Full text of Working Papers (Federal Reserve Bank of Richmond) : Diverging Trends in National and Local Concentration, Working Paper 18-15R

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

Diverging Trends in National and Local Concentration  WP 18-15R  Esteban Rossi-Hansberg Princeton University Pierre-Daniel Sarte Federal Reserve Bank of Richmond Nicholas Trachter Federal Reserve Bank of Richmond  Diverging Trends in National and Local Concentration∗ Esteban Rossi-Hansberg Princeton University  Pierre-Daniel Sarte Federal Reserve Bank of Richmond  Nicholas Trachter Federal Reserve Bank of Richmond  February 27, 2019 Working Paper No. 18-15R Abstract Using U.S. NETS data, we present evidence that the positive trend observed in national productmarket concentration between 1990 and 2014 becomes a negative trend when we focus on measures of local concentration. We document diverging trends for several geographic definitions of local markets. SIC 8 industries with diverging trends are pervasive across sectors. In these industries, top firms have contributed to the amplification of both trends. When a top firm opens a plant, local concentration declines and remains lower for at least 7 years. Our findings, therefore, reconcile the increasing national role of large firms with falling local concentration, and a likely more competitive local environment. JEL codes: D22, D43, L16, L22, R12.  1  Introduction  Most product markets are local. The reason is simply that the transportation of goods and people is costly so firms set up production plants, distribution centers, and stores close to customers. In turn, individuals locate in areas where they can obtain the goods they desire. A coffee shop or restaurant in Manhattan does not compete with similar establishments in Seattle, and probably not even in Brooklyn. The wedge in prices created by the inconvenience and monetary cost of buying a product far away from the desired consumption point shields companies in different locations from direct competition. Of course the size of these costs, and therefore the geographic extent of the market, varies by product. Markets are also product-specific. ∗  Rossi-Hansberg: erossi@princeton.edu. Sarte: pdgs4frbr@gmail.com. Trachter: trachter@gmail.com. We thank Simcha Barkai, Tom Holmes, Greg Kaplan, Simon Mongey, Steve Redding, and participants at numerous seminars and conferences for their feedback. We thank Eric LaRose and Sara Ho for outstanding research assistance. The views expressed herein are those of the authors and do not necessarily represent the views of the Federal Reserve Bank of Richmond or the Federal Reserve System.  1  Producers of a particular product are shielded from competition by producers of distinct but related goods and services to the degree that their consumption requires households to move away from their ideal variety. Much has been written recently about the increase in national market concentration observed over the last two decades, and the role that large national firms have played in driving this trend. The evidence for the rise in concentration is uncontroversial; the share of the largest firms and the Herfindahl-Hirschman Index, among other measures of concentration, have increased consistently in most sectors since 1990.1 A narrative has emerged whereby this increase in national concentration is perceived as the cause of lower product-market competition. This fall in competition is then viewed as the culprit of other apparent trends, such as rising markups and market power (Guti´errez and Philippon, 2017; De Loecker, Eeckhout and Unger, 2018; Hall, 2018), the increasing profits of large firms (Barkai, 2017), declining labor market dynamism and firm entry (Decker, Haltiwanger, Jarmin and Miranda, 2017), and declining wages and declining labor share (e.g. Autor, Dorn, Katz, Patterson, and Van Reenen, 2017)2 . Some studies have called into question the interpretation of these facts as evidence of increasing market power (see Hopenhayn, Neira and Singhania, 2019; Syverson, 2019), and the empirical robustness and validity of some of these trends has also been contested in recent work3 . However, the uncontroversial rise in national market concentration remains as the main empirical foundation of this central narrative. In this paper, we use the National Establishment Time Series (NETS) dataset to document four main facts regarding national and local product-market concentration in the U.S. economy between 1990 and 2014. Our first fact is that the observed positive trend in market concentration at the national level has been accompanied by a corresponding negative trend in average local market concentration. We measure concentration using the Herfindahl-Hirschman index (HHI), but our findings hold for a variety of statistics. We observe an increase in concentration at the national level overall across the vast majority of sectors and industries, but a fall in concentration when it is measured at the CBSA, County, or ZIP code levels.4 The 1  A 2016 report by the Council of Economic Advisers, for instance, finds that the national revenue share of the top 50 firms has increased across most NAICS sectors between 1997 and 2012. The report can be found at https://obamawhitehouse.archives.gov/sites/default/files/page/files/20160414 cea competition issue brief.pdf. Guti´errez and Philippon (2017) show that this increase in U.S.-wide market concentration is not uniform across all sectors and has been most pronounced in non-manufacturing sectors. Barkai (2017), and Autor, Dorn, Katz, Patterson, and Van Reenen (2017), find that the national sales share of top firms has also been rising since 1997 and, in fact, helps explain the decline in the labor share over the same period. 2 Other examples are Azar, Marinescu and Steinbaum (2017), Benmelech, Bergman and Kim (2018) and Qiu and Sojourner (2019). 3 While rising market concentration at the national level is relatively undisputed, the evidence regarding markups is more mixed. De Loecker, Eeckhout and Unger (2018) show evidence of rising markups since the 1980s among publicly traded firms. However, Traina (2018) points out that the evidence on markups depends crucially on the measurement of variable costs. When variable costs include marketing and management costs, as well as other indirect costs of production, markups have been relatively flat since the mid-1980s. Hall (2018) also finds essentially constant markups at the sectoral level using KLEMS productivity data. Similarly, Karabarbounis and Neiman (2018) find generally flat markups over time when also accounting for selling, general, and administrative expenses. Anderson, Rebelo, and Wong (2018) focus on the retail sector and find stable markups since 1979 using scanner data on the price of transactions and measuring marginal costs as replacement costs at the store level. Edmond, Midrigan and Xu (2018) show that when weighted by costs rather than sales, as implied by the microfoundations they lay out, aggreate markups have increased only modestly. 4 In the main text, we focus mostly on ZIP codes as our geographic definition of a local market. An online-only appendix  2  narrower the geographic definition, the faster the decline in local concentration. This is meaningful because the relevant definition of concentration from which to infer changes in competition is, in most sectors, local and not national. The second fact shows that local concentration is falling across SIC 8 industries that together account for 78% of employment and 72% of sales. Furthermore, conditioning on industries where national concentration is rising, industries where local concentration has declined account for the majority of employment overall (72% of employment and 66% of sales) across all major sectors. The presence of these diverging trends is always large but more pronounced in services, retail trade, and FIRE relative to wholesale trade and manufacturing. This ordering is natural given that transport costs are less relevant in the latter two sectors. Together, these first two facts underscore an unmistakable decline in local concentration on average that is pervasive across all sectors. How does one reconcile a positive trend in national concentration with a negative trend in local concentration? Our third fact shows that among SIC 8 industries that exhibit this pattern, top firms have accelerated these trends. That is, excluding the top firm in each industry (in terms of national sales in their SIC 8 industry in 2014), the national increase in concentration becomes naturally less pronounced. Perhaps more surprisingly, the decline in local concentration also becomes less pronounced. Put another way, large firms have materially contributed to the observed decline in local concentration.5 Among industries with diverging trends, large firms have become bigger but the associated geographic expansion of these firms, through the opening of more plants in new local markets, has lowered local concentration thus suggesting increased local competition. In the considerably smaller set of industries where we observe increases in both national and local concentration, top firms have also been responsible for both forms of concentration. Our fourth fact establishes that among industries with falling local concentration, the opening of a plant by a top firm is associated with a decline in local concentration at the time of the opening, and that this lower level of concentration persists for at least the next 7 years. This observation provides further evidence that in those industries, large enterprises do not enter and dominate the local market but instead lower its concentration, either by competing with the previous local monopolist or by simply adding one more establishment that grabs a proportional market share from other local establishments. In any case, the notion that entry by large firms eliminates local producers to the point of increasing concentration is certainly not supported in the vast majority of industries where most of U.S. employment resides. Consider the much-publicized case of Walmart. Most of Walmart’s establishments are in the discount department stores industry, an industry with declining local concentration. Consistent with fact four, when Walmart opens a store, the HHI falls by 0.15 in the associated ZIP code. In contrast, computing the HHI without taking into account the opening of a Walmart establishment, concentration remains constant. One can also consider the effect of Walmart on the number of firms in a market. When Walmart enters, the total number of establishments in the ZIP code increases, though by less than one-to-one (about 3/4). In other presents results with other geographic units. 5 This finding also holds when we exclude the top three firms in each industry instead of just the top firm.  3  words, Walmart generates some exit but the net result of opening a Walmart store is a greater number of competitors in the market for at least 7 years after entry.6 This case is paradigmatic, but there are many others across all major sectors. For example, the expansion of Cemex, the top firm by sales in 2014 in the ready-mixed concrete industry, led to a similar decline in local concentration and an expansion in the local number of establishments in the industry.7 Our findings challenge the view that product-market concentration is increasing in the U.S. They do so not by challenging the evidence that national concentration has increased – we actually provide additional evidence to that effect across many industries – but by observing that this national trend does not imply a positive local trend in concentration. In fact, we show that it implies the opposite in most industries, a declining trend in concentration. Ultimately, concentration matters because it can lead to less competition. Hence, measures of concentration have to be aligned with product markets as well as their geographic and industrial scope. In particular, for the majority industries, concentration is likely more relevant to firm pricing and other strategic behavior at a more local level. Our findings are also consistent with the mixed evidence found in recent literature regarding secular changes in markups across individual industries. If local competition matters, we should not see increases in markups or profits in the markets where local competition is increasing. The measurement of markups in local markets associated with particular industries depends on important assumptions and requires very detailed data. The NETS data does not allow us to calculate these local statistics, but there exists evidence of flat markups over time in specific industries with declining concentration (Anderson, Rebelo, and Wong, 2018), and in the aggregate (Traina, 2018). Finally, our results are also consistent with recent papers contending that labor market concentration is falling in the US economy (Hershbein, Macaluso and Yeh, 2018; Rinz, 2018; Berger, Herkenhoff and Mongey, 2019).8 The National Establishment Time Series (NETS) dataset covers the universe of U.S. firms and their plants.9 The dataset includes sales and employment numbers of all plants at different levels of geographic and industrial disaggregation down to the SIC 8 product code. Neumark, Zhang and Wall (2006), and Barnatchez, Crane and Decker (2017), provide thorough discussions of the advantages and disadvantages of this data source relative to U.S. Census data. The next section discusses many of these and the extent to which they are relevant for our findings. In particular, we show that for the industries covered in our study, the geographic distribution of employment in the data is very highly correlated with that in the U.S. Census 6 Jia (2008) studies competition by Walmart and other discount retail stores. She proposes a structural model of this competition and argues that the profits of previously available retailers decrease when ‘Walmart comes to town’. This is consistent with our view that Walmart lowers concentration by taking market share away from local competitors. Moreover, the exit of firms we observe is also consistent with those of Jia (2008) when measured at the county level. Holmes (2011) studies the expansion strategy of Walmart and, in particular, its geographic expansion strategy. Our findings are exactly consistent with this view of geographic expansion and provide related facts concerning its impact on local concentration. In contrast to these studies, our empirical findings extend to most U.S. industries in addition to the discount retail sector. 7 This industry was singled out in Syverson (2008) as an example of an industry with a local market. 8 Berger, Herkenhoff and Mongey (2019) also develop an equilibrium model of the U.S. labor market and find that, although there are large welfare gains from increasing competition, in their framework market power is unable to explain the decline in the labor share. 9 Throughout the paper we interchangeably use the terms plant and establishment. We also treat firm and enterprise as synonymous.  4  and, moreover, that this correlation is stable over time. Our findings, therefore, cannot be driven by changes in data coverage, accuracy, or the speed of its updates. The small discrepancies we find between NETS and standard Census data sources can evidently not explain the variety of consistent patterns revealed in the data. In addition, a critical feature of the NETS data integral to our analysis is that it allows researchers to explore and share the role of individual enterprises in shaping changes in industry concentration. This defining characteristic of NETS, therefore, permits explorations that would otherwise be infeasible with Census data. The facts we document are directly relevant to the design of antitrust policy and other policies that can prevent successful firms from growing at the national level. We document heterogenous trends across industries and, in some industries, concentration is clearly rising both at the national and local level. However, our results should provide pause for policy-makers that worry about increases in market power. On the whole, and in most industries, large firms are lowering local concentration and, therefore, most likely increasing product market competition. Carl Shapiro, a former Deputy Assistant Attorney General at the Antitrust Division of the Department of Justice, and member of the President’s Council of Economic Advisers under Barack Obama, makes a similar argument. Discussing evidence on the positive trend in national market concentration, he observes: “So, while these data do reflect the fact that large, national firms have captured an increasing share of overall revenue during the past 20 years in many of these 893 ‘industries,’ they do not, in and of themselves, indicate that the relevant local markets have become more concentrated.” In this paper, we provide the empirical evidence supporting the notion that, in the face of rising national concentration, local markets have indeed become on average significantly less concentrated.10 The rest of the paper is organized as follows. Section 2 describes our data, the way we use it, as well as our benchmark measure of national and local concentration. Section 3 presents our main four facts and describes their implications. Section 4 concludes. An online-only appendix presents a large variety of additional calculations using other concentration statistics, and provides additional detail regarding the data and the results in the main text.11  2  Data and Concentration Statistics  Our analysis uses data from the National Establishment Time Series (NETS), which is provided by Walls & Associates, and comprises annual observations on specific lines of business at unique locations over the period 1990 − 2014. In particular, NETS data allow us to observe sales and employment of between 8 and 19 million lines of business each year in our sample. Each line of business is assigned a data universal numbering system (DUNS) identifier that makes it possible to track its sales and employment over time at the SIC 10  De Loecker and Eeckhout (2018) also argue that in order to measure concentration in a way that is meaningful as an indicator of market power, this measurement has to be carried out for specific goods and local markets using the universe of firms. This is exactly what we do in this paper using the NETS data. 11 The online-only supplementary appendix to this paper is available at http://www.princeton.edu/~erossi/DTNLC_ Appendix.pdf.  5  8 level of Standard Industrial Classification (SIC) and at specific latitudes and longitudes. Industries can be mapped into broader SIC 2 classifications or divisions, and locations can be mapped into ZIP codes, counties, or Core-Based Statistical Areas (CBSAs). In addition, each line of business is also assigned a headquarters (HQ) number that gives the particular enterprise to which it reports. Thus, the NETS data encompass the universe of establishments operating in the U.S., and the enterprise to which each belongs, between 1990 and 2014. To better illustrate the nature of the NETS data, consider the case of Walmart as an example of an enterprise. It is headquartered in Bentonville, AR, and in 2014, it is associated with approximately 4700 establishments across all 50 states. Each of these 4700 establishments is assigned its own 8−digit primary SIC code, with 3718 establishments operating mainly as discount department stores (SIC 53119901), 603 establishments operating mainly as warehouse club stores (SIC 53999906), 241 establishments operating primarily as grocery stores (SIC 54110000), and the remaining establishments scattered mostly across various retail classifications. Because each establishment in the NETS data is assigned a unique DUNS identifier, it is possible to track when an establishment enters our sample (for those that enter after 1990) and, if applicable, when it exits. In addition, the DUNS identifier follows each establishment over time even if it is sold from one enterprise to another, or becomes included in a merger of enterprises, so that sales and employment of particular establishments may be tracked irrespective of corporate-level changes. Approximately a quarter of enterprises in the NETS data have only one employee. This feature of the data is typically not accounted for by alternative government sources of local employment as estimated by the County Business Patterns (CBP) or the Quarterly Census of Employment and Wages (QCEW).12 Since these establishments nevertheless report positive sales, we include them in our benchmark analysis. In an online-only supplement to this paper, we show that our results are robust to excluding enterprises with only 1, fewer than 5, or fewer than 10 employees. At the 2−digit SIC code, the data is classified in terms of 11 divisions, including Manufacturing, Services, Retail Trade, Wholesale Trade, and Finance, Insurance and Real Estate (FIRE), that together account for approximately 85 percent of sales and 80 percent of employment respectively in 2014. Because our analysis centers on the relationship between market concentration and the geographic expansion of enterprises, we exclude from our benchmark exercises industries that are intrinsically tied to specific locations because of weather or endowments of natural resources. These industries include Mining, Agriculture, Forestry, and Fishing, Construction, and Transportation and Public Utilities. We also exclude from our benchmark analysis any government establishment, including establishments belonging to enterprises whose headquarters are associated with a public administration SIC code, and establishments associated with education, non-profit endeavors, and central banking. 12  Many enterprises with one employee are non-employer enterprises, or in other words, have no paid employees. While employment at those enterprises is then at times the result of imputations, Barnatchez, Crane, and Decker (2017) show that taking out those imputations leaves measures of local employment that are generally highly correlated with those in the CBP across industries.  6  Throughout the analysis, we consider different levels of industrial and geographic disaggregation. The most basic level of disaggregation we consider is defined as an SIC 8-ZIP code pair. The NETS data cover 18, 000 SIC-8 industries and about 40, 000 ZIP codes. Because we omit particular industries whose operations have intrinsic ties to geographic endowments, our sample includes 15, 305 industries. In each year, we only use SIC 8-ZIP code pairs that have reported both positive sales and positive employment. This leaves us with around 6 million SIC 8-ZIP code pairs for each year on average that we aggregate in different ways below. Because we only consider in each year SIC 8-ZIP code pairs that have at least one establishment with positive sales and employment, the number of industry-geography pairs at a given level of aggregation (e.g. SIC 4-County) can vary from year to year. Below and in the online-only appendix, we show that our results are robust to exercises that only consider industry-geography pairs that have at least one establishment with positive sales and employment in every year. In the latter exercises, the number of industry-geography pairs is constant across time.  2.1  The NETS Data  One important advantage of the NETS data is that it covers every establishment in the U.S. at an exceptionally high level of disaggregation both by industry and geographic area. Unlike the comparable microdata available from the Census Bureau which produces the County Business Patterns (CBP) data, NETS does not require that a research proposal be approved describing how the data is to be used. It also does not require that an approved researcher travel to the location of a Federal Statistical Research Data Center (FSRDC) for secure access or compliance with the disclosure process to protect sensitive information. NETS data only require a subscription fee and can be easily accessed on any machine without undergoing a formal review process. A critical feature of the NETS data is that NETS allows researchers to examine and illustrate the impact of individual enterprises, such as Walmart in the discount department stores industry, or Cemex in the ready-mix concrete industry. This feature of NETS is integral to our analysis in that it permits us to explore and disclose the extent to which findings on concentration are driven by specific and large enterprises in individual industries. This defining characteristics of NETS, which we exploit in this paper, thus permits explorations that would otherwise be infeasible with Census data. Evidently, to the degree that NETS allows us flexibility not permitted with Census data, it is important to benchmark how the two datasets compare. Barnatchez, Crane, and Decker (2017) note that NETS include many non-employer establishments not covered by the CBP which tend to be very small establishments. When removing establishments with 1 employee or fewer than 5 employees, they find that overall trends in employment and establishments counts are closely aligned with the CBP. They suggest, therefore, verifying that findings using NETS data are robust to these sample restrictions. Barnatchez et al. (2017) also observe that differences between NETS and Census data are largely related to discrepancies in Agriculture, Mining, and Construction. In particular, in the latter two sectors, NETS appears to not have captured the full extent of recent employment changes. These include changes resulting from the boom-bust cycles associated with 7  the shale-oil and gas expansion in the early 2000s for mining, and the bursting of the housing bubble in 2007 for construction. As mentioned above, our analysis excludes these sectors in any case since their activity is intrinsically tied to local geographic characteristics. To get an idea of the differences between NETS and the CBP, Figure (1) illustrates standardized (i.e. set to 1 in 1990) aggregate employment in the NETS database and the CBP. Unlike Barnatchez et al. (2017), we remove from both datasets the set of industries described above. Consistent with their observations, in the NETS data, we plot standardized employment when including all enterprises as well as when excluding, in each year, enterprises with 1, fewer than 5, and fewer than 10 employees. As shown in the figure, the NETS data line up almost identically with the CBP data up to around 2002. Small differences arise after that year though all series appear to flatten out. Of note, almost throughout the sample, CBP employment lies in between employment excluding 1 and fewer than 5 employees in the NETS data. The difference in the sampling of small firms between NETS and the Census noted in Barnatchez et al. (2017), therefore, appears bounded by these two cases. In an online-only appendix, we show that all findings in this paper hold for all NETS cases shown in Figure (1). Moreover, as described below, all weighted average estimates in this study are weighted by employment so that, by construction, any differences in the sampling of small firms after 2002 is also given small weight. Observe also that higher frequency changes in Census data, such as the dip in employment following the Great Recession, are present in the various NETS cases shown in Figure (1). Evidently, differences in sampling between NETS and the Census manifest themselves mainly as a level effect after 2002. While differences between NETs and the CBP primarily result from the sampling of small firms, a question arises as to whether these differences might have grown over time. To explore this question, Figure (2a) illustrates the cross-sectional correlation between CBP and NETS total employment across counties in each year of our sample. The correlation is high at above 0.98 in every year irrespective of the definition we adopt for the NETS data (i.e. enterprises with 1, fewer than 5, and fewer than 10 employees). More importantly, to the extent that differences exist, for the set of industries we consider, there is no obvious time trends in those differences. The CBP data allows us to construct these correlations industry by industry only at the county level, where these are very close, but not at the ZIP code level. At the more geographically disaggregated ZIP code level, the CBP only reports aggregate ZIP code employment without any industry breakdown. Consequently, we are not able to remove particular industries such as Agriculture, Mining, or Construction when comparing ZIP code level data between NETS and the CBP. However, even with those industries included, Figure (2b) shows that the correlation between ZIP code employment in NETS versus the CBP remains above 0.85 in every year. As with the county level data, to the extent that this correlation is not exactly 1 and that differences exist between the two datasets, there is no trend in the way that ZIP code employment differs between NETS and the CBP. In summary, our findings are consistent with those of Barnatchez et al. (2017) who find differences between NETS and Census data related to the sampling of small firms. As the authors suggest, these differences appear to be bounded by exercises that remove small firms. Specifically, we find that standardized  8  Standardized Employment .75 1 1.25  1.5  Figure 1: Standardized National Employment in CBP and NETS Data  .5  Including All NETS Enterprises Excluding NETS Enterprises with 1 Employee Excluding NETS Enterprises with Fewer than 5 Employees Excluding NETS Enterprises with Fewer than 10 Employees CBP 1990  1995  2000  Year  2005  2010  2015  Notes: The plot compares standardized aggregate employment for CBP and NETS for the 5 major sectors used in this study: FIRE, Manufacturing, Retail, Services, and Wholesale trade. Each different line represents a different selection of enterprises to be included for the NETS sample. CBP employment is bounded by cases that remove firms with 1 and fewer than 5 employees in the NETS data. Moreover, our analysis removes industries identified by Barnatchez et al. (2017) as those least closely matching employment in the CBP, though we remove them for different reasons as explained above. Finally, while there appears to be an increase in the difference between CBP and NETS employment starting in 2002, this discrepancy emerges mainly as one time change with no visible trends after 2002. All of these observations, therefore, suggest that any findings related to aggregate or industry-wide estimates that we discuss below cannot be driven by the scope or coverage of the two datasets.  2.2  Measuring Concentration  Establishments in our dataset are indexed by industry, i, location, `, and year, t. Industries are defined by an SIC 8 code. Locations are defined by a latitude-longitude pair. We denote collections of industries into broader classifications (for example, SIC 2 or divisions) by d. We denote collections of locations into broader geographies (ZIP codes, CBSAs, Counties, States, or the whole U.S.) by g. When defining locations at the CBSA level, counties that are not within CBSAs are not represented, which amounts to between 5 and 10 percent of establishments in any given year.  9  1 Correlation Coefficient .5 .75 .25  .25  Correlation Coefficient .5 .75  1  Figure 2: County and ZIP-Level Correlations Between NETS and CBP Employment  1990  1995  2000  Year  2005  2010  Including All NETS Enterprises Excluding NETS Enterprises with 1 Employee Excluding NETS Enterprises with Fewer than 5 Employees Excluding NETS Enterprises with Fewer than 10 Employees  0  0  Including All NETS Enterprises Excluding NETS Enterprises with 1 Employee Excluding NETS Enterprises with Fewer than 5 Employees Excluding NETS Enterprises with Fewer than 10 Employees 2015  1995  2000  2005 Year  2010  2015  Notes: Plot (a) produces a time-series of the cross-sectional correlation of county employment between CBP and NETS for the 5 major sectors used in this study: FIRE, Manufacturing, Retail, Services, and Wholesale trade. Plot (b) produces a time-series of the cross-sectional correlation of ZIP code employment between CBP and NETS for the full economy (CBP does not provide an industry classification at the zip-code level). In both plots, each different line represents a different selection of enterprises to be included for the NETS sample. I,G I,G denote the nominal sales of enterprise e in industry i at location ` in year t, and Se,i,g,t = Let Se,i,`,t  P  `∈g  I,G its sales in the broader geography g (i.e. the sum of all its establishments’ sales across all latitudeSe,i,`,t  longitude pairs ` in geography g). The index I refers to the industrial level of aggregation (i.e. by SIC 2, SIC 4, SIC 6, or SIC 8). The index G indicates the geographic level of aggregation (i.e. by ZIP code level, CBSA level, County level, or the whole U.S.) that we use to define a location `. We then denote by sI,G e,i,g,t this enterprise’s share of all sales in industry i located in geography g at date t for the levels of aggregation I and G. We adopt as our benchmark measure of market concentration the Herfindahl-Hirschman Index (HHI), I,G Ci,g,t =  X  sI,G e,i,g,t  2  ,  e I,G I,G I,G where Ci,g,t ∈ [1/Ni,g,t , 1] is the sales concentration, and Ni,g,t the number of enterprises in industry i and  geography g at time t. In the online-only supplement to this paper, we also consider alternative measures of concentration, such as the sales share of the top firm, or the adjusted Herfindahl, and show that all of our findings are robust to these other measures. In particular, to the degree that the number of small firms in an industry differs between NETS and CBP data, and that this difference were to have an increasing trend that materially reduces the lower bound of the Herfindahl index, the range of the adjusted Herfindahl index  10  no longer depends on the number of firms.13  3  National and Local Market Concentration: The Facts  We organize the discussion of our findings into four main facts. The first two facts document the diverging trends in national and local concentration and their importance across sectors and geographic definitions of a ‘local’ market. The third and fourth facts document the role that large firms have played in these trends. As a form of corollary to the last fact, we also present evidence specific to Walmart, a firm that has featured prominently in the debate on the evolution of market concentration.  3.1  Fact 1: Diverging Trends on Average  Fact 1 is summarized in Figures (3) and (4). Figure (3) shows a weighted average of the change in concentration, ∆CtI,G , across all industry-geography pairs (i, g) for different definitions of geography G, namely ZIP code, County, CBSA, and the whole U.S., ∆CtI,G =  X  I,G I,G wi,g,t ∆Ci,g,t ,  (1)  i,g I,G where the weights wi,g,t are given by the employment shares of industry-geography (i, g) in aggregate emI,G ployment in year t, and ∆Ci,g,t denotes the change in market concentration between year t and the first  year for which we observe sales in the location-industry pair (i, g). As mentioned in Section 2, all findings presented in this paper are robust to excluding enterprises with 1, fewer than 5, and fewer than 10 employees. In part, this is because, by construction, our employment-weighted measures of concentration already assign small weights to industry-geography pairs that contain mostly small enterprises.14 As indicated in the 2016 CEA report, Barkai (2017), Guti´errez and Philippon (2017), and others, market concentration at the national level has been steadily increasing since 1990. However, the exact opposite is true for less aggregated measures of concentration. Figure (3) shows that the more geographically disaggregated the measure of concentration, the more pronounced its downward trend over the last two and a half decades. Figure (4) shows a weighted average of the change in concentration across all industry-geography pairs (i, g) within a particular division, d, namely Manufacturing, Services, Retail Trade, Wholesale Trade, and 13 14  I,G The adjusted Herfindahl is given by Cei,g,t =  I,G  I,G  Ci,g,t −1/Ni,g,t I,G  1−1/Ni,g,t  I,G I,G I,G ∈ [0, 1] when Ni,g,t > 1 and Cei,g,t = 1 when Ni,g,t = 1.  Given differences in the number of firms and other industry characteristics, we study changes in HHI instead of the level of the HHI, so that we can compare trends in concentration across industries. This is why we aggregate changes in the HHI instead of aggregating levels of HHI. Using sales shares instead of employment shares as weights yields similar results.  11  FIRE, for geographies defined by ZIP code and the whole U.S., I,G ∆Ct,d =  X  I,G I,G wi,g,t,d ∆Ci,g,t .  i∈d,g  Figure (4) shows that while increasing market concentration at the national level holds broadly across all divisions, it is equally the case that concentration has steadily fallen at the ZIP code level in these divisions. Observe, in particular, that market concentration in the Retail Trade division has been increasing nationally more than in any other division. However, Retail Trade is also among the divisions that show the steepest decline in concentration at the ZIP code level. This fact is especially striking given that physical retail establishments in our dataset are likely to have very local markets.  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0  .05  Figure 3: Diverging economy-wide national and local concentration trends  -.25  National CBSA County ZIP 1990  3.1.1  1995  2000  Year  2005  2010  2014  Concentration by Industrial Classification and Employment  Figure (5) depicts the divergence between national and local concentration at the ZIP code level for different degrees of industrial aggregation I. This growing divergence between national and local concentration is most pronounced at the SIC 8 level but clearly present at lower levels of industrial aggregation as well, including the most coarse SIC 2 classification. Thus, we explore below in more detail how these diverging trends between national and local concentration are related to industrial classification. Figure (6) repeats the exercise in Figure (5) but focusing on employment rather than sales. As the 12  National  ZIP Manufacturing Wholesale Trade Retail Trade FIRE Services  -.25  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0  .05  Figure 4: Diverging division-level national and local concentration trends  1990  1995  2000  Year  2005  2010  2014  figure shows, using employment rather than sales is immaterial for the growing divergence between national and local concentration.15 In the online-only appendix, we show that all of our other findings regarding diverging trends between national and local concentration also hold for employment as well as sales. In summary, Figures (3) through (6) indicate a growing divergence in national and local concentration that holds for broad levels of industrial and geographic definitions. In the online-only appendix, we carry out and present a large number of exercises that highlight the robustness of our findings.16 3.1.2  Concentration and Sample Selection  Before proceeding with the analysis and an exploration of the roots underlying our basic Fact 1, we discuss an important aspect of this fact related to sample selection. In particular, because we omit in each year industry-geography pairs with no establishments, the resulting unbalanced panel can create situations where an industry-geography pair with a single establishment is dropped from one year to the next. Since the omitted observation is one with a single establishment, and thus associated with high concentration, local concentration decreases simply as a result of losing the observation. Conversely, of course, entry 15  In the NETS data, establishment sales numbers are more frequently imputed than employment numbers. However, given that the CBP data does not report sales figures, the similarities in national and local concentration trends for sales and employment in NETS are consistent with the notion that, for a given enterprise, increasing sales by opening new establishments in local markets requires hiring labor in those markets. 16 We also consider county and CBSA geographies, as well as the SIC 4 industrial classification code, that highlight respectively the importance of local markets and well defined industries.  13  National  ZIP SIC 2 SIC 4 SIC 6 SIC 8  -.15  Average Change in HHI from First Year -.1 -.05 0  .05  Figure 5: Diverging economy-wide trends in sales concentration  1990  1995  2000  Year  2005  2010  2015  National  ZIP SIC 2 SIC 4 SIC 6 SIC 8  -.15  Average Change in HHI from First Year -.1 -.05 0  .05  Figure 6: Diverging economy-wide trends in employment concentration  1990  1995  2000  Year  14  2005  2010  2015  has the opposite effect whereby new establishments in markets without a previous incumbent raise local concentration.17 Figures (7) and (8) repeat the exercises in Figures (3) and (4) but only considering industry-geography pairs where at least one establishment is present in every year. The resulting panel, therefore, is balanced. Though slightly less pronounced, the divergence between national and local concentration trends remains unequivocal. Furthermore, it is still the case that this divergence becomes more pronounced as one moves towards more disaggregated definitions of local markets. It is also still the case that the divergence in concentration trends is particularly evident in service industries, such as Retail and FIRE, which make up the bulk of the U.S. economy.  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0  .05  Figure 7: Diverging economy-wide national and local concentration trends with a balanced panel  -.25  National CBSA County ZIP 1990  1995  2000  Year  2005  2010  2014  In the online-only appendix, we show that these balanced-panel findings also hold for other measures of concentration such as the share of the largest firm or the adjusted Herfindahl. It is worth noting that, for more disaggregated definitions of local markets, measures of concentration based on the sales share of the largest X firms become less informative as X increases. The reason lies in strong selection issues. Consider for instance a concentration measure based on the sales share of the top 4 firms. In that case, measured concentration will necessarily be unchanged in all industry-geography pairs that have 4 or fewer firms throughout the sample. However, at the SIC 8-ZIP code level, 90 percent of observations turn out to have 3 firms or fewer and 93 percent of observations have 4 firms or fewer. This case is especially misleading 17  See, for example, the argument and evidence presented in Ganapati (2018).  15  National  ZIP Manufacturing Wholesale Trade Retail Trade FIRE Services  -.25  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0  .05  Figure 8: Diverging division-level national and local concentration trends with a balanced panel  1990  1995  2000  Year  2005  2010  2014  if those pairs are increasingly moving from having 1 firm to 3 firms and concentration is actually falling. For example, in the discount department store industry (which includes Walmart), of the ZIP codes that had 1 firm in 1990 and at least 1 firm in 2014, 56 percent had at least 2 firms competing in 2014 but only 5 percent of those ZIP codes had more than 4 firms. It emerges, therefore, that one has to be cautious when measuring concentration at local levels. Since industry-geography pairs with 4 or fewer firms represent the large majority of local markets, any study that attempts to measure market concentration in disaggregated sectors and highly disaggregated geographies using the share of the largest 4 firms faces strong measurement problems. Such studies (e.g. Ganapati 2018) do not challenge the results in our paper or their interpretation.  3.2  Fact 2: Pervasive Diverging Trends  Fact 2 is presented in Figure (9). Within each SIC 2 classification, the figure gives a breakdown of employment in industries with different market concentration trends. In particular, for a given SIC 2 classification, the height of each bar gives the percentage of employment in all industries within that classification that have rising market concentration at the national level between 1990 and 2014. For each SIC 8 industry P I,G I,G I,G i within an SIC 2 classification, we compute in each year ∆Ci,t = g wi,g,t ∆Ci,g,t , where both g and G I,G denote the whole U.S., and regress ∆Ci,t on t. The height of the bar then represents the percent of labor,  within that SIC 2 and across all years, employed in all SIC 8 industries with positive national concentration  16  time trends. Thus, the major part of U.S. employment resides in industries with rising national concentration across all SIC 2 classifications. Within a bar associated with a given SIC 2 classification in Figure (9), the colors red, blue, and black represent, respectively, the percent of employment that resides in industries with declining, rising, and flat market concentration at the ZIP code level.18 Figure (9) shows the pervasiveness of SIC 8 industries with diverging trends.19 That is, a substantive share of employment resides in industries with rising market concentration at the national level and declining market concentration at the ZIP code level. It also shows the heterogeneity in this share across SIC 2 divisions. For example, in SIC 2 53, which includes General Merchandise Stores, virtually all employment resides in SIC 8 industries with diverging trends (96.38%). In contrast, in SIC 2 21, which includes Tobacco Products, none of the SIC 8 industries exhibit a positive national trend and a negative local trend. Diverging trends are more pronounced in Retail, FIRE and Services, than in Wholesale Trade and Manufacturing, though still very much present in the latter two divisions. The proportion of aggregate U.S. employment located in all SIC 8 industries with increasing national market concentration and decreasing ZIP code level market concentration is 43 percent. Thus, given that some industries have also had declining concentration at both the national and ZIP code level, 78 percent (or over 3/4) of U.S. employment resides in industries with declining local market concentration.20  3.3  Fact 3: The Role of Top Firms  Fact 3 explores the contribution that top firms –in terms of sales share– have made to the diverging trends in each SIC 8 industry. Figures (10) and (133) focus on just those industries whose market concentration has increased at the national level since 1990, represented by the height of the bars in Figure (9). Those industries account for roughly half of all industries in our sample, 61% of aggregate U.S. employment, and 67% of aggregate sales. Within that set of SIC 8 industries, Figure (10) focuses on those that exhibit negative local concentration trends. These industries account for 72% of total employment in industries with positive national trend (66% of sales). The figure presents in solid orange and solid red, respectively, the national HHI and the local ZIP code level HHI among these industries. Given our industry selection, the national concentration (orange) line is increasing by construction and the local concentration (red) line is decreasing by construction. The dashed orange and dashed red lines in that figure depict the same objects but excluding the top enterprise in each SIC 8 industry as measured by sales in 2014.21 We consider only industry-geography pairs (i, g) for which i’s top enterprise has at least one establishment present in g in at least one year. Furthermore, because we are interested in isolating the effect of the top enterprise on market concentration, among those remaining P I,G I,G I,G Specifically, in the calculation of ∆Ci,t = g wi,g,t ∆Ci,g,t , both g and G now represent a ZIP code. 19 We reserve the term of ‘diverging trends’ for a case of positive national trend and a negative local trend. The case of a negative national trend and a positive local trend is also possible, though much less common in virtually all industries. 20 The share of national sales in sectors with decreasing local market concentration is 72%. 21 We show in the online-only appendix that we obtain similar results when we exclude the top 3 firms rather than only the top firm. 18  17  0  Percent of Employment in SIC 2 20 40 60 80  100  Figure 9: Pervasive diverging trends across 2-digit sectors  53 55 56 54 57 59 52 58  60 65 61 67 63 62 64  83 86 72 88 78 75 70 76 80 79 87 73 81 89 84  Retail Trade  FIRE  Services  50 51  26 30 38 36 32 33 37 24 34 27 35 28 20 22 31 39 23 25 29 21  Wholesale Trade  Manufacturing  Percent of Employment in SIC 8s with Increasing National and ZIP Trends Percent of Employment in SIC 8s with Increasing National and Decreasing ZIP Trends Percent of Employment in SIC 8s with Increasing National and Zero ZIP Trends  industry-geography pairs, we then only include observations (i, g, t) where at least one establishment remains after dropping the top enterprise in i and its associated establishments.22 Figure (10) shows that among SIC 8 industries with diverging trends, excluding the top firm results in a national concentration trend that is less pronounced. The fact that large firms have contributed to the national increase in concentration is as expected. More surprising is the observation that the top firms have also contributed to the decline in local concentration. Figure (10) shows that when we exclude the top firm, the negative trend in ZIP-code-level concentration is less pronounced. Hence, the top firm (and more generally the largest firms) in an industry are responsible (though not entirely) for the diverging trends. Figure (133) is constructed exactly as Figure (10) but uses the SIC 8 industries with increasing national trends which are not depicted in Figure (10). In other words, it uses the SIC 8 industries with positive national and local trends. The figure shows that for this set of industries, excluding the top firm lowers both the national and the local trend in concentration. Over the last ten years or so, it also shows that excluding the top firm reduces the trend in national concentration significantly more than that in local concentration. How can the growth of large firms contribute to the divergence in these trends? To a large extent, top enterprises expand by adding new establishments in new locations. The new establishments tend to decrease 22 We also exclude industry-geography pairs whose first year of observed sales results from only one establishment belonging to the top enterprise, since the change in market concentration cannot be computed in that case.  18  -.2  Average Change in HHI from First Year -.15 -.1 -.05 0 .05  .1  Figure 10: The role of top enterprises in national and local concentration trends in diverging industries  Including Top Enterprise Excluding Top Enterprise ZIP Level National Level 1990  1995  2000  Year  2005  2010  2014  -.1  Average Change in HHI from First Year -.05 0 .05 .1  .15  Figure 11: The role of top enterprises in national and local concentration trends in concentrating industries  Including Top Enterprise Excluding Top Enterprise ZIP Level National Level 1990  1995  2000  Year  19  2005  2010  2014  local concentration as they compete with existing establishments in the area, even as the top firm acquires a larger national market share, increasing national concentration. Next, we explore the impact of local entry by a top firm. 3.3.1  Comparing Industries’ Largest Firm to their Runner-Ups  To the extent that lower concentration is associated with more competitive markets, the findings in this paper are suggestive of local markets becoming more competitive despite concentration rising at the national level. Our analysis is indeed consistent with the notion that in many industries, the top enterprise expands into new markets by opening plants that compete with already established local monopolists. The case for increasing competition, however, is less clear if the observed fall in concentration is the result of a few firms entering several markets. As discussed in Bernheim and Whinston (1990), and Bond and Syropoulos (2008), when firms compete in multiple markets simultaneously, the potential for collusion can grow since these firms’ ability to punish any deviation can be enhanced by their multiple ‘contacts’ across markets. Hence, increasing national concentration resulting from the increasingly large positions of two or three enterprises in an industry can result in declines in local concentration, more local contacts between competitors, and a rise in the ability to collude and thus effective market power of the largest firms. If, in contrast, increasing national concentration results from the gradual expansion of a single top firm competing with local firms, a decline in local concentration will be associated with reductions in market power in those local markets. To gain insight into the role of the largest enterprise in a given industry relative to that of the largest 2 or 3 enterprises, Figure (12) repeats the exercise in Figure (10) but excluding the second and third largest enterprise instead of the top enterprise. For industries where national and local concentration trends diverge, excluding the second and third largest enterprises results in an increase in concentration at the local level as in Figure (10). Thus, as with the top enterprise, when the second and third largest enterprises enter new geographical markets, local concentration falls. However, unlike Figure (10), excluding the second and third largest enterprises increases national concentration. Put another way, unlike the top enterprise, the second and third largest enterprises contribute to reducing concentration at the national level. This finding is inconsistent with the view that the largest 2 or 3 enterprises are responsible for a simultaneous rise in national concentration and decline in local concentration as they expand in new untested markets. Instead, Figure (10) shows that entry of the the second and third largest enterprises leads to overall declines in concentration as with any other less dominant firm. Analogous to Figure (12), Figure (13) repeats the exercise in Figure (133) for cases where both national and local concentration have been rising but excluding the second and third largest enterprises instead of the top enterprise. Comparisons with Figure (133) make it even more apparent that the second and third largest firms have an impact that on average differs from that of the most dominant enterprise across industries. In Figure (13), we see that the second and third largest enterprises contribute to lowering concentration both at the national and local level whereas the most dominant enterprise in Figure (133) contributes to higher concentration nationally. 20  -.25  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0 .05  .1  Figure 12: The role of the second and third largest enterprises in diverging industries  Incl. 2nd & 3rd Ranked Enterprises Excl. 2nd & 3rd Ranked Enterprises ZIP Level National Level 1990  1995  2000  Year  2005  2010  2014  -.1  Average Change in HHI from First Year -.05 0 .05 .1  .15  Figure 13: The role of the second and third largest enterprises in concentrating industries  Incl. 2nd & 3rd Ranked Enterprises Excl. 2nd & 3rd Ranked Enterprises ZIP Level National Level 1990  1995  2000  Year  21  2005  2010  2014  0  2.5  Percent 5 7.5  10  12.5  15  Figure 14: Expansion of Top Enterprises into ZIP Codes  Percent Containing Largest Enterprise Percent Containing Largest and Second and/or Third Largest Enterprises 1990  1995  2000  Year  2005  2010  2015  Figures (14) helps further contrast the way in which, across industries, the dominant enterprises have expanded geographically relative to the next two largest enterprises. Specifically, the solid line in Figure (14) shows the proportion of SIC 8-ZIP code pairs, weighted by employment associated with that pair, where the dominant enterprise has at least one establishment. This proportion steadily increased from 5.4 percent in 1990 to 15.5 percent in 2014. In other words, on average across industries, the largest enterprise has unambiguously and steadily expanded into new local markets over the last 25 years. The dashed line in Figure (14) depicts the proportion of SIC 8-ZIP code pairs where not only the largest enterprise but also the second and third largest enterprises have at least one establishment. While the dashed line has gradually increased over the last 25 years, it has done so at a considerably slower rate than the solid line. Put differently, the difference between the two lines represents the proportion of SIC 8-ZIP code pairs in the U.S. where the top enterprise is competing with smaller firms rather than its next two largest competitors, and this difference has itself gotten markedly larger over the last three decades.  3.4  Fact 4: When a Top Firm Comes to Town  To further illustrate the impact of an industry’s top enterprise on market concentration at the local level, Figures (15) and (16) present an event study describing the effect of local entry by an establishment associated with a top firm (defined by 2014 sales in an SIC 8 industry as in Fact 3) in a ZIP code. Specifically, Figures (15) and (16) examine the effect of a top firm opening a new establishment in a ZIP code on local  22  market concentration. The calculations here mimic those in Figures (133) and (10). In Figures (15) and (16), the x−axis plots a 10−year window surrounding a top firm establishment opening in a given ZIP code, with 0 denoting the opening year. To better highlight the net effect of entry on concentration, we normalize the change in concentration to zero in the year prior to the establishment opening. Figure (15) depicts the event study for all SIC 8 industries with increasing market concentration at the national level and decreasing local market concentration, that is, SIC 8 industries with diverging trends. Figure (16) illustrates findings for the remaining SIC 8 industries with increasing national concentration: those where both national and local trends are positive over our sample period. The solid lines in both figures present the evolution of the HHI index when the entering establishment is included; the dashed lines illustrate the same object when excluding the opening establishment owned by the top enterprise within each industry. Among industries with diverging trends, the opening of an establishment in a ZIP code is associated with a fall in market concentration. Moreover, this fall persists at about the same size for at least 7 years after the event. In contrast, among industries with increasing local market concentration, the opening of an establishment leads to a temporary decrease in market concentration but one that reverses quickly. After 4 to 5 years, concentration is higher than it would have been absent the opening. Hence, in the former case, the establishment owned by the top enterprise does not become dominant, while in the latter case it eventually dwarfs the establishments of other firms. The data suggest that on the whole, the case where the top firm does not become dominant at the local level is markedly more relevant.23 The dashed lines in both Figures (15) and (16) suggest that when all shares are re-calculated excluding sales of the opening establishment belonging to the top enterprise in each industry, market concentration does not exhibit a significant trend over the entire 10−year window. Thus, the dashed-lines lend credibility to a central assumption underlying the event study, namely that entry by a top enterprise in a local market is the main event affecting concentration in each market. 3.4.1  The Case of Walmart  The event study presented in Fact 4 averages the effect of entry by a top enterprise across many markets. It is informative, therefore, to further delve into the data within a particular sector. In the last couple of decades, one of the most widely studied cases of an expanding firm has been the case of Walmart.24 Hence, here we repeat the calculations underlying Fact 4 but for the particular case of Walmart and the SIC 8 industries with which it is associated. The solid line depicted in Figure (17) represents a weighted average 23 Neumark, Zhang and Wall (2006) and Barnatchez, Crane and Decker (2017) argue that NETS dataset might at times be slow in reporting the entry and exit of small firms. Given their findings, one might question the extent to which our results are driven by the exit of small firms not being reported accurately. However, the fact that the fall in the HHI persists for up to 7 years diminishes this potential concern. Another potential concern is that ZIP codes are too narrow a geographic definition of a market. In the online-only appendix, we show that the fall in the local HHI as a result of local entry by a top enterprise in industries with diverging trends holds and persists when we using counties rather than ZIP codes. 24 See, for example: Basker (2007), Jia (2008), Ailawadi, Zhang, Krishna, Krueger (2010), Zhu, Singh, Manuszak (2009), Holmes (2011).  23  Average Change in HHI from First Year -.05 -.03 -.01 .01  .03  Figure 15: Effect on concentration when a top enterprise enters a local market in diverging industries  -.07  Including Top Enterprise Excluding Top Enterprise  -3  -2  -1 0 1 2 3 4 5 Years Since SIC 8's Top Enterprise Opening in ZIP Code  6  7  Average Change in HHI from First Year -.05 -.03 -.01 .01  .03  Figure 16: Effect on concentration when a top enterprise enters a local market in concentrating industries  -.07  Including Top Enterprise Excluding Top Enterprise -3  -2  -1 0 1 2 3 4 5 Years Since SIC 8's Top Enterprise Opening in ZIP Code  24  6  7  of concentration within Walmart’s primary industry (discount department stores) across all ZIP codes. The dashed line represents the same object but excluding the opening establishment owned by Walmart (i.e. all shares are re-calculated excluding Walmart’s sales from the new establishment). Our findings for Walmart are qualitatively similar to those in Fact 4 for industries with diverging trends (as is the case for Walmart’s industries). Absent a Walmart opening, there is no trend in concentration, but there is a significant fall in the HHI of a ZIP code in which Walmart opens a new establishment. This lower level of concentration remains about constant for at least 7 years. One advantage of considering a particular firm and its industries is that we can also show, and easily interpret, the effect of entry on the number of establishments in the local market. To do so, Figure (18) illustrates the effect of a Walmart establishment opening in a given ZIP code on the number of establishments in that ZIP code. The solid line in the figure indicates that, when averaged across all ZIP codes (weighted by geography-SIC 8 employment, as in all other figures), the opening of a Walmart establishment is associated with an increase in the number of local establishments. This increase is somewhat less than one-for-one (roughly 0.75) which suggests that the entry of Walmart is associated with some establishment exits across ZIP codes. Consistent with this observation, the dashed line indicates that when the newly established Walmart is excluded from the calculation, the number of establishments falls somewhat across ZIP codes.25  Average Change in HHI from First Year -.1 -.05 0  .05  Figure 17: Effect on concentration when Walmart enters a local market  -.15  With Walmart Entering Without Walmart Entering  -3  -2  -1  0 1 2 3 4 Years Since Walmart Opening in ZIP  25  5  6  7  Consistent with the findings in Jia (2008) and Basker (2007), carrying out these calculations at the county level reveals a more pronounced effect of Walmart’s entry on firm exit. Nevertheless, the decline in the HHI is still large on impact and still negative after 7 years.  25  Average Number of Establishments in ZIP 2.5 3  3.5  Figure 18: Effect on number of establishments when Walmart enters a local market  2  With Walmart Entering Without Walmart Entering  -3  3.4.2  -2  -1  0 1 2 3 4 Years Since Walmart Opening in ZIP  5  6  7  The Case of Cemex  Figure (9) suggests that a very high share of employment in Retail Trade resides in industries with diverging national and local trends, while this phenomenon is much less prevalent in Manufacturing and Wholesale Trade. However, the sector-level of aggregation presented in Figure (9) obscures considerable heterogeneity within industries in a given sector. It is still the case that many Manufacturing industries have diverging trends and see declining local concentration following the arrival of their largest enterprise in a ZIP code. To use one example, Figures (19) and (20) highlight the SIC 8 code 32730000, Ready-Mixed Concrete, whose top enterprise by sales in 2014 is Cemex, a building materials company. Figure (19) shows that the arrival of Cemex into a ZIP code reduces concentration in its industry by about 0.1. Although this effect dissipates after 7 years, concentration measured excluding Cemex remains higher than it otherwise would when included so that this company is still contributing to lower local concentration. Figure (20) shows that, as with the case of Walmart, while some existing establishments do exit when Cemex opens a plant, the overall number of establishments in the ZIP-industry pair rises on average. Although Syverson (2008) documents increasing national concentration within this industry, consistent with our findings, Syverson (2004) and Syverson (2008) also argue that high transport costs make local measures of concentration more relevant. In this paper, we show that these more local measures exhibit a downward trend.  26  Average Change in HHI from First Year -.1 -.05 0 .05  .1  Figure 19: Effect on concentration when Cemex enters a local market  -.15  With Cemex Entering Without Cemex Entering -3  -2  -1  0 1 2 3 4 Years Since Cemex Opening in ZIP  5  6  7  Average Number of Establishments in ZIP 2 2.5 3  3.5  Figure 20: Effect on number of establishments when Cemex enters a local market  1.5  With Cemex Entering Without Cemex Entering  -3  -2  -1  0 1 2 3 4 Years Since Cemex Opening in ZIP  27  5  6  7  4  Conclusions  We have shown, by way of four main facts, that the increase in market concentration observed at the national level over the last 25 years is being shaped by enterprises expanding into new local markets. This expansion into local markets is accompanied by a fall in local concentration as firms open establishments in new locations. These observations are suggestive of more, rather than less, competitive markets. The findings in this paper potentially help reconcile the observation of increasing concentration at the national level and the more mixed evidence on increasing markups and profits. Almost no theory of product market competition associates decreasing concentration with either increasing markups or increasing profits. One exception resides in theories of multi-market collusion where a few firms competing in many markets can have enhanced opportunities to collude. Although this form of collusion could be important in specific industries, we show that an expanding top firm competing with local producers is a much more common occurrence. Thus, our facts indicate that the rising trend in national concentration is not, in and of itself, necessarily a concern for anti-trust policy. By decreasing local concentration, the growth of top firms has likely increased local competition in many industries and, therefore, helped improve the quality and reduce the prices of much of what we buy.  References [1] Ailawadi, Kusum L., Aradhna Krishna, Michael W. Kruger, and Jie Zhang. 2010. “When Wal-Mart Enters: How Incumbent Retailers React and How This Affects Their Sales Outcomes.” Journal of Marketing Research. 47 (4): 577-93. [2] Anderson, Eric, Sergio Rebelo, and Arlene Wong. 2018. “Markups Across Space and Time.” NBER Working Paper 24434. [3] Autor, David, David Dorn, Lawrence F. Katz, Christina Patterson, and John Van Reenen. 2017. “The Fall of the Labor Share and the Rise of Superstar Firms.” NBER Working Paper 23396. [4] Azar, Jose, Ioana Marinescu, and Marshall Steinbaum. 2019. “Labor Market Concentration.” NBER Working Paper 24147. [5] Barkai, Simcha. 2017. “Declining Labor and Capital Shares.” Working paper. [6] Barnatchez, Keith, Leland D. Crane, and Ryan A. Decker. 2017. “An Assessment of the National Establishment Time Series (NETS) Database.” Finance and Economics Discussion Series 2017-110. [7] Basker, Emek. 2007. “The Causes and Consequences of Wal-Mart’s Growth.” Journal of Economic Perspectives 21 (3): 177-98.  28  [8] Benmelech, Efraim, Nittai Bergman, and Hyunseob Kim. 2018. “Strong Employers and Weak Employees: How Does Employer Concentration Affect Wages?” NBER Working Paper 24307. [9] Berger, David, Kyle Herkenhoff, and Simon Mongey. 2019. “Labor Market Power.” Working Paper. [10] Bernheim, B. Douglas, and Michael D. Whinston. 1990. “Multimarket Contact and Elusive Behavior.” The RAND Journal of Economics. 21 (1): 1-26. [11] Bond, Eric W., and Constantinos Syropoulos. 2008. “Trade Costs and Multimarket Collusion.” The RAND Journal of Economics. 39 (4): 1080-1104. [12] Choi, Taelim, John C. Robertson, and Anil Rupasingha. 2013. “High-Growth Firms in Georgia.” Federal Reserve Bank of Atlanta Working Paper 2013-20. [13] Decker, Ryan A., John Haltiwanger, Ron S. Jarmin, and Javier Miranda. 2017. “Declining Dynamism: Allocative Efficiency, and the Productivity Slowdown.” American Economic Review: Papers & Proceedings. 107 (5): 322-6. [14] De Loecker, Jan, and Jan Eeckhout. 2018. “Global Market Power.” NBER Working Paper 24768. [15] De Loecker, Jan, Jan Eeckhout, and Gabriel Unger. 2018. “The Rise of Market Power and the Macroeconomic Implications.” Working Paper. [16] Edmond, Chris, Virgiliu Midrigan, and Daniel Yi Xu. 2018. “How Costly are Markups?” NBER Working Paper 24800. [17] Ganapati, Sharat. 2018. “Growing Oligopolies, Prices, Output, and Productivity.” Working paper. [18] Guti´ errez, Germ´ an, and Thomas Philippon. 2017. “Declining Competition and Investment in the U.S.” NBER Working Paper 23583. [19] Hall, Robert E. 2018. “New Evidence on the Markup of Prices Over Marginal Costs and the Role of Mega-Firms in the US Economy.” NBER Working Paper 24574. [20] Hershbein, Brad, Claudia Macaluso, and Chen Yeh. 2018. “Concentration in U.S. Local Labor Markets: Evidence from Vacancy and Employment Data.” Working paper. [21] Holmes, Thomas J. 2011. “The Diffusion of Wal-Mart and Economics of Density.” Econometrica 79 (1): 253-302. [22] Hopenhayn, Hugo, Julian Neira, and Rish Singhania. 2018. “From Population Growth to Firm Demographics: Implications for Concentration, Entrepreneurship and the Labor Share. ” NBER Working Paper 25382. 29  [23] Jia, Panle. 2008. “What Happens When Wal-Mart Comes to Town: An Empirical Analysis of the Discount Retailing Industry.” Econometrica 76 (6): 1263-1316. [24] Karabarbounis, Loukas, and Brent Neiman. 2018. “Accounting for Factorless Income.” NBER Working Paper 24404. [25] Neumark, David, Junfu Zhang, and Brandon Wall. 2006. “Employment Dynamics and Business Relocation: New Evidence from the National Establishment Time Series.” NBER Working Paper 11647. [26] Neumark, David, Brandon Wall, and Junfu Zhang. 2011. “Do Small Businesses Create More Jobs? New Evidence for the United States from the National Establishment Time Series.” The Review of Economics and Statistics 93 (1): 16-29. [27] Qiu, Yue, and Aaron Sojourner. 2019. “Labor-Market Concentration and Labor Compensation.” IZA Discussion Paper 12089. [28] Rinz, Kevin. 2018. “Labor Market Concentration, Earnings Inequality, and Earnings Mobility. ” Center for Administrative Records Research and Applications Working Paper 2018-10. [29] Syverson, Chad. 2019. “Macroeconomics and Market Power: Facts, Potential Explanations and Open Questions. ” Brookings Institution working paper. [30] Syverson, Chad. 2008. “Markets: Ready-Mixed Concrete.” Journal of Economic Perspectives 22 (1): 217-33. [31] Syverson, Chad. 2004. “Market Structure and Productivity: A Concrete Example.” Journal of Political Economy, 112 (6): 1181-1222. [32] Traina, James. 2018. “Is Aggregate Market Power Increasing? Production Trends Using Financial Statements.” Stigler Center New Working Paper Series No. 26. [33] Walls & Associates. 2014. “National Establishment Time-Series (NETS) Database.” [34] Zhu, Ting, Vishal Singh, and Mark D. Manuszak. 2009. “Market Structure and Competition in the Retail Discount Industry.” Journal of Marketing Research. 46 (4): 453-66.  30  ONLINE APPENDIX - NOT FOR PUBLICATION A  Data Description  A.1  NETS  The National Establishment Time Series (NETS) is made available through Walls & Associates, which relies on data compiled by Dun & Bradstreet (D&B). D&B provides each business establishment, corresponding to a distinct business activity by an enterprise at a specific location, a unique 9-digit Data Universal Numbering System (DUNS) number, which remains with that establishment even in the case of broader corporate-level changes, name changes, and so on.26 Each year, D&B compiles data for its Duns Marketing Information file on business characteristics of every establishment, including its sales, employment, location, primary industry, and the DUNS number of the establishment to which it reports (i.e. its parent company).27 As described by Neumark, Zhang, and Wall (2006), as well as Barnatchez, Crane, and Decker (2017), D&B makes an exhaustive effort to ensure that the file accurately covers the entire universe of business establishments, relying on many sources of information including direct phone calls, Yellow Pages, newspapers, and multiple government agencies. Furthermore, D&B and the establishments from which they gather information both have incentives to ensure information is accurate. While D&B compiles annual cross-sections of establishment characteristics, Walls & Associates aggregates these files into a longitudinal database that makes it possible to track the birth and death of establishments. Neumark et al. (2006) note that this process requires imputation of sales and employment to many establishment-year pairings. We use the NETS database to gather data on employment, sales, and the primary industry (8-digit SIC code) for each establishment for each year from 1990 through 2014, as well as the DUNS number of the establishment’s headquarters in each year (HQ or enterprise number). In each year, an enterprise is then defined as a collection of all the establishments with a given HQ number. We additionally collect the establishment’s county, ZIP code, and legal status, as well as the most recent HQ number of the establishment.28 The data is provided in wide form, with one observation per establishment and separate variables for establishment characteristics in each year, but we reshape the data into long form with one observation per establishment-year. We then drop any observations that have missing employment, sales, industry, or HQ numbers, and consider an establishment to “exist” in a given year if it has an observation associated with it (i.e. it has non-missing employment, sales, industry, and HQ numbers). We can then see the first year in which each establishment exists (its entry year) and the last year (its exit year).29 26  If an establishment goes out of business, its DUNS number will not be re-used. This file contains data on many other establishment characteristics that we do not consider in this paper. 28 The NETS database additionally has a variable for the business name, allowing us to identify the firms associated with specific HQ numbers. 29 Establishments that exist in 1990 are assigned 1990 as their entry year, as we have no data on them prior to this year. Likewise, establishments which exist in 2014 have that as their exit year. 27  31  Among these remaining establishment-year pairs, we identify the industry corresponding to the headquarters for each enterprise, and drop any establishments belonging to an enterprise whose headquarters has an SIC 8 industry code corresponding to the Public Administration division. We then drop any establishments which have an SIC 8 code either equal to 73899999 (Business Activities at Non-Commercial Sites), falling under Public Administration (even if their enterprise headquarters do not), the Educational Services 2-digit SIC sector (SIC 2 82), or the SIC 3-digit code 601, Central Reserve Depository. Additionally, we drop any establishments whose HQ number corresponds to the United States Postal Services (USPS), whose legal status identifies them as a non-profit organization, or which are in counties located outside of the 50 U.S. states and the District of Columbia. Among remaining establishments, we keep only those whose primary industry falls into one of the following five divisions: Manufacturing; Wholesale Trade; Retail Trade; Finance, Insurance, and Real Estate (FIRE); and Services. In the remaining dataset, we have roughly 41 million unique establishments spread across approximately 312 million establishment-year pairs. A.1.1  Data Quality  A number of researchers have attempted to compare the scope and accuracy of the NETS database to official sources such as the County Business Patterns (CBP), the Quarterly Census of Employment and Wages (QCEW), the Longitudinal Business Database (LBD), and Census Nonemployer Statistics (CES). Neumark et al. (2006); Neumark, Wall, and Zhang (2011); and Barnatchez et al. (2017) find that NETS reports substantially higher aggregate employment than these sources. This discrepancy seems to arise primarily from the inclusion of nonemployer establishments, which consist only of business owners and have no paid employees, in the NETS data; such establishments are generally not counted in government employment data. Since nonemployer establishments tend to have very small employment numbers, NETS vastly overstates the number of establishments in the 1-4 employee bin compared to Census counts of establishments with employment in this range, as noted by Neumark et al. (2006) and Barnatchez et al. (2017). Using an extract of the NETS data covering Georgia, Choi, Robertson, and Rupasingha (2013) find that, compared to the QCEW, NETS has nearly 75% more establishments in the state in 2000 and roughly three times as many in 2009. In the main text, we include these nonemployer establishments because they do report positive sales (and employment). Following the advice of Barnatchez et al. (2017), however, we show later in this appendix that the results hold when modifying the data to attempt to remove such establishments. While Neumark et al. (2006) argue that subtracting one from establishments’ employment counts to remove business owners makes the NETS universe comparable to that of official sources and eliminates most nonemployer establishments, Barnatchez et al. (2017) instead propose subtracting one from employment counts at the headquarters of each enterprise, because enterprise owners will generally only be counted at their headquarters. Along those lines, we present an alternative specification in which, for each year, we exclude sales and employment from all enterprises that report only one employee. However, there is some evidence that dropping single-employee enterprises may not remove the entire set 32  of nonemployers. While over one-quarter of establishment-year pairs in our dataset have only one employee, nearly as many report two employees. Furthermore, because the CBP and LBD report establishment counts in employment bins, we can only compare the number of establishments with between 1 and 4 employees in NETS and these other sources. In fact, Barnatchez et al. (2017) find that NETS still overcounts the number of establishments in this range even after dropping single-employee enterprises. Consequently, we explore a second specification in which we exclude sales and employment from all enterprise-year pairs reporting fewer than five employees.30 Additionally, we explore a third specification where we exclude sales and employment from enterprise-year pairs reporting fewer than ten employees. Surely there are many employer establishments with fewer than five and ten employees, so these specifications should be interpreted as conservative attempts to remove the influence of nonemployer establishments in the database. This notion is supported by the fact that, in every year, enterprises with one employee contain relatively immaterial shares of less than 5% and 3% of aggregate employment and sales, respectively, but enterprises with fewer than five employees contain rather substantial shares, between 11% and 23% of aggregate employment and between 10% and 16% of aggregate sales.31 There are two other potential issues regarding the NETS database to be addressed. First, Neumark et al. (2006) observe that NETS can be slow to report the birth and death of establishments, often operating on a two-to-three year lag in such cases. Second, the NETS data reported for each year are collected primarily in the prior year, and unlike official government sources NETS data are collected throughout the year, with establishments potentially reporting data at different months in different years.32 Because our dataset encompasses a 25-year period, such lags and inconsistencies in data collection timing should not affect the long-term trends we observe in the main text.  A.2  SIC 8 Codes  Our benchmark definition of an industry is an 8-digit Standard Industrial Classification (SIC) code. The first four digits of each SIC 8 code (SIC 4 codes) are created and determined by U.S. government agencies and assigned to business establishments. D&B supplements these codes with an additional four digits, providing a much finer level of detail regarding establishments’ primary activities; there are over 18, 000 unique 8-digit SIC codes compared to only about 1, 000 unique SIC codes at the 4-digit level and 84 at the 2-digit level. Each SIC code is also assigned to one of 11 divisions, five of which we consider in this paper.33 30  Since the vast majority of enterprises only have one establishment, results removing establishments based on establishment size rather than enterprise size should be roughly equivalent. 31 Likewise, enterprises with fewer than ten employees contain, in every year, between 19% and 31% of aggregate employment and between 18% and 22% of aggregate sales. 32 Some researchers roll back NETS data one year, but Barnatchez et al. (2017) find more favorable comparisons with government sources leaving years unchanged. 33 In addition to the primary 8-digit SIC code of each establishment, the NETS database reports the establishment’s primary North American Industry Classification System (NAICS) code. While government agencies developed the SIC system in the early 1900s, the Office of Management and Budget developed the NAICS system in 1992 to better reflect changes in the structure of the economy. Although the NAICS system contains a higher share of codes in more service-oriented industries, the most detailed NAICS code level only contains roughly 1, 100 industries, a level of aggregation much more comparable to the 4-digit  33  To better illustrate this “hierarchy” of SIC codes, consider the case of Walmart. As mentioned in the main text, the large majority of Walmart’s establishments in 2014 have SIC 8 53119901, Discount Department Stores, as their primary industry. This SIC 8 is a subset of SIC 4 5311, Department Stores, which also contains three other industries including SIC 8 53119902, Non-discount Department Stores. This SIC 4 code is further contained within the General Merchandise sector, SIC 2 53, which encompasses other industries corresponding to, for instance, Warehouse Club Stores and Miscellaneous General Merchandise. Finally, the Retail Trade division contains these industries and others as diverse as Grocery Stores, Optical Goods Stores, Eating Places, and Hardware Stores. Individual SIC 8 codes vary widely in their sizes as measured by employment, sales, and the number of establishments. For instance, among the 15, 305 SIC 8 codes considered in the main text, over onequarter of these have reported 2014 employment of fewer than 100 employees across all establishments with that primary industry.34 On the other hand, there are 206 industries with greater than 100, 000 reported employees, with the Discount Department Stores industry having over 1.8 million. Overall, industries in Retail Trade, FIRE, and Services have much higher employment on average than industries in Manufacturing and Wholesale Trade. To get a better sense of this heterogeneity across industries, Table (1) shows total employment and the number of SIC 8 industries within each of our SIC 2 sectors in 2014, as well as the division into which each SIC 2 code falls. Even at the sector level, there is substantial heterogeneity in these variables. Over 12 million and 10 million employees work in the Health Services and Business Services sectors, respectively, while many sectors in the Manufacturing division have only a few hundred thousand employees. Sectors in Retail Trade, FIRE, and Services tend to have higher employment than sectors in Manufacturing and Wholesale Trade, while the latter two divisions encompass over three-quarters of all SIC 8 codes. than the 8-digit SIC code. Consequently, we use SIC 8 codes as they offer by far the most granular available definition of an industry. 34 Approximately 90% of these industries are in the Manufacturing division.  34  Table 1: Employment and Number of SIC 8 Codes in Each SIC 2 Sector in 2014 SIC 2 Code  SIC 2 Description  Division  Employment in 2014  Number of SIC 8 codes in 2014  (Thousands)  Mean SIC 8 employment in 2014 (Thousands)  20  Food and Kindred Products  Manufacturing  1637  805  2  21  Tobacco Products  Manufacturing  25  11  2  22  Textile Mill Products  Manufacturing  343  587  1  23  Apparel, Finished Products  Manufacturing  386  385  1  Manufacturing  699  371  2  from Fabrics and Similar Materials 24  Lumber and Wood Products, Except Furniture  25  Furniture and Fixtures  Manufacturing  429  262  2  26  Paper and Allied Products  Manufacturing  680  328  2  27  Printing, Publishing and Allied  Manufacturing  1472  299  5  Industries 28  Chemicals and Allied Products  Manufacturing  1345  643  2  29  Petroleum Refining and Related  Manufacturing  189  75  3  Manufacturing  895  334  3  Industries 30  Rubber and Miscellaneous Plastic Products  31  Leather and Leather Products  Manufacturing  101  162  1  32  Stone, Clay, Glass, and  Manufacturing  576  516  1  33  Primary Metal Industries  Manufacturing  610  345  2  34  Fabricated Metal Products  Manufacturing  1488  736  2  35  Industrial and Commercial  Manufacturing  2144  1123  2  Manufacturing  1967  694  3  Concrete Products  Machinery and Computer Equipment 36  Electronic and Other Electrical Equipment and Components  37  Transportation Equipment  Manufacturing  1802  373  5  38  Measuring, Photographic,  Manufacturing  1401  784  2  Manufacturing  570  627  1  Wholesale Trade  4527  1104  4  Wholesale Trade  2966  653  5  Retail Trade  1400  77  18  Medical, and Optical Goods, and Clocks 39  Miscellaneous Manufacturing Industries  50  Wholesale Trade - Durable Goods  51  Wholesale Trade - Nondurable Goods  52  Building Materials, Hardware, Garden Supplies and Mobile Homes  53  General Merchandise Stores  Retail Trade  3346  11  304  54  Food Stores  Retail Trade  3935  61  65  55  Automotive Dealers and  Retail Trade  2783  68  41  Gasoline Service Stations 56  Apparel and Accessory Stores  Retail Trade  1462  78  19  57  Home Furniture, Furnishings  Retail Trade  1279  125  10  58  Eating and Drinking Places  Retail Trade  10446  81  129  59  Miscellaneous Retail  Retail Trade  4446  358  12  60  Depository Credit Institutions  FIRE  1974  37  53  61  Nondepository Credit  FIRE  749  65  12  62  Security and Commodity  FIRE  901  56  16  and Equipment Stores  Institutions Brokers, Dealers, Exchanges and Services 63  Insurance Carriers  FIRE  1191  77  15  64  Insurance Agents, Brokers and  FIRE  1373  27  51  65  Real Estate  FIRE  4217  63  67  67  Holding and Other Investment  FIRE  1679  48  35  Service  Offices  35  70  Hotels, Rooming Houses, Camps,  Services  2736  49  56  and Other Lodging Places 72  Personal Services  Services  2501  168  15  73  Business Services  Services  10524  487  22  75  Automotive Repair, Services and  Services  1822  106  17 4  Parking 76  Miscellaneous Repair Services  Services  928  227  78  Motion Pictures  Services  535  67  8  79  Amusement and Recreation  Services  2357  306  8  80  Health Services  Services  12200  203  60  81  Legal Services  Services  1869  25  75  83  Social Services  Services  2755  109  25  Museums, Art Galleries and  Services  125  15  8  Services  84  Botanical and Zoological Gardens 86  Membership Organizations  Services  3025  125  24  87  Engineering, Accounting,  Services  7927  198  40  Services  734  46  16  Research, and Management Services 89  Services, Not Elsewhere Classified  A.2.1  Percent of Sector-Level Employment in Industries with Diverging Trends  Table (2) provides more detail for Figure (9) in the main text and further highlights this degree of heterogeneity by displaying the exact percentages of employment in each sector and division across industries with diverging trends. In the column headings, αn and αz refer to the coefficients obtained from regressing the weighted average change in the HHI in each industry on the year with a constant at the national and ZIP code levels, respectively. The first three columns to the right of the sector and division descriptions show the percentage of employment in industries that have a positive national trend and positive, negative, and flat ZIP code trends, respectively. The last column displays the percentage of employment in industries with positive national trends located in industries that also have negative local trends. In all five divisions, over half of employment in industries with positive national trends is also located in industries that have declining concentration over time at the ZIP code level. Table 2: Percent of Sector Employment in Industries with Diverging Trends Division  SIC2  Description  Pct. Emp αn > 0, αz > 0  Pct. Emp αn > 0, αz < 0  Pct. Emp αn > 0, αz = 0  Pct. Emp αz < 0|αn > 0  Manufacturing  27.45  33.81  2.78  52.79  D  20  Food and Kindred Prod.  36.94  29.6  3.66  42.16  D  21  Tobacoo Prod.  17.18  0  3.79  0  D  22  Textile Mill Prod.  43.68  29.26  5.14  37.47  D  23  Apparel, Finished Prod.  52.12  23.91  1.8  30.72  26.51  33.49  1.28  54.65 24.70  D  from Fabrics D  24  Lumber and Wood Prod., Exc. Furn.  D  25  Furniture and Fixtures  52.83  18.19  2.63  D  26  Paper and Allied Prod.  22.74  50.72  5.67  64.1  D  27  Printing and Publishing  45.8  31.69  .46  40.65  D  28  Chemicals and Allied  16.9  30.32  2.78  60.64  Prod. D  29  Petroleum Refining  72.23  11.37  .85  13.46  D  30  Rubber and Misc. Plastic  22.11  49.7  2.95  66.48  Prod.  36  D  31  Leather and Leather Prod.  52.53  26.01  4.4  31.36  D  32  Stone Clay, Glass, and  18.33  39.32  4.07  63.7  Concrete Prod. D  33  Primary Metal Ind.  22.23  34.83  3.79  57.24  D  34  Fabricated Metal Prod.  23.39  32.32  2.57  55.46  D  35  Ind. and Comm.  24.94  31.52  2.61  53.36  14.41  39.46  2.26  70.31  Machinery and Comp. Equip. D  36  Electronic and Electric Equip.  D  37  Transport. Equip.  13.73  33.97  2.94  67.09  D  38  Instruments and Related  21.72  43.76  3.72  63.24  Products D  39  F  Misc. Manufact. Ind.  27.57  25.11  2.71  45.33  Wholesale Trade  34.17  39.01  .18  53.17  F  50  Wholesale- Durable Goods  31.65  39.71  .19  55.5  F  51  Wholesale- Nondurable  38.18  37.89  .16  49.7  Goods G G  52  Retail Trade  14.57  52.71  .01  78.33  Bldg. Materials and  47.91  31.25  .06  39.45 98.14  Garden. Supp. G  53  Gen. Merch. Stores  1.83  96.38  0  G  54  Food Stores  21.69  70.27  0  76.41  G  55  Auto. Dealers and Service  7.27  77.5  .01  91.41  G  56  Apparel and Access. Stores  6.42  77.24  0  92.33  G  57  Furn. and Homefurn.  20.92  55.51  .02  72.6  Stations  Stores G  58  Eating and Drinking Places  4.66  26.43  .01  85  G  59  Misc. Retail  30.94  42.54  0  57.89  Finance, Insurance, and  14.26  46.17  .04  76.35  H  Real Estate H  60  Depository Inst.  12.23  75.44  .01  86.04  H  61  Nondepository Inst.  11.27  38.33  .06  77.19  H  62  Security and Commod.  15.37  13.58  .03  46.86  Brokers H  63  Insurance Carr.  42.88  17.58  .23  28.97  H  64  Ins. Agents, Brokers, and  7.01  12.78  0  64.6  Service H  65  Real Estate  6.77  69.86  .01  91.16  H  67  Holding and Oth. Invest.  19.81  27.48  .04  58.06  Offices I  Services  9.26  40.44  .04  81.31  Hotels and Lodging Places  13.51  40.79  0  75.12  I  70  I  72  Personal Serv.  5.11  53.34  .15  91.03  I  73  Business Serv.  20.91  34.87  .01  62.5  I  75  Auto Repair, Serv., and  8.29  43.55  .01  84  Park. I  76  Misc. Repair Serv.  18.42  40.68  .04  68.78  I  78  Motion Pict.  37.54  47.23  .05  55.68  I  79  Amusement and Rec. Serv.  5.12  36.38  .06  87.52  I  80  Health Services  6.31  38.48  .08  85.74  I  81  Legal Services  .11  26.45  0  99.59  I  83  Social Services  .66  70.49  0  99.07  I  84  Museums, Art Gall., Zoos  2.95  3.99  .18  56.1  I  86  Membership Org.  .67  54.62  0  98.78  I  87  Engineering and Mgmt.  4.77  35.14  0  88.04  Services I  88  Private Households  0  53.19  39.43  57.43  I  89  Misc. Serv.  1.09  17.75  0  94.22  37  B  Robustness  The results in this section show that all the findings in the main text are robust to, in order, specifications removing small enterprises from the NETS database, using alternative measures of concentration, measuring concentration in terms of employment rather than sales, and using a balanced panel of industry-geography pairs. Finally, we expand on some of the results highlighted in the main text.  B.1  Removing Nonemployer Enterprises  As discussed above, this section contains three modifications to the NETS database in order to reduce the occurrence of nonemployer establishments in the data. The first modification, removing enterprise-year pairs with only one employee, likely leaves a large number of nonemployment establishments remaining, while dropping enterprise-year pairs with fewer than five and fewer than ten employees likely overstates the prevalence of such nonemployer establishments. B.1.1  Removing Enterprises with Only 1 Employee  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0  .05  Figure 21: Removing Enterprises with One Employee: Diverging economy-wide national and local concentration trends  -.25  National CBSA County ZIP 1990  1995  2000  Year  38  2005  2010  2014  National  ZIP Manufacturing Wholesale Trade Retail Trade FIRE Services  -.25  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0  .05  Figure 22: Removing Enterprises with One Employee: Diverging division-level national and local concentration trends  1990  1995  2000  Year  2005  2010  2014  National  ZIP SIC 2 SIC 4 SIC 6 SIC 8  -.15  Average Change in HHI from First Year -.1 -.05 0  .05  Figure 23: Removing Enterprises with One Employee: Diverging economy-wide trends in sales concentration  1990  1995  2000  Year  39  2005  2010  2015  National  ZIP SIC 2 SIC 4 SIC 6 SIC 8  -.15  Average Change in HHI from First Year -.1 -.05 0  .05  Figure 24: Removing Enterprises with One Employee: Diverging economy-wide trends in employment concentration  1990  1995  2000  Year  2005  2010  2015  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0  .05  Figure 25: Removing Enterprises with One Employee: Diverging economy-wide national and local concentration trends with a balanced panel  -.25  National CBSA County ZIP 1990  1995  2000  Year  40  2005  2010  2014  National  ZIP Manufacturing Wholesale Trade Retail Trade FIRE Services  -.25  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0  .05  Figure 26: Removing Enterprises with One Employee: Diverging division-level national and local concentration trends with a balanced panel  1990  1995  2000  Year  2005  2010  2014  0  Percent of Employment in SIC 2 20 40 60 80  100  Figure 27: Removing Enterprises with One Employee: Pervasive diverging trends across 2-digit sectors  53 55 56 54 57 59 52 58  Retail Trade  60 65 61 67 63 62 64  50 51  83 88 72 86 78 75 70 80 79 87 73 81 76 89 84  FIREWholesale Trade  Services  26 30 38 32 36 27 33 37 24 35 28 34 20 22 31 39 23 25 29 21  Manufacturing  Percent of Employment in SIC 8s with Increasing National and ZIP Trends Percent of Employment in SIC 8s with Increasing National and Decreasing ZIP Trends Percent of Employment in SIC 8s with Increasing National and Zero ZIP Trends  41  -.25  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0 .05  .1  Figure 28: Removing Enterprises with One Employee: The role of top enterprises in national and local concentration trends in diverging industries  Including Top Enterprise Excluding Top Enterprise ZIP Level National Level 1990  1995  2000  Year  2005  2010  2014  -.1  Average Change in HHI from First Year -.05 0 .05 .1  .15  Figure 29: Removing Enterprises with One Employee: The role of top enterprises in national and local concentration trends in concentrating industries  Including Top Enterprise Excluding Top Enterprise ZIP Level National Level 1990  1995  2000  Year  42  2005  2010  2014  -.3  Average Change in HHI from First Year -.25 -.2 -.15 -.1 -.05 0 .05  .1  Figure 30: Removing Enterprises with One Employee: Effect on concentration when the second and third largest enterprises enter a market in diverging industries  Incl. 2nd & 3rd Ranked Enterprises Excl. 2nd & 3rd Ranked Enterprises ZIP Level National Level 1990  1995  2000  Year  2005  2010  2014  -.1  Average Change in HHI from First Year -.05 0 .05 .1  .15  Figure 31: Removing Enterprises with One Employee: The role of the second and third largest enterprises in concentrating industries  Incl. 2nd & 3rd Ranked Enterprises Excl. 2nd & 3rd Ranked Enterprises ZIP Level National Level 1990  1995  2000  Year  43  2005  2010  2014  0  5  Percent 10  15  20  25  Figure 32: Removing Enterprises with One Employee: Expansion of top enterprises into ZIP codes  Percent Containing Largest Enterprise Percent Containing Largest and Second and/or Third Largest Enterprises 1990  1995  2000  Year  2005  2010  2015  Average Change in HHI from First Year -.05 -.03 -.01 .01  .03  Figure 33: Removing Enterprises with One Employee: Effect on concentration when a top enterprise enters a local market in diverging industries  -.07  Including Top Enterprise Excluding Top Enterprise  -3  -2  -1 0 1 2 3 4 5 Years Since SIC 8's Top Enterprise Opening in ZIP Code  44  6  7  Average Change in HHI from First Year -.09 -.07 -.05 -.03 -.01 .01  .03  Figure 34: Removing Enterprises with One Employee: Effect on concentration when a top enterprise enters a local market in concentrating industries  -.11  Including Top Enterprise Excluding Top Enterprise -3  -2  -1 0 1 2 3 4 5 Years Since SIC 8's Top Enterprise Opening in ZIP Code  6  7  Average Change in HHI from First Year  0  Figure 35: Removing Enterprises with One Employee: Effect on concentration when Walmart enters a local market  -.2  With Walmart Entering Without Walmart Entering  -3  -2  -1  0 1 2 3 4 Years Since Walmart Opening in ZIP  45  5  6  7  Average Number of Establishments in ZIP 2.5 3  3.5  Figure 36: Removing Enterprises with One Employee: Effect on number of establishments when Walmart enters a local market  2  With Walmart Entering Without Walmart Entering  -3  -2  -1  0 1 2 3 4 Years Since Walmart Opening in ZIP  5  6  7  Average Change in HHI from First Year -.1 -.05 0 .05  .1  Figure 37: Removing Enterprises with One Employee: Effect on concentration when Cemex enters a local market  -.15  With Cemex Entering Without Cemex Entering -3  -2  -1  0 1 2 3 4 Years Since Cemex Opening in ZIP  46  5  6  7  Average Number of Establishments in ZIP 2 2.5 3  3.5  Figure 38: Removing Enterprises with One Employee: Effect on number of establishments when Cemex enters a local market  1.5  With Cemex Entering Without Cemex Entering  -3  B.1.2  -2  -1  0 1 2 3 4 Years Since Cemex Opening in ZIP  5  6  7  Removing Enterprises with Fewer than 5 Employees  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0  .05  Figure 39: Removing Enterprises with Fewer than 5 Employees: Diverging economy-wide national and local concentration trends  -.25  National CBSA County ZIP 1990  1995  2000  Year  47  2005  2010  2014  National  ZIP Manufacturing Wholesale Trade Retail Trade FIRE Services  -.25  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0 .05  Figure 40: Removing Enterprises with Fewer than 5 Employees: Diverging division-level national and local concentration trends  1990  1995  2000  Year  2005  2010  2014  National  ZIP SIC 2 SIC 4 SIC 6 SIC 8  -.1  Average Change in HHI from First Year -.05 0  .05  Figure 41: Removing Enterprises with Fewer than 5 Employees: Diverging economy-wide trends in sales concentration  1990  1995  2000  Year  48  2005  2010  2015  National  ZIP SIC 2 SIC 4 SIC 6 SIC 8  -.1  Average Change in HHI from First Year -.05 0  .05  Figure 42: Removing Enterprises with Fewer than 5 Employees: Diverging economy-wide trends in employment concentration  1990  1995  2000  Year  2005  2010  2015  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0  .05  Figure 43: Removing Enterprises with Fewer than 5 Employees: Diverging economy-wide national and local concentration trends with a balanced panel  -.25  National CBSA County ZIP 1990  1995  2000  Year  49  2005  2010  2014  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0 .05  Figure 44: Removing Enterprises with Fewer than 5 Employees: Diverging division-level national and local concentration trends with a balanced panel  National  ZIP  -.25  Manufacturing Wholesale Trade Retail Trade FIRE Services 1990  1995  2000  2005  Year  2010  2014  0  Percent of Employment in SIC 2 20 40 60 80  100  Figure 45: Removing Enterprises with Fewer than 5 Employees: Pervasive diverging trends across 2-digit sectors  65 60 61 67 62 63 64  55 53 56 54 52 57 59 58  75 72 83 78 86 76 87 70 79 80 73 89 81 88 84  FIRE  Retail Trade  Services  50 51  26 30 38 32 24 36 27 20 28 33 37 31 34 22 35 23 25 39 29 21  Wholesale Trade  Manufacturing  Percent of Employment in SIC 8s with Increasing National and ZIP Trends Percent of Employment in SIC 8s with Increasing National and Decreasing ZIP Trends Percent of Employment in SIC 8s with Increasing National and Zero ZIP Trends  50  -.25  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0 .05  .1  Figure 46: Removing Enterprises with Fewer than 5 Employees: The role of top enterprises in national and local concentration trends in diverging industries  Including Top Enterprise Excluding Top Enterprise ZIP Level National Level 1990  1995  2000  Year  2005  2010  2014  -.1  Average Change in HHI from First Year -.05 0 .05 .1  .15  Figure 47: Removing Enterprises with Fewer than 5 Employees: The role of top enterprises in national and local concentration trends in concentrating industries  Including Top Enterprise Excluding Top Enterprise ZIP Level National Level 1990  1995  2000  Year  51  2005  2010  2014  -.3  Average Change in HHI from First Year -.25 -.2 -.15 -.1 -.05 0 .05  .1  Figure 48: Removing Enterprises with Fewer than 5 Employees: The role of the second and third largest enterprises in diverging industries  Incl. 2nd & 3rd Ranked Enterprises Excl. 2nd & 3rd Ranked Enterprises ZIP Level National Level 1990  1995  2000  Year  2005  2010  2014  -.1  Average Change in HHI from First Year -.05 0 .05 .1  .15  Figure 49: Removing Enterprises with Fewer than 5 Employees: The role of the second and third largest enterprises in concentrating industries  Incl. 2nd & 3rd Ranked Enterprises Excl. 2nd & 3rd Ranked Enterprises ZIP Level National Level 1990  1995  2000  Year  52  2005  2010  2014  0  5  Percent 10  15  20  25  Figure 50: Removing Enterprises with Fewer than 5 Employees: Expansion of top enterprises into ZIP codes  Percent Containing Largest Enterprise Percent Containing Largest and Second and/or Third Largest Enterprises 1990  1995  2000  Year  2005  2010  2015  Average Change in HHI from First Year -.07 -.05 -.03 -.01 .01  .03  Figure 51: Removing Enterprises with Fewer than 5 Employees: Effect on concentration when a top enterprise enters a local market in diverging industries  Including Top Enterprise Excluding Top Enterprise  -3  -2  -1 0 1 2 3 4 5 Years Since SIC 8's Top Enterprise Opening in ZIP Code  53  6  7  Average Change in HHI from First Year -.09 -.07 -.05 -.03 -.01 .01  .03  Figure 52: Removing Enterprises with Fewer than 5 Employees: Effect on concentration when a top enterprise enters a local market in concentrating industries  -.11  Including Top Enterprise Excluding Top Enterprise -3  -2  -1 0 1 2 3 4 5 Years Since SIC 8's Top Enterprise Opening in ZIP Code  6  7  Average Change in HHI from First Year -.15 -.1 -.05 0  .05  Figure 53: Removing Enterprises with Fewer than 5 Employees: Effect on concentration when Walmart enters a local market  -.2  With Walmart Entering Without Walmart Entering  -3  -2  -1  0 1 2 3 4 Years Since Walmart Opening in ZIP  54  5  6  7  Average Number of Establishments in ZIP 2 2.5  3  Figure 54: Removing Enterprises with Fewer than 5 Employees: Effect on number of establishments when Walmart enters a local market  1.5  With Walmart Entering Without Walmart Entering  -3  -2  -1  0 1 2 3 4 Years Since Walmart Opening in ZIP  5  6  7  Average Change in HHI from First Year -.1 -.05 0 .05  .1  Figure 55: Removing Enterprises with Fewer than 5 Employees: Effect on concentration when Cemex enters a local market  -.15  With Cemex Entering Without Cemex Entering -3  -2  -1  0 1 2 3 4 Years Since Cemex Opening in ZIP  55  5  6  7  Average Number of Establishments in ZIP 2 2.5 3  3.5  Figure 56: Removing Enterprises with Fewer than 5 Employees: Effect on number of establishments when Cemex enters a local market  1.5  With Cemex Entering Without Cemex Entering  -3  B.1.3  -2  -1  0 1 2 3 4 Years Since Cemex Opening in ZIP  5  6  7  Removing Enterprises with Fewer than 10 Employees  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0  .05  Figure 57: Removing Enterprises with Fewer than 10 Employees: Diverging economy-wide national and local concentration trends  -.25  National CBSA County ZIP 1990  1995  2000  Year  56  2005  2010  2014  National  ZIP Manufacturing Wholesale Trade Retail Trade FIRE Services  -.25  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0 .05  Figure 58: Removing Enterprises with Fewer than 10 Employees: Diverging division-level national and local concentration trends  1990  1995  2000  Year  2005  2010  2014  National  ZIP SIC 2 SIC 4 SIC 6 SIC 8  -.1  Average Change in HHI from First Year -.05 0  .05  Figure 59: Removing Enterprises with Fewer than 10 Employees: Diverging economy-wide trends in sales concentration  1990  1995  2000  Year  57  2005  2010  2015  National  ZIP SIC 2 SIC 4 SIC 6 SIC 8  -.1  Average Change in HHI from First Year -.05 0  .05  Figure 60: Removing Enterprises with Fewer than 10 Employees: Diverging economy-wide trends in employment concentration  1990  1995  2000  Year  2005  2010  2015  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0  .05  Figure 61: Removing Enterprises with Fewer than 10 Employees: Diverging economy-wide national and local concentration trends with a balanced panel  -.25  National CBSA County ZIP 1990  1995  2000  Year  58  2005  2010  2014  National  ZIP Manufacturing Wholesale Trade Retail Trade FIRE Services  -.25  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0 .05  Figure 62: Removing Enterprises with Fewer than 10 Employees: Diverging division-level national and local concentration trends with a balanced panel  1990  1995  2000  Year  2005  2010  2014  0  Percent of Employment in SIC 2 20 40 60 80  100  Figure 63: Removing Enterprises with Fewer than 10 Employees: Pervasive diverging trends across 2-digit sectors  55 52 53 56 54 57 59 58  Retail Trade  65 60 61 67 63 64 62  51 50  72 78 83 86 75 76 73 79 70 80 87 89 84 88 81  FIREWholesale Trade  Services  26 38 30 25 32 27 20 36 35 28 33 37 31 34 24 23 22 39 29 21  Manufacturing  Percent of Employment in SIC 8s with Increasing National and ZIP Trends Percent of Employment in SIC 8s with Increasing National and Decreasing ZIP Trends Percent of Employment in SIC 8s with Increasing National and Zero ZIP Trends  59  -.25  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0 .05  .1  Figure 64: Removing Enterprises with Fewer than 10 Employees: The role of top enterprises in national and local concentration trends in diverging industries  Including Top Enterprise Excluding Top Enterprise ZIP Level National Level 1990  1995  2000  Year  2005  2010  2014  -.1  Average Change in HHI from First Year -.05 0 .05 .1  .15  Figure 65: Removing Enterprises with Fewer than 10 Employees: The role of top enterprises in national and local concentration trends in concentrating industries  Including Top Enterprise Excluding Top Enterprise ZIP Level National Level 1990  1995  2000  Year  60  2005  2010  2014  -.3  Average Change in HHI from First Year -.25 -.2 -.15 -.1 -.05 0 .05 .1  Figure 66: Removing Enterprises with Fewer than 10 Employees: The role of the second and third largest enterprises in diverging industries  Incl. 2nd & 3rd Ranked Enterprises Excl. 2nd & 3rd Ranked Enterprises ZIP Level National Level 1990  1995  2000  Year  2005  2010  2014  -.1  Average Change in HHI from First Year -.05 0 .05 .1  .15  Figure 67: Removing Enterprises with Fewer than 10 Employees: The role of the second and third largest enterprises in concentrating industries  Incl. 2nd & 3rd Ranked Enterprises Excl. 2nd & 3rd Ranked Enterprises ZIP Level National Level 1990  1995  2000  Year  61  2005  2010  2014  0  5  Percent 10  15  20  25  Figure 68: Removing Enterprises with Fewer than 10 Employees: Expansion of top enterprises into ZIP codes  Percent Containing Largest Enterprise Percent Containing Largest and Second and/or Third Largest Enterprises 1990  1995  2000  Year  2005  2010  2015  Average Change in HHI from First Year -.07 -.05 -.03 -.01 .01  .03  Figure 69: Removing Enterprises with Fewer than 10 Employees: Effect on concentration when a top enterprise enters a local market in diverging industries  Including Top Enterprise Excluding Top Enterprise  -3  -2  -1 0 1 2 3 4 5 Years Since SIC 8's Top Enterprise Opening in ZIP Code  62  6  7  Average Change in HHI from First Year -.09 -.07 -.05 -.03 -.01 .01  .03  Figure 70: Removing Enterprises with Fewer than 10 Employees: Effect on concentration when a top enterprise enters a local market in concentrating industries  -.11  Including Top Enterprise Excluding Top Enterprise -3  -2  -1 0 1 2 3 4 5 Years Since SIC 8's Top Enterprise Opening in ZIP Code  6  7  Average Change in HHI from First Year -.15 -.1 -.05 0  .05  Figure 71: Removing Enterprises with Fewer than 10 Employees: Effect on concentration when Walmart enters a local market  -.2  With Walmart Entering Without Walmart Entering  -3  -2  -1  0 1 2 3 4 Years Since Walmart Opening in ZIP  63  5  6  7  Average Number of Establishments in ZIP 2 2.5  3  Figure 72: Removing Enterprises with Fewer than 10 Employees: Effect on number of establishments when Walmart enters a local market  1.5  With Walmart Entering Without Walmart Entering  -3  -2  -1  0 1 2 3 4 Years Since Walmart Opening in ZIP  5  6  7  Average Change in HHI from First Year -.1 -.05 0 .05  .1  Figure 73: Removing Enterprises with Fewer than 10 Employees: Effect on concentration when Cemex enters a local market  -.15  With Cemex Entering Without Cemex Entering -3  -2  -1  0 1 2 3 4 Years Since Cemex Opening in ZIP  64  5  6  7  Average Number of Establishments in ZIP 2 2.5 3  3.5  Figure 74: Removing Enterprises with Fewer than 10 Employees: Effect on number of establishments when Cemex enters a local market  1.5  With Cemex Entering Without Cemex Entering  -3  B.2  -2  -1  0 1 2 3 4 Years Since Cemex Opening in ZIP  5  6  7  Alternative Measures of Concentration  While the figures in the main text all rely on the Herfindahl-Hirschman Index (HHI), here we replicate them for alternative measures of concentration. In particular, we look at the adjusted HHI, which modifies the HHI for the number of enterprises in a market, as well as the share of the top enterprise, as measured by sales, for each industry-geography grouping in every year. For reasons discussed in more detail below, we believe the HHI used in the main text remains the best measure of concentration; however, the results in this section show that all the findings in the main text still hold using these alternative measures.35 B.2.1  Adjusted HHI  I,G Let Ci,g,t denote the HHI for industry i in geography g in year t (using the level of industrial and geoI,G graphic aggregations I and G, respectively), and let Ni,g,t denote the number of enterprises in this industryh i I,G I,G I,G geography-year grouping. Then Ci,g,t ∈ 1/Ni,g,t , 1 . Because Ci,g,t is bounded below by the inverse of the  number of enterprises, comparisons of the HHI between groupings with different numbers of enterprises can be somewhat difficult. While looking at changes in concentration rather than levels, as we do in the main text, allows for efficient comparisons between groups with different numbers of enterprises, such comparisons can also be made using the adjusted Herfindahl-Hirschman Index, which for any pair with more than 35  When using alternative measures of concentration, we do not replicate Figures (14), (18), and (20) in the main text as those figures are not dependent on the measure of concentration we use.  65  1 enterprise can take on any value between 0 and 1, inclusive. In particular, the adjusted HHI of industry i in geography g in year t, C˜ I,G , can be defined as i,g,t  I,G C˜i,g,t =       I,G Ci,g,t −     1  1−  N  1 I,G i,g,t  1 I,G N i,g,t  I,G Ni,g,t >1 I,G Ni,g,t  .  (2)  =1  For groupings with a very large numbers of enterprises (for example, most groupings with a geography defined at the national level), the adjusted and unadjusted HHIs will be very close. However, groupings defined at the ZIP code level typically have a small number of enterprises, leading to potentially large differences between the adjusted and unadjusted measures. In such cases, the unadjusted HHI is preferable because, in some sense, the number of an enterprises in a market itself partly determines concentration. That is, a market with, say, 3 enterprises is arguably more concentrated than a market with 10, even if all enterprises have equal sales in both markets. The figures below replicate the figures in the main text using changes in the adjusted HHI.  Average Change in Adjusted HHI from First Year -.2 -.15 -.1 -.05 0 .05  Figure 75: Adjusted HHI: Diverging economy-wide national and local concentration trends  -.25  National CBSA County ZIP 1990  1995  2000  Year  66  2005  2010  2014  National  ZIP Manufacturing Wholesale Trade Retail Trade FIRE Services  -.25  Average Change in Adjusted HHI from First Year -.2 -.15 -.1 -.05 0 .05  Figure 76: Adjusted HHI: Diverging division-level national and local concentration trends  1990  1995  2000  Year  2005  2010  2014  Average Change in Adjusted HHI from First Year -.15 -.1 -.05 0 .05  Figure 77: Adjusted HHI: Diverging economy-wide trends in sales concentration  National  ZIP SIC 2 SIC 4 SIC 6 SIC 8  1990  1995  2000  Year  67  2005  2010  2015  Average Change in Adjusted HHI from First Year -.15 -.1 -.05 0 .05  Figure 78: Adjusted HHI: Diverging economy-wide trends in employment concentration  National  ZIP SIC 2 SIC 4 SIC 6 SIC 8  1990  1995  2000  Year  2005  2010  2015  Average Change in Adjusted HHI from First Year -.2 -.15 -.1 -.05 0 .05  Figure 79: Adjusted HHI: Diverging economy-wide national and local concentration trends with a balanced panel  -.25  National CBSA County ZIP 1990  1995  2000  Year  68  2005  2010  2014  National  ZIP Manufacturing Wholesale Trade Retail Trade FIRE Services  -.25  Average Change in Adjusted HHI from First Year -.2 -.15 -.1 -.05 0 .05  Figure 80: Adjusted HHI: Diverging division-level national and local concentration trends with a balanced panel  1990  1995  2000  2005  Year  2010  2014  0  Percent of Employment in SIC 2 20 40 60 80  100  Figure 81: Adjusted HHI: Pervasive diverging trends across 2-digit sectors  53 56 55 54 57 59 52 58  60 65 61 67 63 62 64  83 86 72 88 78 75 80 76 70 79 87 73 81 89 84  Retail Trade  FIRE  Services  50 51  26 30 38 32 36 24 37 33 35 34 27 28 20 22 31 39 25 23 29 21  Wholesale Trade  Manufacturing  Percent of Employment in SIC 8s with Increasing National and ZIP Trends Percent of Employment in SIC 8s with Increasing National and Decreasing ZIP Trends Percent of Employment in SIC 8s with Increasing National and Zero ZIP Trends  69  -.2  Average Change in Adjusted HHI from First Year -.15 -.1 -.05 0 .05 .1  Figure 82: Adjusted HHI: The role of top enterprises in national and local concentration trends in diverging industries  Including Top Enterprise Excluding Top Enterprise ZIP Level National Level 1990  1995  2000  Year  2005  2010  2014  -.1  Average Change in Adjusted HHI from First Year -.05 0 .05 .1 .15  Figure 83: Adjusted HHI: The role of top enterprises in national and local concentration trends in concentrating industries  Including Top Enterprise Excluding Top Enterprise ZIP Level National Level 1990  1995  2000  Year  70  2005  2010  2014  -.25  Average Change in Adjusted HHI from First Year -.2 -.15 -.1 -.05 0 .05 .1  Figure 84: Adjusted HHI: Effect on concentration when the second and third largest enterprises enter a market in diverging industries  Incl. 2nd & 3rd Ranked Enterprises Excl. 2nd & 3rd Ranked Enterprises ZIP Level National Level 1990  1995  2000  Year  2005  2010  2014  -.1  Average Change in Adjusted HHI from First Year -.05 0 .05 .1 .15  Figure 85: Adjusted HHI: Effect on concentration when the second and third largest enterprises enter a market in concentrating industries  Incl. 2nd & 3rd Ranked Enterprises Excl. 2nd & 3rd Ranked Enterprises ZIP Level National Level 1990  1995  2000  Year  71  2005  2010  2014  Average Change in Adjusted HHI from First Year -.07 -.05 -.03 -.01 .01 .03  Figure 86: Adjusted HHI: Effect on concentration when a top enterprise enters a local market in diverging industries  Including Top Enterprise Excluding Top Enterprise  -3  -2  -1 0 1 2 3 4 5 Years Since SIC8's Top Enterprise Opening in ZIP Code  6  7  Average Change in Adjusted HHI from First Year -.09 -.07 -.05 -.03 -.01 .01 .03  Figure 87: Adjusted HHI: Effect on concentration when a top enterprise enters a local market in concentrating industries  -.11  Including Top Enterprise Excluding Top Enterprise -3  -2  -1 0 1 2 3 4 5 Years Since SIC8's Top Enterprise Opening in ZIP Code  72  6  7  Average Change in Adjusted HHI from First Year -.2 -.1 0  Figure 88: Adjusted HHI: Effect on concentration when Walmart enters a local market  -.3  With Walmart Entering Without Walmart Entering  -3  -2  -1  0 1 2 3 4 Years Since Walmart Opening in ZIP  5  6  7  Average Change in Adjusted HHI from First Year -.2 -.15 -.1 -.05 0 .05 .1 .15  Figure 89: Adjusted HHI: Effect on concentration when Cemex enters a local market  -.25  With Cemex Entering Without Cemex Entering -3  -2  -1  0 1 2 3 4 Years Since Cemex Opening in ZIP  73  5  6  7  B.2.2  Share of Top Enterprise  Another common measure of market concentration is the concentration ratio, which looks at the total market share accounted for by a certain number of top firms in a market. Here, we measure concentration in a geography-industry-year grouping as that pair’s share of total sales in the top enterprise measured by sales. This share will obviously equal 1 for any geography-industry-year groupings with only one enterprise. We prefer the HHI as a measure of concentration because the HHI captures in a more precise way the entire distribution of market shares. The share of the top enterprise fails to capture any variation in the structure of market shares among enterprises beyond the top enterprise. For instance, this measure would conclude that among two markets in which the top enterprises control 60% of total sales, a market in which there is only one other enterprise comprising the remaining 40% of sales is just as concentrated as one in which ten enterprises each have 4% of sales. In contrast, the HHI would indicate considerably more concentration in the first market.  Average Change in Top 1 Share from First Year -.2 -.15 -.1 -.05 0  .05  Figure 90: Share of Top Enterprise: Diverging economy-wide national and local concentration trends  -.25  National CBSA County ZIP 1990  1995  2000  Year  74  2005  2010  2014  National  ZIP Manufacturing Wholesale Trade Retail Trade FIRE Services  -.25  Average Change in Top 1 Share from First Year -.2 -.15 -.1 -.05 0 .05  Figure 91: Share of Top Enterprise: Diverging division-level national and local concentration trends  1990  1995  2000  Year  2005  2010  2014  National  ZIP SIC 2 SIC 4 SIC 6 SIC 8  -.15  Average Change in Top 1 Share from First Year -.1 -.05 0 .05  Figure 92: Share of Top Enterprise: Diverging economy-wide trends in sales concentration  1990  1995  2000  Year  75  2005  2010  2015  National  ZIP SIC 2 SIC 4 SIC 6 SIC 8  -.15  Average Change in Top 1 Share from First Year -.1 -.05 0  .05  Figure 93: Share of Top Enterprise: Diverging economy-wide trends in employment concentration  1990  1995  2000  Year  2005  2010  2015  Average Change in Top 1 Share from First Year -.2 -.15 -.1 -.05 0  .05  Figure 94: Share of Top Enterprise: Diverging economy-wide national and local concentration trends with a balanced panel  -.25  National CBSA County ZIP 1990  1995  2000  Year  76  2005  2010  2014  National  ZIP Manufacturing Wholesale Trade Retail Trade FIRE Services  -.25  Average Change in Top 1 Share from First Year -.2 -.15 -.1 -.05 0 .05  Figure 95: Share of Top Enterprise: Diverging division-level national and local concentration trends with a balanced panel  1990  1995  2000  2005  Year  2010  2014  0  Percent of Employment in SIC 2 20 40 60 80  100  Figure 96: Share of Top Enterprise: Pervasive diverging trends across 2-digit sectors  53 55 54 57 56 59 58 52  60 65 61 67 63 64 62  83 72 86 88 78 87 76 80 79 75 73 81 70 89 84  Retail Trade  FIRE  Services  50 51  26 28 32 38 30 36 34 27 24 20 37 35 33 22 39 31 23 25 29 21  Wholesale Trade  Manufacturing  Percent of Employment in SIC 8s with Increasing National and ZIP Trends Percent of Employment in SIC 8s with Increasing National and Decreasing ZIP Trends Percent of Employment in SIC 8s with Increasing National and Zero ZIP Trends  77  -.2  Average Change in Top 1 Share from First Year -.15 -.1 -.05 0 .05 .1  Figure 97: Share of Top Enterprise: The role of top enterprises in national and local concentration trends in diverging industries  Including Top Enterprise Excluding Top Enterprise ZIP Level National Level 1990  1995  2000  Year  2005  2010  2014  -.1  Average Change in Top 1 Share from First Year -.05 0 .05 .1  .15  Figure 98: Share of Top Enterprise: The role of top enterprises in national and local concentration trends in concentrating industries  Including Top Enterprise Excluding Top Enterprise ZIP Level National Level 1990  1995  2000  Year  78  2005  2010  2014  -.2  Average Change in Top 1 Share from First Year -.15 -.1 -.05 0 .05 .1  Figure 99: Share of Top Enterprise: The role of the second and third largest enterprises in diverging industries  Incl. 2nd & 3rd Ranked Enterprises Excl. 2nd & 3rd Ranked Enterprises ZIP Level National Level 1990  1995  2000  Year  2005  2010  2014  -.1  Average Change in Top 1 Share from First Year -.05 0 .05 .1  .15  Figure 100: Share of Top Enterprise: The role of the second and third largest enterprises in concentrating industries  Incl. 2nd & 3rd Ranked Enterprises Excl. 2nd & 3rd Ranked Enterprises ZIP Level National Level 1990  1995  2000  Year  79  2005  2010  2014  Average Change in Top 1 Share from First Year -.05 -.03 -.01 .01  .03  Figure 101: Share of Top Enterprise: Effect on concentration when a top enterprise enters a local market in diverging industries  -.07  Including Top Enterprise Excluding Top Enterprise  -3  -2  -1 0 1 2 3 4 5 Years Since SIC 8's Top Enterprise Opening in ZIP Code  6  7  Average Change in Top 1 Share from First Year -.05 -.03 -.01 .01  .03  Figure 102: Share of Top Enterprise: Effect on concentration when a top enterprise enters a local market in concentrating industries  -.07  Including Top Enterprise Excluding Top Enterprise -3  -2  -1 0 1 2 3 4 5 Years Since SIC 8's Top Enterprise Opening in ZIP Code  80  6  7  Average Change in Top 1 Share from First Year -.1 -.05 0  .05  Figure 103: Share of Top Enterprise: Effect on concentration when Walmart enters a local market  -.15  With Walmart Entering Without Walmart Entering  -3  -2  -1  0 1 2 3 4 Years Since Walmart Opening in ZIP  5  6  7  Average Change in Top 1 Share from First Year -.1 -.05 0 .05  .1  Figure 104: Share of Top Enterprise: Effect on concentration when Cemex enters a local market  -.15  With Cemex Entering Without Cemex Entering -3  -2  -1  0 1 2 3 4 Years Since Cemex Opening in ZIP  81  5  6  7  B.3  Employment Concentration  In the main text, we measure concentration in terms of sales in each geography-industry pair. In this section, we replicate the figures in the main text where, for each geography-industry pair in each year, we instead calculate the HHI of employment, rather than sales.36  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0  .05  Figure 105: HHI of Employment: Diverging economy-wide national and local concentration trends  -.25  National CBSA County ZIP 1990  1995  2000  Year  36  2005  2010  2014  When measuring the HHI of employment, we do not replicate Figures (14), (18), and (20) in the main text as those figures are not dependent on which measure for which we calculate concentration.  82  National  ZIP Manufacturing Wholesale Trade Retail Trade FIRE Services  -.25  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0  .05  Figure 106: HHI of Employment: Diverging division-level national and local concentration trends  1990  1995  2000  Year  2005  2010  2014  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0  .05  Figure 107: HHI of Employment: Diverging economy-wide national and local concentration trends with a balanced panel  -.25  National CBSA County ZIP 1990  1995  2000  Year  83  2005  2010  2014  National  ZIP Manufacturing Wholesale Trade Retail Trade FIRE Services  -.25  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0  .05  Figure 108: HHI of Employment: Diverging division-level national and local concentration trends with a balanced panel  1990  1995  2000  Year  2005  2010  2014  0  Percent of Employment in SIC 2 20 40 60 80  100  Figure 109: HHI of Employment: Pervasive diverging trends across 2-digit sectors  53 55 56 54 57 59 52 58  60 62 61 65 63 67 64  Retail Trade  FIRE  50 51  26 38 32 30 35 24 34 36 37 20 28 22 33 31 23 27 25 39 29 21  Whls. Trd.  Manufacturing  72 88 75 83 86 78 76 87 73 79 81 70 80 89 84  Services  Percent of Employment in SIC 8s with Increasing National and ZIP Trends Percent of Employment in SIC 8s with Increasing National and Decreasing ZIP Trends Percent of Employment in SIC 8s with Increasing National and Zero ZIP Trends  84  -.2  Average Change in HHI from First Year -.15 -.1 -.05 0 .05  .1  Figure 110: HHI of Employment: The role of top enterprises in national and local concentration trends in diverging industries  Including Top Enterprise Excluding Top Enterprise ZIP Level National Level 1990  1995  2000  Year  2005  2010  2014  -.1  Average Change in HHI from First Year -.05 0 .05 .1  .15  Figure 111: HHI of Employment: The role of top enterprises in national and local concentration trends in concentrating industries  Including Top Enterprise Excluding Top Enterprise ZIP Level National Level 1990  1995  2000  Year  85  2005  2010  2014  -.25  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0 .05  .1  Figure 112: HHI of Employment: Effect on concentration when the second and third largest enterprises enter a market in diverging industries  Incl. 2nd & 3rd Ranked Enterprises Excl. 2nd & 3rd Ranked Enterprises ZIP Level National Level 1990  1995  2000  Year  2005  2010  2014  -.1  Average Change in HHI from First Year -.05 0 .05 .1  .15  Figure 113: HHI of Employment: The role of the second and third largest enterprises in concentrating industries  Incl. 2nd & 3rd Ranked Enterprises Excl. 2nd & 3rd Ranked Enterprises ZIP Level National Level 1990  1995  2000  Year  86  2005  2010  2014  Average Change in HHI from First Year -.05 -.03 -.01 .01  .03  Figure 114: HHI of Employment: Effect on concentration when a top enterprise enters a local market in diverging industries  -.07  Including Top Enterprise Excluding Top Enterprise  -3  -2  -1 0 1 2 3 4 5 Years Since SIC 8's Top Enterprise Opening in ZIP Code  6  7  Average Change in HHI from First Year -.05 -.03 -.01 .01 .03  Figure 115: HHI of Employment: Effect on concentration when a top enterprise enters a local market in concentrating industries  -.07  Including Top Enterprise Excluding Top Enterprise -3  -2  -1 0 1 2 3 4 5 Years Since SIC 8's Top Enterprise Opening in ZIP Code  87  6  7  Average Change in HHI from First Year -.15 -.1 -.05 0  .05  Figure 116: HHI of Employment: Effect on concentration when Walmart enters a local market  -.2  With Walmart Entering Without Walmart Entering  -3  -2  -1  0 1 2 3 4 Years Since Walmart Opening in ZIP  5  6  7  Average Change in HHI from First Year -.1 -.05 0 .05  .1  Figure 117: HHI of Employment: Effect on concentration when Cemex enters a local market  -.15  With Cemex Entering Without Cemex Entering -3  -2  -1  0 1 2 3 4 Years Since Cemex Opening in ZIP  88  5  6  7  B.4  Using a Balanced Panel  In the main text, in every year we consider all industry-geography pairs which have at least one establishment present. Because establishments are entering and exiting local markets over time, some industry-geography pairs may have establishments present in some years but not others. In this section, we reproduce the figures in the main text considering only those industry-geography pairs which have at least one establishment present in all 25 years in our sample.37  National  ZIP SIC 2 SIC 4 SIC 6 SIC 8  -.15  Average Change in HHI from First Year -.1 -.05 0  .05  Figure 118: Balanced Panel: Diverging economy-wide trends in sales concentration  1990  1995  2000  Year  37  2005  2010  2015  In this section, we do not reproduce Figures (3), (4), (7), and (8) in the main text as Figures (7) and (8) already reproduce Figures (3) and (4), respectively, with a balanced panel.  89  National  ZIP SIC 2 SIC 4 SIC 6 SIC 8  -.1  Average Change in HHI from First Year -.05 0  .05  Figure 119: Balanced Panel: Diverging economy-wide trends in employment concentration  1990  1995  2000  Year  2005  2010  2015  0  Percent of Employment in SIC 2 20 40 60 80  100  Figure 120: Balanced Panel: Pervasive diverging trends across 2-digit sectors  53 55 54 56 57 59 58 52  65 60 67 61 63 62 64  83 86 72 78 80 76 87 79 75 81 73 89 84 70  32 38 24 34 22 20 26 36 27 30 33 28 35 37 31 23 39 25 29 21  Retail Trade  FIRE  Services  Manufacturing  Percent of Employment in SIC 8s with Increasing National and ZIP Trends Percent of Employment in SIC 8s with Increasing National and Decreasing ZIP Trends Percent of Employment in SIC 8s with Increasing National and Zero ZIP Trends  90  50 51  Wholesale Trd.  -.2  Average Change in HHI from First Year -.15 -.1 -.05 0 .05  .1  Figure 121: Balanced Panel: The role of top enterprises in national and local concentration trends in diverging industries  Including Top Enterprise Excluding Top Enterprise ZIP Level National Level 1990  1995  2000  Year  2005  2010  2014  -.1  Average Change in HHI from First Year -.05 0 .05 .1  .15  Figure 122: Balanced Panel: The role of top enterprises in national and local concentration trends in concentrating industries  Including Top Enterprise Excluding Top Enterprise ZIP Level National Level 1990  1995  2000  Year  91  2005  2010  2014  -.25  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0 .05  .1  Figure 123: Balanced Panel: Effect on concentration when the second and third largest enterprises enter a market in diverging industries  Incl. 2nd & 3rd Ranked Enterprises Excl. 2nd & 3rd Ranked Enterprises ZIP Level National Level 1990  1995  2000  Year  2005  2010  2014  -.1  Average Change in HHI from First Year -.05 0 .05 .1  .15  Figure 124: Balanced Panel: Effect on concentration when the second and third largest enterprises enter a market in concentrating industries  Incl. 2nd & 3rd Ranked Enterprises Excl. 2nd & 3rd Ranked Enterprises ZIP Level National Level 1990  1995  2000  Year  92  2005  2010  2014  Average Change in HHI from First Year -.05 -.03 -.01 .01  .03  Figure 125: Balanced Panel: Effect on concentration when a top enterprise enters a local market in diverging industries  -.07  Including Top Enterprise Excluding Top Enterprise -3  -2  -1 0 1 2 3 4 5 Years Since SIC 8's Top Enterprise Opening in ZIP Code  6  7  Average Change in HHI from First Year -.05 -.03 -.01 .01  .03  Figure 126: Balanced Panel: Effect on concentration when a top enterprise enters a local market in concentrating industries  -.07  Including Top Enterprise Excluding Top Enterprise -3  -2  -1 0 1 2 3 4 5 Years Since SIC 8's Top Enterprise Opening in ZIP Code  93  6  7  Average Change in HHI from First Year -.15 -.1 -.05 0  Figure 127: Balanced Panel: Effect on concentration when Walmart enters a local market  -.2  With Walmart Entering Without Walmart Entering  -3  -2  -1  0 1 2 3 4 Years Since Walmart Opening in ZIP  5  6  7  Average Number of Establishments in ZIP 2.5 3  3.5  Figure 128: Balanced Panel: Effect on number of establishments when Walmart enters a local market  2  With Walmart Entering Without Walmart Entering  -3  -2  -1  0 1 2 3 4 Years Since Walmart Opening in ZIP  94  5  6  7  Average Change in HHI from First Year -.1 -.05 0 .05  .1  Figure 129: Balanced Panel: Effect on concentration when Cemex enters a local market  -.15  With Cemex Entering Without Cemex Entering -3  -2  -1  0 1 2 3 4 Years Since Cemex Opening in ZIP  5  6  7  Average Number of Establishments in ZIP 2.5 3 3.5  4  Figure 130: Balanced Panel: Effect on number of establishments when Cemex enters a local market  2  With Cemex Entering Without Cemex Entering -3  -2  -1  0 1 2 3 4 Years Since Cemex Opening in ZIP  95  5  6  7  B.5  Other Results  The results in this section expand on select figures in the main text. B.5.1  Effect of Top Enterprises on Number of Establishments  Figure (131) expands on Figures (15) and (16) in the main text by looking at the number of establishments in industry-ZIP code pairs over time in response to the arrival of an industry’s top enterprise into that ZIP code. The red line displays the weighted average number of establishments in the years before and after an opening of an industry’s top enterprise across industries with diverging trends; the blue lines display the same number average across industries with positive trends at the national and local levels. When a top enterprise opens in an industry with positive local trends, there is on average no exit of existing establishments, while there is close to one-to-one exit of existing establishments in industries with diverging trends. Over time, however, for both sets of industries the number of establishments both including and excluding establishments belonging to the top enterprise increases following an opening. Because these lines are weighted by employment in a geography-industry-year grouping, which is highly correlated with the number of establishments, the results of this figure should be interpreted with caution.  3  Average Number of Establishments in ZIP 5 7 9 11 13  15  Figure 131: Number of Establishments When Top Enterprise Enters  Including Top Enterprise Excluding Top Enterprise SIC 8s with Positive ZIP Trend SIC 8s with Negative ZIP Trend -3  B.5.2  -2  -1 0 1 2 3 4 5 Years Since SIC 8's Top Enterprise Opening in ZIP Code  6  7  Replicating Figures (10) and (11) with Top 3 Enterprises  Here, we replicate Figures (10) and (11) in the main text using the top 3 enterprises (as measured by sales in 2014) in each industry as opposed to just the top enterprise. That is, we look at geography-industry 96  pairs where at least one of these enterprises is present in at least one year. Within this subset of pairs, we drop geography-industry-year groupings where there are no enterprises in that group remaining after dropping the top three enterprises in that industry. We then calculate, for each grouping, the HHI both including and excluding the top 3 enterprises. Figure (132) shows that when averaged across SIC 8 industries with diverging trends, removing the top 3 enterprises makes the increase in the national trend much less pronounced, but increases concentration at the local level. In contrast, Figure (133) shows that across industries with increasing trends at the national and local levels, excluding the top 3 enterprises brings down concentration at both levels. These observations are consistent with Figures (10) and (11) in the main text.  -.2  Average Change in HHI from First Year -.15 -.1 -.05 0 .05  .1  Figure 132: The role of the top three enterprises in national and local concentration trends in diverging industries  Including Top 3 Enterprises Excluding Top 3 Enterprises ZIP Level National Level 1990  1995  2000  Year  97  2005  2010  2014  -.1  Average Change in HHI from First Year -.05 0 .05 .1  .15  Figure 133: The role of the top three enterprises in national and local concentration trends in concentrating industries  Including Top 3 Enterprises Excluding Top 3 Enterprises ZIP Level National Level 1990  C  1995  2000  Year  2005  2010  2014  Geography and Industrial Classification  In this section, we explore how the overall findings in the main text respond to changing, in order, the level of geography at which we define a market, and the level of industrial classification at which we define a market. We only vary the levels of industrial and geographic aggregations separately. That is, whenever possible we try to keep either our level of industrial aggregation at the SIC 8 level, or the level of geographic aggregation at the ZIP code level. The results show that the broad thrusts of our findings still hold, though to slightly lesser extents than when defining an industry-geography pair as a grouping of an SIC 8 code and a ZIP code. Hence, these results further highlight the importance of defining markets locally.  C.1  Results for Other Geographic Measures  As shown in Figure (3) in the main text, while the decline in concentration is most pronounced at the ZIP code level, concentration is also declining over time at the County and Core-Based Statistical Area (CBSA) levels.38 This section shows that diverging trends are still prevalent at these two geographic levels. 38 A CBSA is defined as either a Metropolitan or Micropolitan Statistical Area and is a collection of counties. Although CBSA boundaries can change over time as new counties are added to or removed from them, for all years here we classify counties into CBSAs based on 2014 CBSA definitions. When reproducing these figures at the CBSA level, we drop any observations located in counties outside of CBSAs.  98  National  CBSA Manufacturing Wholesale Trade Retail Trade FIRE Services  -.25  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0  .05  Figure 134: CBSA Level: Diverging division-level national and local concentration trends  1990  1995  2000  Year  2005  2010  2014  National  CBSA SIC 2 SIC 4 SIC 6 SIC 8  -.06  Average Change in HHI from First Year -.04 -.02 0 .02  .04  Figure 135: CBSA Level: Diverging economy-wide trends in sales concentration  1990  1995  2000  Year  99  2005  2010  2015  National  County Manufacturing Wholesale Trade Retail Trade FIRE Services  -.25  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0  .05  Figure 136: County Level: Diverging division-level national and local concentration trends  1990  1995  2000  Year  2005  2010  2014  National  County SIC 2 SIC 4 SIC 6 SIC 8  -.1  Average Change in HHI from First Year -.05 0  .05  Figure 137: County Level: Diverging economy-wide trends in sales concentration  1990  1995  2000  Year  100  2005  2010  2015  C.2  SIC 4 Level Results  The two figures below show that diverging trends are still prevalent at the SIC 4 level.  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0  .05  Figure 138: SIC 4 Level: Diverging economy-wide national and local concentration trends  -.25  National ZIP 1990  1995  2000  Year  2005  2010  2014  National  ZIP Manufacturing Wholesale Trade Retail Trade FIRE Services  -.25  Average Change in HHI from First Year -.2 -.15 -.1 -.05 0  .05  Figure 139: SIC 4 Level: Diverging division-level national and local concentration trends  1990  1995  2000  Year  101  2005  2010  2014