 Paper manufacturing -0.030 (0.045) 0.254** (0.122) 3.004*** (0.668) 3,687 0.013 Printing and related support activities -0.056* (0.030) 0.010 (0.019) 1.756*** (0.584) 28,987 0.001 Chemical manufacturing -0.084** (0.037) -0.044 (0.058) 4.135* (2.403) 5,942 0.002 Plastic and rubber products -0.017 (0.040) 0.020 (0.101) 4.338*** (1.251) 9,061 0.002 Nonmetallic mineral products manufacturing 0.008 (0.023) 0.267*** (0.085) 3.253*** (1.225) 23,875 0.008 Primary metal manufacturing -0.066 (0.061) -0.127 (0.331) 1.099 (0.981) 1,666 0.003 Fabricated metal products -0.039*** (0.013) 0.076** (0.036) 3.912*** (0.501) 76,622 0.002 Machinery manufacturing -0.053 (0.047) 0.113 (0.213) 3.855*** (1.336) 4,005 0.007 Computer and electronics -0.028 (0.086) 0.261 (0.260) -0.352 (1.827) 1,111 0.053 Electrical equipment, appliance and components 0.192*** (0.066) 1.500*** (0.266) -3.033 (2.562) 1,952 0.031 Transport equipment -0.219*** (0.044) -1.261*** (0.246) 1.633*** (0.614) 3,250 0.032 Furniture and related products 0.054* (0.030) 0.219*** (0.077) 0.560 (0.666) 28,145 0.002 Miscellaneous manufacturing -0.111*** (0.029) -0.123 (0.103) 4.749*** (1.464) 13,324 0.006 Note: Coefficients displayed horizontally. All regressions include Fixed Effects at the Metropolitan Area level The economic impact of the interactions term depends on the values of the concentration and share of export for each specific industry. Table 9 displays the economic significance of increases on the level of exposure to international markets as we keep the concentration of the sector in the economic area fixed. We can see that the effect of being more exposed to the international markets is able to compensate, and in most cases, reverse the negative effects concentration has on productivity.22 22For example, an increase in one standard deviation in the level of exposure to international markets increases productivity in 0.1 standard deviations in the Chemical sector. 19 Table 9. Economic significance of exposure to international markets at the sector level Sector Economic Significance Food Manufacturing -0.035 Textile Mills 0.018 Apparel Manufacturing 0.028 Paper manufacturing 0.022 Printing and related support activities 0.048 Chemical manufacturing 0.099 Plastic and rubber products 0.010 Nonmetallic mineral products manufacturing 0.064 Fabricated metal products 0.065 Machinery manufacturing 0.220 Electrical equipment, appliance and components 0.073 Transport equipment -0.013 Furniture and related products 0.002 Miscellaneous manufacturing 0.041 Note: A sector is only displayed when statistically significant results are present. Source: Authors calculations bases on Economic Census Robustness tests Olley and Pakes bias correction The first of these robustness tests is to account for the possibility our estimation of each firm productivity is still bias after the inclusion of region, sector and year fixed effects. The correction methods developed on the Industrial Organization literature for this endogeneity problem require the use of a panel at the firm level. Unfortunately, the Mexican Economic Census only allows us to construct a firm-level panel for the last two rounds of the census, depriving us the possibility to implement them in the full dataset. Following the literature, we implemented the Olley and Pakes (1996) correction using as proxy for productivity the firm’s level of investment. This approach allows us to estimate the firm-level productivity and test how the results presented in the previous sections change. Since we only use the last two rounds of the census our sample is reduced to 115,802 firms. Moreover, since our definition of a firms TFP is constructed from the residuals of equation (1). An important element to consider is that the estimation relies on the levels of capital, labor and inputs. This makes that firms with a high residual are considered to have a higher level of productivity. It can be the case that even after including sector and metropolitan area fixed effects this residual still captures the mark-up level a firm has. If that is the case, a firm with a high residual, given the level of the control variables, can report a higher productivity due, in part, to its advantage in price, and not solely for its productivity. This makes that the reported coefficients for the concentration variable are under reported on Table 4 and Figure 5. We explore this when we apply the Olley and Pakes correction 20 using the two latter rounds of the Economic Census. However, at least for the smaller sample used in that case, it does not seem the case. Table 10 compares how the coefficient of the regression used to estimate the TFP under three estimation technics: Olley and Pakes (1996), simple regression with no controls, and simple regression including year and sector fix effects. As Colum (3) shows, the coefficients used in the previous sections of the paper to predict firm-level TFP are, in general, very close to the ones obtained after using the correction technic. Our main results are also robust when we introduce as a control a measure of firm- level markup defined as revenues over total costs.23 Table 10. TFP estimation under different methodologies Olley and Year and sector Simple regression Pakes fixed effects (1) (2) (3) Capital 0.043*** 0.041*** 0.040*** (0.001) (0.001) (0.001) Expenditure 0.848*** 0.853*** 0.850*** (0.000) (0.001) (0.001) Labor 0.155*** 0.154*** 0.162*** (0.002) (0.002) (0.002) Proxy 0.045*** (0.003) FE Year No No Yes FE sector No No Yes FE firm No No No Observations 115,802 115,802 115,802 R-squared 0.971 0.964 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Next, we use the corrected productivity measure to replicate the results presented found in the paper. Table 12 shows how even if the magnitude of the effect changes, their direction and significance are comparable to the results that do not use the corrected TFP. Both coefficients are statistically so it is not possible to affirm, in this smaller sample, how much firm unobserved mark-up being captured by the residual of equation (1) was making that the results reported were a lower bound of the true impact of concentration. 23 See Annex 4. 21 Table 12. Olley and Pakes Correction: The effects of market concentration on economic productivity in the manufacturing sector in Mexico, 2009-2014 TFP corrected (1) (2) by Olley and Pakes Yes No HHI -0.104** -0.117*** (0.043) (0.009) Constant 2.405*** -0.006** (0.013) (0.003) Controls Yes Yes FE Year Yes Yes FE sector Yes Yes Observations 90,457 90,457 R-squared 0.010 0.005 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Extending the results for the role of exposure to international markets using the sample in the last two census, we find that our main results hold (Table 13). However, this is not true when we use the same sample but omit to apply the Olley and Pakes (1996) correction. Table 13. Olley and Pakes Correction: The joint effect of concentration and exposure to the external market on economic productivity in the manufacturing sector in Mexico, 2004-2009 TFP corrected by Olley and TFP not corrected Pakes (1) (2) HHI 0.007 -0.058*** (0.041) (0.009) Share exports 0.460*** 0.240*** (0.078) (0.017) Share exports * HHI -4.906*** 0.234 (0.852) (0.185) Constant 2.371*** -0.030*** (0.012) (0.003) Controls Yes Yes FE Yes Yes Observations 115,802 115,802 R-squared 0.011 0.005 22 Robustness to the use of other concentration measures We also explore the impact of other concentration measures at the metropolitan area level in productivity. Table 13 explores this effect using as concentration measure the share that the five biggest firms of the metropolitan area have on their sector. In general, all of our results hold. Table 13. The effects of market concentration on economic productivity in the manufacturing sector in Mexico using alternative independent variables, 1993-2014 (1) (2) (3) (4) (5) (6) (7) (8) Simple Quadratic HHI -0.052*** -0.047*** -0.053*** -0.062*** -0.115*** -0.099*** -0.071*** -0.022 (0.002) (0.002) (0.002) (0.003) (0.009) (0.010) (0.011) (0.019) Constant 1.085*** 1.085*** 1.075*** 1.071*** 1.098*** 1.096*** 1.079*** 1.063*** (0.001) (0.001) (0.002) (0.002) (0.002) (0.002) (0.003) (0.004) Controls No Yes Yes Yes No Yes Yes Yes FE year No No Yes Yes No No Yes Yes FE Sector No No Yes Yes No No Yes Yes IV No No No Yes No No No Yes Observations 867,126 687,356 687,356 687,218 867,126 687,356 687,356 687,218 R-squared 0.001 0.001 0.001 0.002 0.001 0.001 0.001 0.002 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 As with the previous results, we find heterogeneous effects at the industry level (Table 15). The effects are negative for the same sectors as previous results show. The results present a similar pattern when we calculate their economic impact: Concentration has a negative, and economically significant impact, in eight economic sectors. Considering the stagnant levels of productivity growth in the country, understanding the channels that create, and that can be used to mitigate this negative impact, are a first order policy question. 23 Table 15. Effect of market concentration measured by the total output concentrated by the 5 biggest establishments on economic productivity, at the 3-digit category. Source: Authors calculations bases on Economic Census Table 16. Economic significance at the sector level Sector Economic Significance Food Manufacturing -0.010 Apparel Manufacturing -0.033 Leather and allied manufacturing -0.059 Wood products -0.006 Paper manufacturing -0.018 Printing and related support activities -0.006 Chemical manufacturing -0.005 Computer and electronics -0.009 Source: Authors calculations bases on Economic Census 6. Conclusions: Market concentration lowers productivity, but trade exposure matters With only few exceptions, most middle-income countries have struggled in their quest of sustained economic growth through higher productivity compared to a model based mostly on factor accumulation. The economic literature has seen a proliferation of both theoretical and empirical studies that have aimed at estimating the effects of different factors, such as openness to trade, access to credit, innovation, and factor misallocation, on economic productivity. In the case of market concentration, some studies have focused on the effects of industry concentration on firm productivity ignoring the role of external competition. By using information from economic census 24 and mitigating potential issues of double causality, this paper finds evidence of a concave relationship between industry concentration and economic productivity. We show that there is a level of concentration where appropriation of extranormal rents at the local level would have a positive effect on economic productivity. However, depending the specific sector of analysis, there is a threshold level of industry concentration after which less competition has unambiguous negative effects of economic productivity. The heterogenous effects found in the paper must be taken with caution since further evidence on the channels that drive sector-specific impacts require further analysis. Our preliminary hypothesis is that a sector relative competitiveness with respect to its main export rivals drives most of the heterogenous results caused by the local level of exposure. However, testing this hypothesis is beyond the scope of this paper. Moreover, this study finds that the potential negative effects of market concentration on productivity are neutralized, and depending on the sector reversed, when exposure to external markets exists. This can be interpreted as if local market concentration would only affect economic productivity if the relevant demand for those firms is domestic. These results have important implications for policy since they suggest that programs that seek to boost productivity in developing countries should focus on fostering firm entry, and thus enhancing industry competition, in those sectors and geographical areas that are more exposed to international trade. Our results thus have important implications for public policy, particularly for middle income countries with low levels of economic competition and less exposition to external markets, as these may be experiencing a double negative effect on economic productivity. Thus, policies that aim to reduce the cost of entry for new firms combined with promotion of trade will be the most effective to boost productivity in manufacturing in these types of countries. Our results are in line with those by Atkin, Khandelwal, and Osman (2016) as we find that firms more exposed to external markets are also more productive. We also find that these effects are of greater magnitude for nine (out of 17) sectors at the 3-digit level that concentrate about 72 percent of the total export value in manufacturing in Mexico. 25 References Aghion, P., R. Burgess, S. Redding, and F. Zilibotti, 2005, “Entry Liberalization and Inequality in Industrial Performance,” Journal of the European Economic Association, Papers and Proceedings, vol. 3, no. 2–3, pp. 291–302. Aghion, P., C. Harris, P. Howitt, and J. Vickers, 2001, “Competition, Imitation and Growth with Artuc, Erhan, Gladys Lopez-Acevedo, Raymond Robertson, and Daniel Samaan. 2019. Exports to Jobs: Boosting the Gains from Trade in South Asia. South Asia Development Forum. Washington, DC: World Bank. doi:10.1596/978-1-4648-1248-4. Step-by-Step Innovation,” Review of Economic Studies, vol. 68, pp. 467–492. Aghion, P., and P. Howitt, 1992, A Model of Growth. Through Creative Destruction." Econometrica 60, no. 2: 323-351. Alvarez, R., Lopez, R., 2004. Exporting and performance: evidence from Chilean plants. Working Paper. Indiana University. Atkin, D.; Khandelwal, A. K.; and A. Osman. Exporting and Firm Performance: Evidence from a Randomized Trial NBER Working Paper No. 20690. 201y Barseghyan, L., 2008, “Entry Costs and Cross-Country Differences in Productivity and Output,” Journal of Economic Growth, vol. 13, pp. 145–167. Bartik, Timothy J. "Instrumental variable estimates of the labor market spillover effects of welfare reform." (2001). Barro, R. J., and X. Sala-i-Martin. 1995. Economic Growth. New York: McGraw-Hill. Bernard, A. B. and B. Jensen (1999): “Exceptional exporter performance: cause, effect, or both?” Journal of International Economics, 47, 1–25. Boustan, L.; Ferreira, F., Winkler, H. and E. M. Zolt (2008). The effect of rising income inequality on taxation and public expenditures: evidence from U.S. municipalities and school districts, 1970 – 2000. The Review of Economics and Statistics. Busso M., M. V. Fazio, and S. Levy (2012). (In)Formal and (Un)Productive: The Productivity Costs of Excessive Informality in Mexico. IDB Working Paper Series No. 341. Clerides, S. K., S. Lach, and J. R. Tybout (1998): “Is Learning By Exporting Important? Micro- Dynamic Evidence From Colombia, Mexico, And Morocco,” The Quarterly Journal of Economics, 113, 903–947. 26 Conway, P., V. Janod, and G. Nicoletti (2005). “Product market regulation in OECD countries: 1998 to 2003”. Economics Department Working Paper 419, Paris, OECD. Cordova, J. E. and J. Rebolledo Márquez-Padilla (2016). “Productivity in Mexico: Trends, drivers, and institutional framework.” International Productivity Monitor (30): 28. Cusolito, Ana Paula, and William F. Maloney. 2018. Productivity Revisited: Shifting Paradigms in Analysis and Policy. Washington, DC: World Bank. De Loecker, J. (2007). Do exports generate higher productivity? Evidence from Slovenia. Journal of International Economics 73 (1), 69-98. De Loecker, J. (2011). “Product Differentiation, Multiproduct Firms, and Estimating the Impact of Trade Liberalization on Productivity.” Econometrica 79 (5): 1407−51. Dunne, T., S. Klimek, and J. Schmitz. 2008. “Does Foreign Competition Spur Productivity? Evidence from Post-WWII US Cement Manufacturing.” Working Paper, Federal Reserve Bank of Minneapolis. Evans, W.N. and Kessides, I.N. (1994), “Living by the "Golden Rule": Multimarket Contact in the U.S. Airline Industry,” The Quarterly Journal of Economics, vol. 109, no. 2, pp. 341-366. Galindo, A., F. Schiantarelli, and A. Weiss, “Does Financial Liberalization Improve the Allocation of Investment? Micro-Evidence from Developing Countries,” Journal of Development Economics, 83 (2007), 562–587. Garret, G (2004). “Globalization’s Missing Middle.” Foreign Affairs 83: 84-96. Levy, Santiago (2018). “Under-Rewarded Efforts: The Elusive Quest for Prosperity in Mexico.” Inter- American Development Bank. Melitz, M. J. (2003). ‘The Impact of Intra-industry Trade Reallocations and Aggregate Industry Productivity’, Econometrica 71(6),pp.1695–1725. Nickell, S. (1996), “Competition and Corporate Performance”, Journal of Political Economy, Vol. 104, pp. 724-746. Nickell, S., Vainiomaki, J. and Wadhwani, S. (1994). “Wage and product market power”. Economica, 61 (244), 457-473. OECD, 2014, “Perspectives on Global Development 2014: Boosting Productivity to Meet the Middle- Income Trap.” Olley, Steven G., and Ariel Pakes (1996). “The Dynamics of Productivity in the Telecommunications Equipment Industry.” Econometrica 64 (6): 1263–97. 27 Ospina, S. and M. Schiffbauer (2006). Competition and Firm Productivity: Evidence from Firm-Level Data. IMF working paper. Schmitz Jr., J. A. (2005). “What Determines Productivity? Lessons from the Dramatic Recovery of the US and Canadian Iron Ore Industries Following their Early 1980s Crisis.” Journal of Political Economy 113 (3): 582–625. Shea, J. (1997). “Instrument Relevance in Multivariate Linear Models: A Simple Measure.” Review of Economics and Statistics 79 (2): 348–52. Van Biesebroeck, J. (2006). “Exporting raises productivity in sub-Saharan African manufacturing firms”. Journal of International Economics, vol 67 (2), 373–391. World Bank (2018). Mexico Systematic Country Diagnostics. Washington DC. 28 Annex Annex 1. Sectors at the 3-digit level on manufacturing in Mexico: Total establishments per year, preferred sample. Sector Industry 1994 1999 2004 2009 2014 Total code 311 Food Manufacturing 43,624 57,158 60,478 70,535 79,690 311,485 312 Beverage and Tobacco Product Manufacturing 1,649 2,002 3,728 7,903 11,367 26,649 313 Textile Mills 1,759 1,937 1,040 1,201 874 6,811 314 Textile Product Mills 1,362 1,946 1,813 3,449 3,425 11,995 315 Apparel Manufacturing 12,161 14,748 10,779 15,674 12,397 65,759 316 Leather and Allied Product Manufacturing 5,626 8,070 4,804 7,478 7,192 33,170 321 Wood Product Manufacturing 3,891 7,269 7,540 9,109 8,626 36,435 322 Paper Manufacturing 1,149 1,962 1,805 2,306 2,531 9,753 323 Printing and Related Support Activities 11,051 11,882 11,421 14,396 15,013 63,763 325 Chemical Manufacturing 2,081 3,396 2,468 3,230 3,424 14,599 326 Plastics and Rubber Products Manufacturing 2,672 4,581 3,878 4,231 4,765 20,127 327 Nonmetallic Mineral Product Manufacturing 11,991 15,296 11,641 13,344 12,780 65,052 331 Primary Metal Manufacturing 1,081 1,546 976 785 677 5,065 332 Fabricated Metal Product Manufacturing 23,533 29,645 28,372 36,529 38,842 156,921 333 Machinery Manufacturing 1,702 1,981 2,030 1,887 1,854 9,454 334 Computer and Electronic Product Manufacturing 1,179 670 612 581 599 3,641 335 Electrical Equipment, Appliance, and Component Manufacturing 1,277 1,281 799 1,014 967 5,338 336 Transportation Equipment Manufacturing 871 1,777 1,529 1,597 1,634 7,408 337 Furniture and Related Product Manufacturing 9,591 13,663 10,519 13,317 15,611 62,701 339 Miscellaneous Manufacturing 2,401 5,301 5,258 7,877 7,597 28,434 Source: Authors calculations bases on Economic Census 29 Annex 2. List of metropolitan areas in Mexico, 2000 Source: CONAPO 30 Annex 3. Total number of manufacturing companies by Economic Area Mexico Metropolitan Area 1994 1999 2004 2009 Aguascalientes 2,697 2,983 2,894 3,560 Chihuahua 2,161 2,300 1,792 2,334 Juárez 2,272 2,782 2,400 2,310 Valle de México 45,809 57,036 54,257 65,031 Moroleón-Uriangato 858 1,862 1,329 2,166 León 5,665 8,734 6,991 8,904 San Francisco del Rincón 696 1,231 924 1,572 Acapulco 1,203 2,060 1,890 2,681 Pachuca 1,017 1,378 1,267 2,063 Tulancingo 602 867 810 1,003 Tula 459 641 752 979 Tijuana 2,197 2,694 2,514 2,934 Guadalajara 11,011 17,568 15,129 17,572 Ocotlán 538 713 776 818 PuertoVallarta 426 609 604 851 Toluca 2,720 4,644 4,854 8,181 Zamora-Jacona 618 771 869 1,067 LaPiedad 451 626 646 786 Morelia 2,486 2,995 3,187 3,997 Cuautla 841 1,140 1,346 1,785 Cuernavaca 2,025 2,806 2,934 3,885 Tepic 1,015 1,328 1,214 1,717 Monclova-Frontera 676 832 782 1,217 Monterrey 8,211 10,827 9,468 11,005 Oaxaca 1,974 3,603 2,974 4,248 Puebla-Tlaxcala 9,996 13,094 11,045 15,502 Querétaro 1,726 2,252 2,266 3,225 Cancún 817 1,081 1,033 1,489 Rioverde-Ciudad Fernández 320 468 413 456 San Luis Potosí-Soledad 2,712 3,495 3,091 4,111 Guaymas 394 475 483 763 Villahermosa 1,111 1,574 1,516 1,752 Piedras Negras 264 284 295 425 Tampico 1,715 2,021 1,871 2,327 Matamoros 690 902 918 1,273 Nuevo Laredo 493 756 749 644 Reynosa-RíoBravo 898 1,178 1,227 1,738 Apizaco 499 775 800 1,127 Tlaxcala 1,070 1,692 1,512 2,421 Acayucan 302 379 401 459 Coatzacoalcos 654 907 973 1,189 Minatitlán 508 808 908 1,064 Córdoba 713 977 1,005 1,272 Saltillo 1,955 2,236 1,940 2,484 Xalapa 1,521 2,140 1,967 2,351 Orizaba 912 1,681 1,309 1,656 PozaRica 1,511 1,544 1,398 1,468 Veracruz 1,497 1,945 1,693 1,967 Mérida 3,338 3,220 2,987 3,944 Zacatecas-Guadalupe 697 897 785 1,140 LaLaguna 2,885 3,250 2,961 3,407 Colima-VilladeÁlvarez 856 909 948 1,192 Source: Authors calculations bases on Economic Census 31 Annex 4. Robustness check, controlling by markups defined as revenues over costs Table A.1. Local industry concentration and productivity, controlling by firm-level markups, 1994-2014 TFP_ind TFP_ind TFP_ind fe TFP_ind fe TFP_ind IV TFP_ind IV VARIABLES TFP_ind TFP_ind Controls Controls ALL ALL COMPLEX COMPLEX - hhi_pbt_zone 0.086*** -0.086*** -0.083*** -0.084*** -0.105*** -0.106*** -0.129*** -0.129*** (0.009) (0.009) (0.010) (0.010) (0.010) (0.010) (0.015) (0.015) share_pbt_sect3 0.036*** 0.037*** 0.038*** 0.039*** 0.039*** 0.040*** (0.008) (0.008) (0.009) (0.009) (0.009) (0.009) ch_total_firms 0.002 0.002 -0.004 -0.004 -0.004* -0.004 (0.002) (0.002) (0.003) (0.003) (0.003) (0.003) mark_up 0.003*** 0.003*** 0.002*** 0.002*** (0.000) (0.000) (0.000) (0.000) Constant 1.116*** 1.110*** 1.110*** 1.104*** 1.113*** 1.107*** 1.096*** 1.091*** (0.002) (0.002) (0.002) (0.002) (0.004) (0.004) (0.005) (0.005) Observations 884,823 881,146 701,265 698,323 701,265 698,323 701,127 698,188 R-squared 0.000 0.000 0.000 0.000 0.001 0.001 0.001 0.002 Number of sector_3 20 20 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 32