DEWEY MIT LIBRARIES 3 9080 0331 6 3392 ho- 07-31 3.0(0 Technology Department of Economics Working Paper Series Massachusetts Institute of AGGLOMERATION SPILLOVERS: EVIDENCE FROM MILLION DOLLAR PLANTS IDENTIFYING Michael Greenstone Richard Hornbeck Enrico Moretti Working Paper 07-31 December 19, 2007 Revised: April 26, 2010 Room E52-251 50 Memorial Drive Cambridge, MA 021 42 This paper can be downloaded without charge from the Network Paper Collection http://ssm.com/abstract=1078027 Social Science Research at Identifying Agglomeration Spillovers: Evidence from Winners and Losers of Large Plant Openings Michael Greenstone Massachusetts Institute of Technology Richard Hornbeck Harvard University Enrico Moretti University of California, Berkeley April 2010 An earlier version of this paper was distributed under the title "Identifying Agglomeration Spillovers: Evidence from Million Dollar Plants." We thank Daron Acemoglu, Jim Davis, Ed Glaeser, Vernon Henderson, William Kerr, Chad Syverson, seminar participants at several and comments. Steve Levitt and two anonymous Jeffrey Kling, Jonathan Levin, Stuart Rosenthal, Christopher Rohlfs, conferences and many universities for helpful conversations referees provided especially insightful comments. Elizabeth Greenwood provided valuable research assistance. Any opinions and conclusions expressed herein are those of the authors and do not necessarily represent the views of the U.S. Census Bureau. All results have been reviewed to ensure that no confidential information for this research at the Boston RDC from NSF (ITR-0427889) is also gratefully acknowledged. is disclosed. Support Abstract We quantify agglomeration spillovers by estimating the impact of the opening of a large in the same compare incumbent plants in the county where the new plant ultimately chose to locate (the "winning county"), with incumbent plants in the runner-up county (the "losing county"). Incumbent plants in winning and losing counties have similar trends in TFP in the seven years before the new plant opening. Five years after the new plant opening, TFP of incumbent plants in winning counties is 12% higher than TFP of incumbent plants in losing counties. Consistent with some theories of agglomeration economies, this effect is larger for incumbent plants that share similar labor and technology manufacturing plant on the county. We use pools with the total factor productivity new on profits is model, we find evidence of a winning counties. This indicates that the ultimate plant. Consistent with a spatial equilibrium relative increase in skill-adjusted labor costs in effect (TFP) of incumbent plants the location rankings of profit-maximizing firms to smaller than the direct increase in productivity. Digitized by the Internet Archive in 2011 with funding from Boston Library Consortium Member Libraries http://www.archive.org/details/identifyingaggloOOgree Introduction most countries, economic In concentration activity is spatially concentrated. explained by the presence of natural is While some of this advantages that constrain specific productions to specific locations, Ellison and Glaeser (1999) and others argue that natural advantages alone cannot account for the observed degree of agglomeration. Spatial concentration is particularly remarkable for industries that produce nationally traded goods, because the areas where economic activity is concentrated are typically characterized by high costs of labor and land. Since at least Marshall (1890), economists economic activity may have speculated that this concentration be explained by cost or productivity advantages enjoyed by firms of when they locate near other firms. The potential sources of agglomeration advantages include: cheaper and faster supply of intermediate goods and services; proximity to workers or consumers; better quality of worker-firm matches in thicker labor markets; lower risk of unemployment for workers and lower risk of unfilled vacancies for firms following idiosyncratic shocks; and knowledge The tantalizing, that 1 spillovers. possibility because of documenting productivity advantages through agglomeration could provide insights into a series of important questions. it produce nationally traded goods willing London that are characterized to locate in cities like by extraordinary production costs? Why and what explains their historical development? New Why is are firms York, San Francisco, or In general, why do do income differences cities exist persist across regions and countries? Beside an obvious interest for urban and growth economists, the existence of agglomeration spillovers has tremendous practical relevance. Increasingly, local governments compete by offering The main economic substantial subsidies to industrial plants to locate within their jurisdictions. rationale for these incentives depends on whether the attraction plants generates agglomeration externalities. In the absence of positive externalities, to justify the use it is of new difficult of taxpayer money for subsidies based on economic efficiency grounds. The optimal magnitude of incentives depends on the magnitude of agglomerations spillovers, if they 2 exist. The existence and exact magnitude of agglomeration 1 spillovers are considered open See Duranton and Puga (2004), Glaeser and Gottlieb (2009), and Moretti (forthcoming) for recent surveys. e discuss in more detail the policy implications of local subsidies in Greenstone and Moretti (2004). See also W Card, Hallock, and Moretti (2007), Glaeser (2001), and Glaeser and Gottlieb (2008). questions by many, despite their enormous theoretical and practical relevance. three objectives. First, estimating county. how We we test for and quantify agglomeration spillovers the productivity of incumbent plants changes when from the Annual Survey of Manufactures. Second, including input and output flows, we measure technological linkages. Third, by the spillover are reflected Because the new opening and identification of af new county that shed light on the possible plant. We economic consider different measures of measures of labor flows between firms, and which the productivity gains generated the extent to plant's location decision is in future periods. plant, using plant- higher local factor prices. in likely to differ substantially we new the magnitude of the spillovers depends on linkages between the incumbent plant and the is manufacturing by estimate augmented Cobb-Douglas production functions that allow the total factor mechanisms by investigating whether linkages, This paper has a Jarge plant opens in their productivity (TFP) of incumbent plants to depend on the presence of the level data in 3 made to maximize profits, the from an average or randomly chosen county, both chosen county at the time of Valid estimates of the plant opening's spillover effect require the is identical to the county where the plant decided to locate in the determinants of incumbent plants' TFP. These determinants are likely to include factors that affect the new infrastructure, plant's TFP and that are difficult to measure, such as local transportation current and future costs of factors of production, quality of the workforce, presence of intermediate input suppliers, and any other local cost This paper's solution is shifter. to rely on the reported location rankings of profit-maximizing firms to identify a valid counterfactual for what would have happened to incumbent plants' in the absence of the plant opening. These rankings come from the corporate Site Selection, which includes a regular a large plant decided where to locate. real estate journal feature titled "Million Dollar Plants" that describes When TFP how firms are considering where to open a large plant, they typically begin by considering dozens of possible locations. They subsequently narrow the 3 To two primary approaches in testing for spillovers. The first tests for an unequal geographic These "dartboard" style tests reveal that firms are spread unevenly and that co-agglomeration rates are higher between industries that are economically similar (Ellison, Glaeser, and Kerr 2009). This approach is based on equilibrium location decisions and does not provide a direct measure of spillovers. The second approach uses micro data to assess whether firms' total factor productivity (TFP) is higher when similar firms are located nearby (see, for example, Henderson 2003). The challenge for both approaches is that firms base their location decisions on where their profits will be highest, and this could be due to spillovers, natural advantages, or other cost shifters. A causal estimate of the magnitude of spillovers requires a solution to this problem of identification. date, there are distribution of firms. list to roughly 10 among which sites, 2 or 3 finalists are selected. county that the plant ultimately chose articles report the two runner-up counties (i.e., the "losers"). The (i.e., The "Million Dollar Plants" the "winner"), as well as the losers are counties that one or have survived a long selection process, but narrowly lost the competition. The identifying assumption is incumbent plants that the in the losing counties form a valid counterfactual for the incumbents in the winning counties, after conditioning on differences in pre-existing trends, plant fixed effects, variables. Compared to the rest industry-by-year fixed effects, and other control of the country, winning counties have higher rates of growth income, population, and labor force participation. But compared to losing counties before the opening of the variables. This finding is new plant, winning counties have similar trends in in in the years most economic consistent with both our presumption that the average county is not a credible counterfactual and our identifying assumption that the losers form a valid counterfactual for the winners. We first measure the effect of the new Million Dollar Plant (MDP) on total factor productivity of all incumbent manufacturing plants in winning counties. In the 7 years before the MDP opened, we find statistically equivalent trends in TFP for incumbent plants in winning and losing counties. This finding supports the validity of the identifying assumption. After the MDP relative increase in opened, incumbent plants in winning counties experienced a sharp TFP. Five years later, the MDP increase in incumbent plants' TFP. This effect substantial: on average, incumbent plants' output is in opening is associated with a statistically significant winning counties is is equivalent to moving a county from increase in the distribution of county TFP. We and economically A 12% it is five increase in TFP 10th percentile of the county-level the distribution to the 27th percentile; alternatively, relative $430 million higher years later (relative to incumbents in losing counties), holding constant inputs. TFP 12% equivalent to a 0.6 standard deviation interpret this finding as evidence of large productivity spillovers generated by increased agglomeration. Notably, the estimated productivity gains experienced by incumbent plants counties are highly heterogeneous. The average county-level instances, small in some increase other cases, and even negative for a non-negligible Having found evidence to the question TFP in favor is very large We winning in some number of counties. of the existence of agglomeration spillovers, of what might explain these spillovers. in we then turn follow Moretti (2004b) and Ellison, Glaeser, and Kerr (2009) and investigate how the magnitude of the spillovers depends on MDP. measures of economic proximity between the incumbent plant and the test whether incumbents that are geographically and economically linked to the we Specifically, MDP experience larger spillovers, relative to incumbents that are geographically close but economically distant MDP. We from the use several measures of economic links including input and output flows, measures of the degree of sharing of labor pools, and measures of technological linkages. We find that spillovers are larger for MDP flows with the transition is industry. A incumbent plants 4 worker in industries that share one standard deviation increase in our measure of worker associated with a 7 percentage point increase in the magnitude of the spillover. Similarly, the measures of technological linkages indicate statistically meaningful increases in the spillover effect. Surprisingly, flows we find little support for the importance of input and output determining the magnitude of the spillover. Overall, this evidence provides support for in the notion that spillovers occur between firms that share workers and use similar technologies. To we interpret the results, set out a straightforward Roback (1982) style model that incorporates spillbvers between producers and derives an equilibrium allocation of firms and workers across locations. In the model, the entry of a new firm produces spillovers. This leads to entry of firms that are interested in gaining access to the spillover. and subsequent new entry leads and other local for labor, land, cannot raise output prices obtained when in to competition for inputs, so inputs. In the activity in the sloping the model, firms produce nationally traded goods and is is equal to the increased costs prices. Consistent with these predictions, economic opening response to higher input prices. Thus, the long-run equilibrium of production due to higher input MDP original plant incumbent firms face higher prices the value of the increase in output due to spillovers following The we find increases in quality-adjusted labor costs openings. These higher wages are consistent with the documented increase in (at least in the model predicts winning counties and with a local labor supply curve medium run). We that is upward also find positive net entry in winning counties, which will occur if there are sufficiently large positive spillovers to generate an overall increase in profitability. The findings We in this are deeply indebted to paper are related to two earlier studies that use a similar approach to Glenn Ellison, these measures of economic distance. Edward Glaeser, and William Kerr for providing their data for five of productivity identify spillovers the at local Henderson level: who documents (2003), agglomeration spillovers for the machinery and high-tech industries, and Moretti (2004b), estimates productivity spillovers generated by increased concentration of human who capital in a location. Consistent with the findings in this paper, the findings in Moretti (2004b) point to the economically non-trivial, of productivity spillovers that are existence depending on economic distance, and are largely offset by increased labor Our have two findings of important sets implications for local economic development policies. spillovers welfare is crucial to understanding the consequences. intervention may In a implications. be efficient from the point of view of a economic development discuss our findings have The magnitude and form of agglomeration economic rationale We significantly costs. First, for location-based policies world with significant agglomeration point of view of aggregate welfare. vary how spillovers, and their government although not always from the locality, our results inform the debate on local policies. Second, our findings have implications for understanding industrial clusters. Urban economists have long noted that economic activity industrial concentration appears to economic activity in we find is it county-wide, while the productivity spillovers decline economically further away that are profitable. In the long-run, this process may result MDP location. The industry. This appears to be fairly stable over time. Because the economic distance, incumbent firms each by spatially concentrated be a pervasive feature of the geographical distribution of most countries and increase in labor costs that is interaction in increased may become in less agglomeration of similar plants in between spillovers and input costs may therefore help explain the existence and persistence of industrial clusters. The remainder of Section II the paper is organized as follows. Section discusses the identification strategy. Section presents the econometric model. Sections interprets the results I. We V III I presents a simple model. introduces the data sources. Section IV and VI describe the empirical results. Section VII and discusses implications for policy. Section VIII concludes. Theories of Agglomeration and Theoretical are interested in identifying how the opening of a productivity, profits, and input use of existing plants in the Framework new plant in a county affects the same county. We begin by briefly reviewing theories of agglomeration. 5 We then present a simple theoretical framework that guides the subsequent empirical exercise and aids in interpreting the results. * A. Theories of Agglomeration Economic activity is geographically concentrated (Ellison the forces that can explain such agglomeration of economic and Glaeser 1997). What are Here we summarize activity? five possible reasons for agglomeration, and briefly discuss what each of them implies for the relationship between productivity and the density of economic activity. (1) First, it is possible that firms (and workers) are attracted to areas with a high concentration of other firms (and other workers) by the size of the labor market There are . least two different reasons more productive 6 larger labor markets may be attractive. First, if there are search and jobs and workers are heterogeneous, then a worker-firm match will be on average frictions jobs. why at in areas where there Second, large labor markets the firm side or unpredictable On the worker demand shocks are may side many firms offering jobs and looking for provide insurance against idiosyncratic shocks, either on (Krugman 1991a). that lead to layoffs If firms experience idiosyncratic and moving/hiring then thicker labor markets will reduce the probability that a worker unfilled vacancies. many workers is is and costly for workers/firms, unemployed and a firm has 7 These two hypotheses have different implications for the relationship between the concentration of economic activity and productivity. If the size of the labor market leads only to better worker-firm matches, we should see that firms located in denser areas are more productive than otherwise identical firms located in less dense areas. The exact form of this productivity gain depends on the shape of the production function. 5 8 See Moretti, forthcoming, and Glaeser and Gottlieb, 2009, for comprehensive surveys of this For A a related point in a different context, see third alternative hypothesis has to example, in Acemoglu do with spillovers (1996), plants have literature. Petrongolo and Pissarides (2005). more capital that arise because of endogenous capital accumulation. For and better technology in areas where the number of skilled and workers find each other via random matching and breaking the match is costly, even without learning or technological externalities. The intuition is simple. The privately optimal amount of skills depends on the amount of physical capital a worker expects to use. The privately optimal amount of physical capital depends on the number of skilled workers. If the number of skilled workers in a city increases, firms in that city, expecting to employ these workers, will invest more. Because search is costly, some of the workers end up working with more physical capital and earn more than similar workers in other cities. For example, it is possible that the productivities of both capital and labor benefit from the improved match in denser areas. It is also possible that the improved match caused by a larger labor market benefits only labor workers is larger. If firms externalities will arise naturally On the other hand, if the only effect of thickness in the labor market unemployment for a lower risk of is workers and a lower risk of unfilled vacancies for firms, there should not be improved differences in productivity between dense and less dense areas. Unlike the case of matching described above, the production function does not change: for the same capital inputs, the output of firms in set of labor and denser areas should be similar to the output of firms in less dense areas. While productivity would not vary, wages would vary across areas depending on the thickness of the labor market, although the exact effect of density on ambiguous. 9 wages is a priori This change in relative factor prices will change the relative use of labor and capital. (2) A second reason why the concentration of economic activity may be beneficial has to do with transportation costs (Krugman 1991a and 1991b; Glaeser and Kohlhase 2003). Because in this paper we focus on firms that produce nationally traded goods, transportation costs of Only a small finished products are unlikely to be the relevant cost in this paper's setting. of buyers of the The relevant final costs intermediate goods . product are the is likely to be located transportation Firms located in costs in the same area fraction as our manufacturing plants. of suppliers of local services and local denser areas are likely to enjoy cheaper and faster delivery of local services and local intermediate goods. For example, a high-tech firm that needs a specialized technician to fix a is machine located in Silicon Valley than in the is likely to get service Nevada more quickly and at lower cost if it desert. This type of agglomeration spillover does not imply that the production function varies as a function of the density of economic activity: for the same set of labor and capital inputs, the output of firms in denser areas should be similar to the output of firms in less dense areas. However, production costs should be lower (3) A third reason why the with knowledge spillovers . in denser areas. concentration of economic activity There are at least two may be different versions of beneficial has to do this hypothesis. First, economists and urban planners have long speculated that the sharing of knowledge and through formal and informal interaction may skills generate positive production externalities across productivity. This has different implications for the relative use of labor and capital, but total factor productivity will be higher regardless. 9 Its sign depends on the relative magnitude of the compensating differential that workers are willing to pay for lower risk of unemployment (generated by an increase in labor supply in denser areas) and the cost savings that firms experience due to lower risk of unfilled vacancies (generated by an increase in labor demand in denser areas). workers. 10 more high-tech industries. For example, patent citations are metropolitan area as the originating patent (Jaffe et geographic proximity of high-tech firms flow of may Empirical evidence indicates that this type of spillover new likely to come from is associated with a ideas and ultimately causes faster innovation." Second, proximity results sharing of information on in new the in same some state or 1993). Saxenian (1994) argues that al. Valley in Silicon be important it is more efficient also possible that technologies and therefore leads to faster technology adoption. This type of social learning phenomenon applied to technology adoption was first proposed by Griliches (1958). If of economic activity density results we agglomeration would lead to higher productivity. In particular, in externalities, intellectual in this form of should see that firms located denser areas are more productive than otherwise identical firms located in less dense areas. As with the search model, this higher productivity could benefit both labor and capital, or only one of the two depending on the form of the production function. factors, On the other hand, if density of economic activity only results in faster technology adoption and the price of technologies refrects their higher productivity, should there new no relationship between be productivity and density, after properly controlling for the quality of capital. (4) is It spillover, but possible that firms concentrate spatially not because of any technological because local amenities valued by workers are concentrated. For example, skilled workers may employ relatively prefer certain amenities more skilled more than unskilled workers. This would workers to concentrate in locations where these amenities are available. In this case, there should not be differences in productivity less dense areas, although there would be differences in wages lead firms that between dense areas and that reflect the compensating differential (5) Finally, spatial concentration of some industries natural advantages or productive amenities limited number of wine industry is states . For example, the concentrated See, for example, Marshall be explained by the presence of oil industry because those states have the most accessible in California due to manufacturing productions, the presence of a harbor 10 may (1 suitable may be 890), Lucas (1988), Jovanovic and Rob is concentrated in a oil fields. Similarly, the weather and land. For some important. Natural advantages imply (1989), Grossman and Helpman (1991), Saxenian (1 994), Glaeser (1 999), and Moretti (2004a, 2004b and 2004c). " The entry decisions of new biotechnology firms in a city depend on the stock of outstanding scientists there, as measured by the number of relevant academic publications (Zucker et al. 1998). Moretti (2004b) finds stronger human capital spillovers between pairs of firms in the same city that are economically or technologically closer. that firms located in areas with a high concentration of similar firms are course this correlation is unrelated to agglomeration spillovers. Since most natural advantages are fixed over time, this explanation which exploit variation over time B. in is not particularly relevant for our empirical estimates, agglomeration. A Simple Model We begin by considering the case where incumbent firms are homogenous technology. Later we consider what happens when incumbent firms Throughout the paper, we focus on the case of factor-neutral Homogeneous Incumbents. (i) We assume capital, fixed and is 1. land, T, to maximize the following expression: normalized to r, and q are input prices and A factors that affect the productivity agglomeration spillovers, we if and heterogeneous. are spillovers. that all Incumbent firms choose Maxijc,T f(A, where w, in size incumbent firms use a production and land to produce a nationally traded good whose price technology that uses labor, all more productive, but of of is L, K, T) - their amount of labor, L, capital, K, is and wL-rK- qT a productivity shifter (TFP). Specifically, labor, capital, A includes and land equally, such as technology and they exist. In particular, to explicitly allow for agglomeration effects, allow A to depend on the density of economic activity in an area: A = A{N) (1) where N is the number of firms that are active in a county, and define factor-neutral agglomeration spillovers as the case where dA/dN = instead 0, we all A counties have equal size. increases in N: dA/dN > We 0. If say that there are no factor-neutral agglomeration spillovers. Let L*(w,r,q) be the optimal level of labor inputs, given the prevailing wage, cost of and cost of industrial land. Similarly, capital, capital K*(w,r,q) and T*(w,r,q) be the optimal level of and land, respectively. In equilibrium, V, K", and T* are of each of the three factors We demand local let assume is equal to that capital or supply conditions. economic conditions. infinitely elastic at the As in is its county marginal product price. internationally traded, so However, we allow In particular, set so that the we its for the price price does not depend on of labor and land to local depend on allow the supply of labor and land to be less than level. Moretti (forthcoming), we attribute the upward sloping labor supply curve to the We existence of preferences for location. assume that workers' wages, cost of housing and idiosyncratic preferences for location, and that we marginal workers are indifferent across locations. For simplicity, decisions within a given location and assume that To new illustrate this, plant. In particular, consider that there are m such is that, all and staying rising, in the original and some workers find county. it county c amount of labor. before the opening of the given the distribution of wages and the housing costs across localities, the marginal worker in another county c in equilibrium in ignore labor supply residents provide a fixed m workers depends on indirect utility When optimal to new a move is indifferent between moving plant opens in county to county c. c, to county wages there start The number of workers who move, and therefore the slope of the labor supply function, depend on the importance of preferences for location (see Moretti, forthcoming, for details). Let w(7V) be the inverse of the reduced-form labor supply function that links the wage level, firms, TV, active in a county to the local nominal w. Similarly, county number of level. we allow the supply of industrial land to be less than infinitely elastic For example, it is possible that the supply of land land-use regulations. Alternatively, industrial land has already more expensive it may been developed, so that the marginal land that links the number of therefore write the equilibrium level of profits, 77* = fixed because of geography or not be completely fixed, but to develop. Irrespective of the reason, form land supply function is 77*, we call at the it is is possible that the best of decreasing quality or q(N) the inverse of the reduced firms, N, to the price of land, q. We can as /[/l(JV),L'(w(JV),r,gW),)('(w(N),r,(|(W)),r*(w(N),r,(,(W))] - w(N)L'(w(N),r,q(N))where we now make explicit the fact that r/C'(w(yV),r,q(iV)) - q(N)T'(w(N),r,q(N)) TFP, wages, and land prices depend on the number of firms active in a county. Consider the total derivative of incumbents' profits with respect to a change in the number of firms: (2) dn'/dN = (df/dA x dA/dN) + dw/dN {[df/dw (df/dL - w) + dq/dN {[dL'/dq (df/BL - w)] + If all firms are price takers and considerably and can be written all L'] + [dK'/dw (df/dK - [dK*/dq (df/dK - r)] + r)] + [dT'/dw (df/dT - [dT'/dq (df/dT - q) q)]} - 7*]} factors are paid their marginal product, equation (2) simplifies as: 10 dn'/dN = (df/dA x dA/dN) - [dw/dN V + dq/dN T ]. (3) Equation (3) makes clear that the effect of an increase First, if there are positive spillovers, the effect on TFP represented by the is it inputs. Formally, df/dA > sum of two first (df/dA x dA/dN). This term, effect L* + dq/dN T*], dw/dN > that of economic activity is in the Intuitively, > 0, entrant. The and to change it an increase dK*/dN > negative N in is an increase in the local demand in the level for labor and two prices is for locally scarce resources with effects on incumbents. mechanically raises production costs. Second, in N, the firm on the use of labor and land productivity of all factors increases. On it First, for a leads the firm is likely to end up using more is ambiguous. On one hand, the the other hand, the price of labor and land increases. on the magnitude of the (i.e., compete in factor 0. contrast, the effect the production function to prices has not affected by an increase capital than before: capital, is use of the different production inputs. In particular, given that its is net effect depends now have wages and land increase in the price of capital By term illustrated in a similar context in Moretti (2004b). given level of input utilization, to re-optimize 0. while the magnitudes depend on the county and therefore an increase costly for incumbent firms, because they new dA/dN > this Unlike the beneficial effect of agglomeration spillovers, the increase the unambiguously represents the negative effect from increases and dq/dN of the supply of labor and land. land. This point opposite effects. is if there are positive spillovers, of production, specifically the prices of labor and land. Formally, we have assumed elasticity the is productivity of all factors increases. In equation (3), this by assumption and, The second term, -[dw/dN because N allows an incumbent firm to produce more output using the same amount of positive, because in the cost in factor price increases, as well as The on the exact shape of the strength of technological complementarities between labor, and land). It is instructive to apply these derivations to the case positive spillovers. would occur when this case, the We initially MDP opening that causes consider the case where for incumbent firms dFl*/dN the agglomeration spillover MDP's of a is < 0. This smaller than the increase in production costs. In opening would not lead to entry and could cause some existing firms to exit. The alternative case spillover due to the is that dU* /dN > 0, which occurs when the magnitude of the MDP opening exceeds the increase in factor prices due to the MDP's demand 11 be positive for for local inputs. In the short run, profits will of local will disappear over time as the price In the long run, there is new such as land and possibly labor, factors, likely that wages bid up. is will increase. This may occur due to a supply, as noted above. These adjustments amount of land locales. Since the of productivity are likely to be capitalized into land fixed, the higher levels profits an equilibrium such that firms and workers are indifferent between the county where the new plant has opened and other is These positive entrants. prices. less than infinite elasticity make marginal workers also It is of local labor between the indifferent county with the new plant and other counties. Similarly, the changes in factor prices mean that same firms earn the spillovers) context to and profits in the in other locations. know when new county with the From plant (even in the presence a practical perspective, it is of the not possible in our empirical the short run ends and the long run begins. There are two empirical predictions that apply when there are positive spillovers. the magnitude of the spillovers is new large enough, MDP's firms will enter the First, if county to gain access to the spillover. This prediction of increased economic activity holds at any point after new potential entrants have had sufficient time to respond. prices of locally traded inputs will rise as the (ii) Assume 12 some some extent from the type of workers overlap. Assume if that the new that the is for these inputs. the population of incumbent firms where there are two types of that for technological reasons, the type differs to is the case prediction MDP and the new entrants bid Heterogeneous Incumbents. What happens homogeneous? Consider The second is 12 non- firms: high-tech and low-tech. of workers employed by high-tech firms, L H employed by low-tech entrant is firms, L L , , although there a high-tech firm. Equations (4) and (5) This model focuses on the case where the productivity benefits of the agglomeration spillovers are distributed equally across all factors. What happens when agglomeration Assume, spillovers are factor biased? example, that for agglomeration spillovers raise the productivity of labor, but not the productivity of capital. As before, the technology is f(A,L,K,T), but now L represents physical workers and 9 We 0. If dA/dN = of the economic , where H is the economic activity in the county 6 and factors are paid their marginal product, then the effect of an increase activity in county on incumbent firms simplifies a [dw/dN H* + dq/dNT']. The increased productivity of labor. effect It is on number of define factor-biased agglomeration spillover as the case where the productivity shifter 8 depends positively on the density of the 68/ dN > = 6H units of effective labor. In particular, L a productivity shifter. is profits can be decomposed in to two — 0(N) and in the density dU'/dN = (df/dH x dd/dN)H — parts. The the product of the sensitivity of output to labor first term represents the (df/dH > 0), times the magnitude of the agglomeration spillover (38 /dN > by definition), times the number of workers. The second term is the same as in equation (3), and represents the increase in the costs of locally supplied inputs. The increase in N changes the optimal use of the production inputs. Labor is now more productive, and its equilibrium use increases: dT'/dN < dL'/dN < 0. 0. Land is equally productive but its Neither the price nor the productivity of capital depends on technology; specifically, it depends on the elasticity 12 price is increases, so its equilibrium affected by an increase in N. of substitution between labor and Its use declines: equilibrium use capital. . characterize the effect of the new high-tech firm on high-tech and low-tech incumbents: (4) dn-H /dNH = (dfH /dA H x 3A H /dNH ) - [dw H /3NH VH + dq/8NH (5) dnt/dNH = (dfL /dA L x 8A L /dNH ) - [dwL /dNH + dq/dNH T'}. plausible to expect that the beneficial effect of agglomeration spillovers generated It is new l\ T*] high-tech entrant is by a larger for high-tech firms than for low-tech firms: CdfH /dA H x dA H /dNH ) > {dfL /dA L x 3A L /dNH ). (5') At the same time, one might expect that the increase in labor costs is also higher for the high- now competing tech incumbents, given that they are for workers with an additional high-tech firm: dwH /dNH > dwL /dNH (5") The effect . on land prices should be similar for both firm types, since the assumption of a single land market seems reasonable. This model of heterogeneous incumbents has two main implications. First, it reasonable to expect larger spillovers on firms that are economically "closer" to the Second, the relative impact of the new plant on profits is may new be plant. unclear, because the economically "closer" plants are likely to have both larger spillovers and larger increases in production costs. C. Empirical Predictions The simple theoretical framework above generates four predictions that we bring to the data. Specifically if there are positive spillovers, then: 1 the opening of a 2. the increase in new plant will increase TFP may TFP of incumbent plants; be larger for firms that are economically "closer" to the new plant; 3. the density of economic activity in the county will increase as firms move in to gain access to the positive spillovers (if the spillovers are large enough); and 4. the price of locally supplied factors of production will increase. in the price of quality-adjusted labor, factor for manufacturing plants. 13 which is We test for changes arguably the most important local II. Plant Location Decisions and Research Design main econometric challenge In testing the four empirical predictions outlined above, the is do not choose that firms where their expectation their location randomly. Firms maximize profits and choose to locate of the present discounted value of future profits is many present value varies tremendously across locations depending on This net greatest. including: factors, transportation infrastructure, the availability of workers with particular skills, subsidies, etc. These factors are frequently unobserved and, problematically, they are with the TFP of existing plants. TFP of incumbents Therefore, a naive comparison of the plant opening with the TFP of incumbents likely to yield biased estimates plant opening on the location TFP effect 13 intent of incumbent plants require the identification of a location that is to locate in the determinants how BMW may circumvent these similar to is of incumbent plants' TFP. picked the location for one of to demonstrate the empirical difficulties that arise of plant openings on the TFP of incumbent our research design plants. Further, when illustrates it BMW announced in months counties. Six informally BMW announced Greenville-Spartanburg, South Carolina, and they would site that the decision. did The list two of potential candidates finalists Omaha, Nebraska. BMW first was by the choose Greenville-Spartanburg? BMW's in the In 1992, to technology. state and Two factors expected future costs of production According to BMW, Spartanburg more attractive than the other 250 the local in characteristics sites initially a supply of qualified workers; numerous global firms 13 This plant we is in 14 20 U.S. BMW announced that governments. were important in this Greenville-Spartanburg, that made BMW's Greenville- considered were: low union density; in the Greenstone and Moretti's (2004) set of 82 MDP plants. Due this plant is part of this paper's analysis. cannot report whether new competition were which are presumably a function of the county's expected supply of inputs and production its would receive a package of the plant in Greenville-Spartanburg and that they incentives worth approximately $115 million funded Why how difficulties. 1991 that they had narrowed the later, its estimating the After overseeing a worldwide competition and considering 250 potential sites for plant, is of productivity spillovers. Credible estimates of the impact of a where the plant decided The counties that experience a in do not experience a plant opening in counties that This section provides a case study for plants. be correlated likely to to area, including 58 Census confidentiality German restrictions, companies; high quality transportation infrastructure, including air, rail, highway, and port access; and access to key local services. For our purposes, the important point to note here While these potential source of unobserved heterogeneity. the is that these county characteristics are a characteristics are well BMW case, they are generally unknown and unobserved. the growth of TFP of existing plants, a standard regression that documented in If these characteristics also affect compares Greenville-Spartanburg with the other 3,000 United States counties will yield biased estimates of the effect of the plant A opening. standard regression will overestimate the effect of plant openings on outcomes example, counties that have more attractive characteristics infrastructure) tend to have faster underestimate the effect if, TFP for improving transportation (e.g., growth. Conversely, a standard regression would example, incumbent plants' declining for if, TFP encourages new entrants (e.g., cheaper availability of local inputs). A second important factor in BMW's decision was the value of the subsidy Presumably Greenville-Spartanburg was willing to provide because it facility's this $2 expected economic benefits from BMW's was expected indirectly. In principle, these presence. According to local officials, the to create 2,000 jobs directly As 2,000 additional jobs could reflect the entry of be a function of the expected gains from agglomeration for the county. is a part of and another 2,000 jobs new plants or the expansion of existing plants caused by agglomeration economies. Thus, the subsidy This possibility received. BMW with $115 million in subsidies ex ante expected five-year economic impact on the region was $2 billion. billion, the plant it is likely to 14 relevant for this paper's identification strategy, because the magnitude of the spillover from a particular plant depends on the level and growth of a county's industrial structure, labor force, determine the represent a heterogeneity and a series of other unobserved variables. For total size second is this reason, the factors that of the potential spillover (and presumably the size of the subsidy) potential source of unobserved heterogeneity. If this unobserved correlated with incumbent plants' TFP, standard regression equations will be misspecified due to omitted variables, just as described above. In order to 14 The valid inferences in the presence of the heterogeneity associated with the fact that business organizations the case with model make BMW) such as the Chambers of Commerce support these incentive plans (which was suggests that incumbent firms expect such increases. Greenstone and Moretti (2004) present a that describes the factors that determine local governments' bids for these plants and whether successfully attracting a plant will be welfare-increasing or welfare-decreasing for the county. 15 plant's expected local production costs and the county's value of attracting the plant, knowledge of the exact form of the selection rule necessary. As BMW the decisions are generally difficult to that determines plants' location decisions example demonstrates, the two unknown where they are known, are to researchers and, in the rare cases factors that determine the plants' in generally factors that determine plant location measure. Thus, the effect of a plant opening on incumbents' confounded by differences is TFP very likely to be is profitability at the chosen location. As a solution to this identification problem, we rely on the reported location rankings of profit-maximizing firms to identify a valid counterfactual for what would have happened to incumbent plants in winning counties in the absence of the plant opening. We implement the research design using data from the corporate real estate journal Site Selection. Each issue of this journal includes an article titled "Million Dollar Plants" that describes where to locate. These that in many the "losers"). (i.e., winner and losers are usually chosen from an cases a large plant decided always report the county that the plant chose articles and usually report the runner-up county or counties indicates, the how number more than one hundred. The initial 15 As (i.e., the the "winner"), BMW case study sample of "semi-finalist" losers are counties that sites have survived a long selection process, but narrowly lost the competition. We use the losers to identify what would have happened to the productivity of incumbent plants in the winning county incumbent firms' in the absence of the plant opening. Specifically, TFP would have trended identically of winning and losing counties belonging so our identifying assumption is to the approach incumbent plants in is more the absence of the plant opening in pairs case. In practice, we if this assumption adjust for covariates fails to hold, reliable than using regression adjustment to counties with that weaker. The subsequent analysis provides evidence that supports the validity of this assumption. Even this pair-wise same in we assume new we presume that compare the TFP of plants to the other 3,000 United States counties, or to using a matching procedure based on observable variables. In some were able is we did the original 82, we instances the "Million Dollar Plants" articles do not identify the runner-up county. For these cases, a Lexis/Nexis search for other articles discussing the plant opening and to identify the losing counties. Comprehensive data on unavailable in the Site Selection articles. 16 in four cases, the subsidy offered among by winning and losing counties Data Sources and Summary III. Statistics Data Sources A. The "Million Dollar Plants" county where the articles typically reveal the new firm (the "Million Dollar Plant") ultimately chooses to locate (the "winning county") and one or two runner-up counties (the "losing counties"). The articles tend to focus on large manufacturing plants that are the target of local is An government subsidies. important limitation of these articles magnitude of subsidy offered by winning counties that the offered by losing counties losing county, there We is is often unobserved and the subsidy is almost always unobserved. In addition, when there no indication of the plants' relative preferences among is more than one the losing counties. identify the Million Dollar Plants in the Standard Statistical Establishment List (SSEL), which is the Census Bureau's "most complete, business establishment," 16 and matched the plants to the and consistent data for U.S. current, Annual Survey of Manufactures (ASM) 17 MDP openings in all industries used in Greenstone and Moretti (2004), we identified 47 useable MDP openings in the manufacturing data. In order to qualify as a useable MDP manufacturing opening, we imposed and the Census of Manufactures (CM) from 1973-1998. the following criterion: 1) there had to be a reported firm, appearing in the the the MDP MDP SSEL new plant in the manufacturing sector, SSEL had owned by the to be located in the county indicated in had to be incumbent plants present for each of the previous 8 years. identified 10 openings in the retail the 82 within 2 years before and 3 years after the publication of article; 2) the plant identified in the article; and, 3) there Of Among the 35 in both winning and losing counties MDP we openings that did not qualify, and wholesale trade sectors whose effects we examine in robustness specifications. To the ASM obtain information on incumbent establishments in winner and loser counties, and CM. The ASM and materials, total value of shipments, location are also reported CM we use contain information on employment, capital stocks, and firm identifiers. The and these play a key role manufacturing data contain a unique plant identifier, 4-digit the in making it SIC code and county of analysis. Importantly, the possible to follow individual The SSEL is confidential and was accessed in a Census Data Research Center. The SSEL is updated continuously and incorporates data from all Census Bureau economic and agriculture censuses and current business surveys, quarterly and annual Federal income and payroll tax records, and other Departmental and Federal statistics and administrative records programs. 17 The sample is because of minor cut at 1998 because sampling methods in the known inconsistencies with the 1972 CM. 17 ASM changed for 1999. The sample begins in 1973 plants over time. ASM Our main were continuously present analysis uses a sample of plants that in the the 8 years preceding the year of the plant opening plus the year of the opening. in Additionally, we drop all plants owned by firms that own sampling scheme was positively related to firm and plant MDP. a Any size. In this period, the ASM was establishment that part of a company with manufacturing shipments exceeding $500 million was sampled with certainty, as were establishments with 250 or. more employees. There are a few noteworthy features of this sample of potentially affected plants. First, the focus on existing plants allows for a test of spillovers on a fixed sample of pre-existing plants, which eliminates concerns compositional bias. Second, a disadvantage is it is that the results related to 54% plants and possible to form a genuine panel of manufacturing plants. Third, may not be externally valid to smaller incumbent plants that are not sampled with certainty throughout this period. Nevertheless, plants accounts for new the endogenous opening of relevant that this sample of it is of county-wide manufacturing shipments in the last CM before the MDP opening. In addition to testing for agglomeration effects are larger some measure of economic categories. First, to an average spillover effect, We to the MDP based on focus on six measures of economic distance in three measure supplier and customer linkages, industry's manufactured inputs that also test whether the estimated more closely linked in industries that are distance. we come from each we use data on the fraction of each 3-digit industry and the fraction of each industry's outputs sold to manufacturers that are purchased by each 3-digit industry. Second, to measure the frequency of worker mobility between transitions industries, we use data on labor market from the Current Population Survey (CPS) outgoing rotation measure the fraction of separating workers from each 2-digit industry 2-digit industry. Third, to measure technological proximity, we that file. In particular, move we to firms in each use data on the fraction of patents manufactured in a 3-digit industry that cite patents manufactured in each 3-digit industry. We also use data digit industries. ,b on the amount of R&D expenditure in a 3-digit industry that is used in other 3- 18 have two sources of information on the date of the plant opening. The first is the MDP articles, which often when ground is broken on the plant but at other times are written when the location decision is made or when the plant begins operations. The second source is the SSEL, which in principle reports the plant's first year of operation. However, it is known that plants occasionally enter the SSEL after their opening. Thus, there is We are written uncertainty about the date of the plant's opening. Further, the date at which the plant could affect the operations of existing plants depends on the channel for agglomeration spillovers. 18 If the agglomeration spillovers are a Summary B. Table Statistics 1 summary presents the basis of the analysis. MDP openings that we As statistics on the sample of plant location decisions that forms discussed in the previous subsection, there are 47 manufacturing can match to plant level data. There are plants same in the 2-digit SIC industry in both winning and losing counties in the 8 years preceding the opening for just 16 of these openings. The some table reveals accompanying other facts about the plant openings. 19 We refer to the winner and opening as a "case." There are two or more loser(s) associated with each plant losers in 16 of the cases, so there are a total of 73 losing counties along with 47 winning counties. Some average county the MDP counties appear multiple times in the sample (as winner and/or loser), and the in the sample appears a total of 1.09 times. article publication across the categories -2 to refer to cases where the and the year the plant appears -1 years, years, and in the to 3 years. 1 appears after the plant article The is difference between the year of SSEL For is roughly spread evenly clarity, positive differences identified in the SSEL. The dates of the plant openings range from the early 1980s to the early 1990s. The remainder of Table assigned opening date. These 1 provides MDPs summary statistics on the MDPs, are quite large: they are five years after their more than twice the size of the average incumbent plant and account for roughly nine percent of the average county's total output one year prior to their opening. Table 2 provides summary descriptions of these variables. In statistics all on the measures of industry linkages and further cases, the proximity between industries is increasing in the value of the variable. For ease of interpretation in the subsequent regressions, these variables are normalized to have a mean of zero and a standard deviation of one. Table 3 presents the means of county-level and plant-level variables across counties. consequence of supplier relationships, then they could occur as soon as the plant is announced. For example, the new plant's management might visit existing plants and provide suggestions on operations. Alternatively, the agglomeration spillovers may be driven by the labor market and therefore may depend on sharing labor. In this case, agglomeration spillovers there is possibility, we emphasize year that the matched 19 A may not clear guidance on number of not be evident until the plant when the new is operating. Based on these data and conceptual issues, plant could affect other plants. results using the earliest To be conservative and allow for each of (1) the publication year of the magazine article and (2) the MDP appears in the SSEL. the statistics in Table 1 are reported in broad categories to confidentiality restrictions and to avoid disclosing the identities 19 of any individual comply with plants. the Census Bureau's These means are reported and (3), respectively. among 20 for winners, losers, In the and the entire United States the incumbent plants present in the ASM in the 8 years preceding the assigned opening across years to produce statistics for the year of the average Further, the plant characteristics are only calculated Column 9 consecutive years. and (2) are equal, (3). Columns the same (6) 2-digit calculated among while (1), (2), winner and loser columns, the plant-level variables are calculated date and the assigned opening date. All entries in the entire United States least Columns in Column (5) repeats this for a test MDP. as the the plants in the same among opening in are weighted our sample. plants that appear in the (4) presents the t-statistics through (10) repeat this exercise SIC industry among MDP column from a ASM for at test that the entries in (1) of equality between Columns (1) and the cases where there are plants within In these columns, the plant characteristics are 2-digit industry. This exercise provides an opportunity to assess the validity of the research design, as measured by pre-existing observable county and plant observable characteristics are balanced characteristics. among winning and To losing counties, this should lend The comparison between winner counties and credibility to the, analysis. the extent that these the rest of the United States provides an opportunity to assess the validity of the type of analysis that undertaken in the absence of a quasi-experiment. The top panel reports county-level characteristics measured in the year before the assigned plant opening and the percentage change between 7 years and opening. Compared to the rest Among compared are at the 1% to losers: 3 level. rates and growth, and a higher share of labor in the 8 variables in this panel, 6 of the 8 differences are statistically significant at conventional levels. are year before the 1 of the country, winning counties have higher incomes, population and population growth, labor force participation manufacturing. would be of the These differences are substantially mitigated when the winners 8 variables are statistically different at the 5% level, and none Notably, the raw differences between winners and losers within the subset of cases where there are plants in the same 2-digit SIC industry are generally smaller, and none are statistically significant. 20 The The second panel reports losing county entries in Column the inverse of their number is (2) are in that case. multiplied by the inverse of the plant within each county) on the number of sample plants and provides information on weighted in the following manner. Losing counties are weighted Losing plants are weighted by the inverse of number of losing counties in their case. given equal weight within the case and then 20 all The result is their by number per-county, that each county (and each cases are given equal weight. some of their characteristics. In light On special interest. of our sample selection The plants in winning and losing counties; in fact, there are among plants or all among shows Overall, Table 3 (although not all) number of plants is of average, there are 18.8 plants in the winner counties and 25.6 in the loser counties (and just 8.0 in the average U.S. county). either criteria, the same plants within the MDP that the covariates are well-balanced no between statistically significant differences 2-digit industry. 21 winner-loser research design balances Of observable county-level and plant-level covariates. many course, this exercise does not guarantee that unobserved variables are balanced across winner and loser counties or their plants. In the subsequent analysis, losing counties prior to the we MDP opening, which section outlines our full econometric TFP were find that trends in similar in winning and The next lends further credibility to this design. model and highlights the exact assumptions necessary for consistent estimation. IV. Econometric Building on the model in Section we I, start Model by assuming that incumbent plants use the following Cobb-Douglas technology: w=v4CC* (6) where p references changes in inventories; capital stock M pij t plant, to Kpij t , i industry, A pijt is ;' case, t TFP; and we allow is total labor the total value of shipments minus hours of production L pijt Kpijt , , building and the dollar value of materials have separate impacts on output. In practice, the two capital stock variables are method and deflated values of subsequent investment. Recall that equation (1) in Section Roughly Ypijt machinery and equipment capital stock calculated with the permanent inventory 21 year; and 20% of the winners were defined as MI, IN, OH, PA, NJ, IL, I that uses earlier years of data on book values 22 allows for agglomeration spillovers by assuming that in the Rust Belt, compared to roughly 25% of the losers (where the Rust Belt is WI, NY). Roughly 65% of the winners were in the South, compared to roughly 45% of the losers. For the first date available, plants' historical capital stock book values are deflated to constant dollars using BEA data by 2-digit industry. In all periods, plants' investment is deflated to the same constant dollars using Federal Reserve data by 3-digit industry. Changes in the capital stock are constructed by depreciating the initial deflated capital stock using Federal Reserve depreciation rates and adding deflated investment. In each year, productive capital stock is defined as the average over the beginning and ending values, plus the deflated level of capital rentals. The analysis is performed separately for building capital and machinery capital. This procedure is described further by Becker et al. (2005), Chiang (2004), and Davis, Haltiwanger, and Schuh (1996), from whose files we gratefully obtained deflators. 21 TFP number of a function of the is some also allow for varying shocks to TFP fi it ln The goal do we need so, in TFP and a stochastic error term , 0W ) £ pijt ap , we ). Here we generalize equation (1) by cases X p , industry-specific time- : = a v + ftt + h + £vm + ^C^pyt )• of winning a plant on incumbent plants' TFP. impose some structure on A(Npijt to In particular, . across plants to estimate the causal effect is ). In particular, we To use a specification that plant in winning counties to affect both the level of TFP as well as new allows for the A piJt additional heterogeneity in allowing for permanent differences A pijt = A(Npijt firms that are active in a county: its growth overtime: ln(i4 yj ) p (7) = 81(Winner) pj + \pTrendjt + fXJrend x l(Winner)) pjt > k1(t + ft (1 (Winner) 0) ;t + a p + n it where l(Winner) Pj year, but The it a > 0)) ;t x 1(t > 0)) p;t + 6 2 (Trend x +Ap + e dummy l(W inner) x 1(t > 0)) pjt pijt equal to 1 if plant p is located in a winner county; and r denotes normalized so that for each case the assigned year of the plant opening is variable is 1(t + y (Trend X + TrendJt is is t = 0. a simple time trend. Combining equations and (6) (7) and taking we logs, obtain the regression equation that forms the basis of our empirical analysis: (8) \n(Ypijt ) =A ln(L pyt + ft ln(^;t ) + ) ln(^yt ) + ft ln(Mpiyt ) ft + 61(Winner) pj + xpTrendjt + H(Trend x l(Winner)) pjt > k1(t + ft (1 (dinner) 0) jt x 1(t + a p + n + Ap + it Equation (8) is 1(t > 0)) ;t 0)) pyt + 9 2 (Trend x l{Winner) x 1(t + y(Trend x + > an augmented Cobb-Douglass production function that allows labor, building machinery focus the estimation of the spillover effects of the capital, parameters of interest are and materials 9i and 82. to have The former differential impacts TFP among the same In practice, we new plant tests for a plants in the winning county after the opening of the in 0)) p;t E pijt capital, is > on incumbent mean MDP, on output. The paper's shift in while the plants' TFP, so the TFP among incumbent latter tests for a trend break plants. estimate two variants of equation more parsimonious model that simply tests for a 22 mean (8). In shift. some In this specifications, we fit a model, any productivity effect is assumed to the restrictions that occur immediately and to remain constant over time. Specifically, i/> = /2 = y = £> = 2 0, which rules out differential trends. This specification and essentially a difference-in-difference estimator adjustment for the inputs, 1 (W inner) pj , restrictions we in > refer to , it Model as 0)) p;t spij t |a p ,^i t Ap] , = Formally, after 1. the consistency of is in this X model 0. estimate the entirety of equation (8) without imposing such on the trends, and label and trend break we and l(r>0) ;t requires the assumption that E[(l(Winner) x 1(t In other specifications, we make this Model productivity. In theory, 2. This specification allows for both a Model 2 allows mean shift us to investigate whether any productivity effect occurs immediately and whether the impact evolves over time. In practice, disentangling these effects t = is demanding of the data because our sample and there are only six years per case to estimate 6 X and 6 2 5 difference between incumbent plants' Model 1 and Model 2 is in equation (8) control for unobserved determinants of MDP TFP in winning and losing counties losing counties after the to the MDP important MDP opening (y), opening (H), This way after the MDP opening opening. These terms control for to assess the validity (k), (ip), a a trend break in winning and and a differential time trend differential pre-trend in might that differences in winning counties (5), a time trend in winning and losing counties change in winning counties prior winning counties (fi) will serve as an of this research design. The specification also includes three of fixed effects: plant fixed effects (a p ), so the comparisons are within a plant; 2-digit SIC industry by year fixed effects (n it ) to account for industry-specific fixed effects for each case (A ) to ensure that the impact of the p comparisons within a winner-loser framework the across all A intuitive appeal pair. These case fixed MDP effects TFP shocks; and separate opening is recreate identified in from a regression of pair-wise differencing within cases, averaging this effect cases. few further estimation to affect input utilization, (see, e.g., Griliches are practical that the latter allows for differential pre-trends in otherwise be confounded with the spillover effects of the sets The other main . TFP. The other terms TFP only balanced through is unobserved demand shocks are likely this raises the possibility that the estimated P's are inconsistent and Mairesse 1995). This has been a topic of considerable research and we we implement the unaware of a complete standard fixes, and details bear noting. First, including: solution. In a variety of robustness specifications, modeling the inputs with alternative functional forms 23 (e.g., the translog); fixing the P's equal to their cost shares at the plant of investment, flexible functions capital, materials, and and industry-level; controlling for and instrumenting for current labor; 2004a and 2004b; van Biesebroeck 2004; Olley inputs with lagged changes in inputs (Syverson and Pakes 1996; Levinsohn and Petrin 2003; Ackerberg, Caves, and Frazer 2006; Blundell and Bond we 1998). Additionally, experiment with adding fixed effects for region-by-year or region- by-industry-by-year, and allowing the effect of inputs to differ by industry or by winner and post-MDP The status. basic results are unchanged by these alterations in the specification. also note that unobserved demand shocks main parameters of interest and (9i counties in the years after the are only a concern for the consistent estimation 02) if they systematically affect MDP opening, incumbent plants in We of our winning controlling for the rich set of covariates in equation (8). Second, some cases in is estimated on a sample of plants from the entire most specifications the sample country, but in ASM for counties in the this equation every year from x = is in these dates of the MDP openings, this we probe Third, changes is we prices of incumbents' output. corroborating evidence for Fourth, in the years between t = -7 and A TFP how of sample may all we of the reported standard all Due to the cases. 24 number of supplementary be influenced by unobserved many of the factor input demand provides biases associated with estimating outcomes among plants in the same county, both within periods and over focus on weighted versions of equation total (8). data from the entire country is used, the sample value of shipments is time. Specifically, the specifications are in t = -8 to heteroskedasticity associated with differences in plant size. This weighting also limited to plants that are in the /ISM for account for means at least that the 14 consecutive years. Data from in errors are clustered at the county level to account for the weighted by the square root of the When 5. mismeasurement of TFP, and changes complementary analysis of plants' increases, without = which we have data from the estimates plants, t we TFP. correlation in Fifth, of the inputs and the industry the longest period for investigate in plant inputs, attrition plant-level for the impact the validity and robustness of our estimates with a For example, specifications. This smaller sample of plants counties from the rest of the country. For most of the analysis, sample to observations further restrict the 23 = 0. -8 through x from only winning and losing counties allows shocks to differ limited to plants from winning and losing all cases is also available for x = -8, but shipments 24 in this period are used to weight the regressions. results measure the change more meaningful than in productivity for the average dollar of output, which in our the impact of the view is MDP on the average plant. 25 V. Results This section effect the is divided into three subsections. The first reports baseline estimates of the of the opening of a new "Million Dollar Plant" on the productivity of incumbent plants same county through the estimation of equation (8). The second subsection explores in potential channels for the agglomeration effects by testing whether the estimated spillovers vary as a function of economic distance. The of the estimates for third subsection explores the implications the profits of local firms. A. Baseline Estimates Columns (1) and (2) of Table 4 report estimated parameters and from a version of equation log of inputs, year by (8). Specifically, 2-digit and the event time indicators the natural log of output SIC industry fixed in a sample that is Column and losing counties (Column 1) (3) reports the yearly difference 2), regressed on the natural is effects, plant fixed effects, case fixed effects, restricted to the years t reported coefficients on the event time indicators reflect yearly (Column their standard errors = —7 mean TFP through t in in 5. The winning counties MDP relative to the year before the between estimated mean TFP = opened. winning and losing counties. Figure mean TFP in 1 graphs the estimated coefficients from Table winning and losing counties (Columns 1 trends The top panel and 2 of Table the differences in the estimated winner and loser coefficients The Figure has 4. (Column 4). 3 The bottom panel of Table three important features. First, in the years before the among incumbent plants were very similar statistical test fails to reject that the trends in separately plots plots 4). MDP opening, TFP winning and losing counties. Indeed, a were equal. This finding supports the validity of our identifying assumption that incumbent plants in losing counties provide a valid counterfactual for incumbents 25 Finally, we in winning counties. hired a graduate student at Princeton to review publicly disclosed and annotated versions of all STATA programs. This person was not associated with the authors or their institutions prior to serving as the program proofreader. To the best of his knowledge, the computer codes were correct. The authors remain fully responsible for any coding errors in the analysis. 25 MDP Second, beginning in the year of the TFP between difference in improvement relative flattening of the availability TFP to the continued TFP a sharp is upward break The top panel shows the winning and losing counties. mainly due is opening, there in the that this decline in losing counties and a trend in winning counties. This underscores the importance of the of losing counties as a counterfactual. For example, a naive comparison of winning counties before and after the MDP opening would suggest that TFP in had a negligible impact it on incumbents' TFP. Overall, these graphs reveal much of the paper's primary finding. This relative increase in TFP among incumbent plants in winning counties TFP variety of tests in the remainder of the paper. Third, is confirmed throughout a We displays a negative trend. discuss this feature in detail in Section VI. Turning fitting different parameter, B Panel which 9i, to the statistical models, the versions of equation (8). For and its four columns of Table 5 present results from Model 1, Panel A determined by the reported 0i TFP evaluated at t = reports the estimated "Mean standard error (in parentheses) in the reports the estimated change in is first Model ("Level Change" row) and 62 ("Trend Break" row). plants in winning and losing counties. In is Shift" row. For all shift 2, 5 in the "Effect after 5 years" row, "Pre-Trend" row contains the coefficient measuring the difference opening mean in pre-existing trends specifications, the estimated change determined during the period where t ranges from -7 through 5, The between after the as the 26 MDP sample is balanced during these years. In Columns (1) and (2), the sample includes all report data for at least 14 consecutive years, excluding Column (3), the sample is manufacturing plants all plants owned by restricted to include only plants in counties that ASM that MDP firm. In in the the won or lost a MDP. This restriction means that the input parameters and the industry-year fixed effects are estimated solely from plants in these counties. Incumbent plants are -8 < x < Column now required to be in the data only for (not for 14 consecutive years, though this does not change the results). Finally, in (4), the sample is restricted further to include only plant-year observations within the period of interest (where x ranges from -7 through 5). This forces the input parameters and industry-year fixed effects to be estimated solely on plant-by-year observations that identify the spillover parameters. This sample estimation details are noted 26 This is calculated as 9, at is used throughout the remainder of the paper. Further the bottom of the table and apply to both + 69 2 because we allow , the MDP to affect outcomes 26 from x Models = 1 and through x = 2. 5. The is associated with a substantial relative increase in Model counties. Specifically, on TFP appears more Model appropriate. Results from approximately 12% increase in in In change Column 4, in millions of 2006 percent change by the = -1. For Model 1, dollars. large, row demonstrate in also owned by from both models are all the of equal trends winning counties plants' total shipments in $170 increase in total output of results TFP following a same years and own the TFP at i = 5 nearly the its set randomly chosen ASM for plants. opening in year t average level of = 5. MDP is These output. magnitude. that is based on using plant of 47 plant openings was randomly industries as the in the MDP in t The Model 2 estimate million. from a "naive" estimator manufacturing plants firms that statistically These numbers are calculated by multiplying the estimated with the Model 2 effect ASM in associated with an square brackets evaluate the average magnitude of openings without an explicit counterfactual. To begin, a the sample includes is an increase in output of roughly $429 million (5) presents chosen from the opening that the null hypothesis Section VII discusses the interpretation of this change and Column MDP this calculation indicates that the increase in larger, suggesting numbers are roughly 4.8%. As the figure and are unaffected by the specification changes. mean value of incumbent was associated with an annual even winning plants in winning and losing counties cannot be rejected. in numbers the MDP opening that the be increasing over time so Model 2 seems 2 suggest that the criteria, the "Pre-trend" TFP among incumbents TFP of five years later. Estimates from zero by conventional Furthermore, entries in TFP to 1 TFP among incumbent implies an increase in 1 highlights, however, the impact different from Figure entries in Table 5 confirm the visual impression MDP openings. The remainder of fourteen consecutive years, and not With these data, we fit a regression of the natural log of output on the natural log of inputs, year by 2-digit SIC fixed effects, and plant fixed effects. In in a Model winning county 7 to reported mean two additional 1, 1 dummy variables are included for whether the plant years before the randomly chosen opening or shift is the difference in these following the opening). In Model and post-trend variables. The 2, the two same two shift in level coefficients dummy (i.e., mean and the average change in The TFP variables are included along with pre- and trend are reported, along with the pre-trend and the total effect evaluated after five years. Finally, this procedure the reported parameters are the to 5 years after. is is implemented 1,000 times and standard deviation of those estimates. This naive "first-difference" style estimator indicates that the opening of a 27 new plant is associated with a -3% estimates from the MDP TFP of incumbent plant openings. This in plants 15% (Model 1) to was on a downward trend similar to is incumbent plants' TFP, depending on the model. If the research design are correct, then this naive approach understates the 10% (Model extent of spillovers by that -5% change to what is observed in 2). in our The estimated advance of the randomly selected new MDP sample of winners. Overall, the absence of a credible research design can lead to misleading inferences It is "pre-trend" indicates in this setting. important to document the degree of heterogeneity in the treatment effects from the 47 separate case studies that underlie the estimates presented thus far. Figure 2 explores this heterogeneity by plotting case-specific estimates of parameter 8] in Model confidence intervals. Specifically, the Figure plots results from a version of Model interacts the (l{Winner) x 1(t > 0)) with indicators variable specification yields 45 estimates of 0i, as results from Census Bureau's confidentiality the estimated impacts on TFP of incumbent MDP from zero at the 5% characteristics, but the limited we with confidence. Specifically, size, MDP whether the When is We level. to substantial heterogeneity in Twenty-seven of the 45 estimates are explored whether this heterogeneity number of cases provide insufficient is power statistically related to the to detect regressed the estimates against three measures of the owned by a foreign that comply with the and 9 of the negative estimates are positive. Thirteep of the positive estimates different plants. is 1 each of the cases. This two cases were omitted Figure 2 reveals that there rules. for 95% and their 1 company, and whether it much MDP's an auto company. is these multiple measures were included jointly, none were significantly related to the estimated effect of the Ultimately, alternative way to MDP's TFP is opening. 27 a residual, and residual labeling must be done cautiously. examine the MDP impact, we As an estimate directly the changes in incumbent plant output (unadjusted for inputs) and inputs following a MDP opening. Contrasting changes in outputs and inputs can shed light on whether productivity increased without imposing the structural assumptions of the production function. Put another way, producing more with 27 less after the MDP opening? Factor input decisions also Separate regressions of the case specific effects on the statistically significant negative coefficients. This result MDP's is total output or the MDP's total consistent with the possibility that large incumbents are left to hire labor and other inputs that are inferior in unobserved ways. incumbents the are reflect firms' labor force generated when On the MDP is very the other hand, we any significant differences when separately testing whether the productivity effect varied by the ratio of the MDP's output to county-wide manufacturing output, whether the MDP is owned by a foreign company, or whether the MDP is an auto company. failed to find 28 many of optimization decisions, and do not share technology (e.g., Model exclude the inputs as covariates. For Model to 13% (Columns increase less. Across specifications, all Table MDP less after the 28 Furthermore, 5. 1, is it Model in output. Overall, is it output increases by 2, change in in all MDP (8), but 1) and inputs 8% and inputs of the inputs is roughly appears that incumbent plants produced consistent with the some evidence of increased is 12% (Column output increases by striking that the opening, which there and Model 2 versions of equation 1 2 through 5). For equal to or less than the increase more with changes incumbent plant output and inputs following a in opening. These estimates are from the by 4 potential biases as output price effects). Table 6 reports estimated changes increase same the TFP increases uncovered in input use, reflecting firms' optimization in the face of higher potential productivity. Estimates of Spillovers by Economic Distance B. What mechanisms might discussed some mechanisms subsection attempts to shed explain the productivity gains estimated above? Section IA may that some light be responsible for agglomeration spillovers. This on the possible mechanisms by investigating how the estimated spillover effect varies as a function of economic distance. A similar approach has been used by Moretti (2004b) and Ellison, Glaeser, and Kerr (2009). (i) By Industry. Table 7 shows separate estimates from the baseline model for samples of incumbent plants in the MDP's 2-digit industry and all other industries. In general, one might expect agglomeration spillovers to decline with economic distance (equation it is natural to explore whether spillovers are larger within an industry. substantial heterogeneity in technologies industry, only 16 [5']). As a first pass, While there can be and labor forces among plants within a 2-digit SIC of the 47 cases have incumbent plants in the MDP's 2-digit industry. Thus, the research design and available data do not permit a discrete analysis at finer industry definitions. The mode! suggests that firms should substitute away from labor and toward capital. supportive of this prediction, though directly estimating changes making 19 The in the capital/labor ratio 29 point estimates are not gives imprecise estimates, definitive conclusions unwarranted. In the spirit of work by Bloom et al. (2007) and Jaffe (1986), technological overlap between industries. Lacking patent data, share of industrial output that is we we explore defining a continuous measure of define at the 3-digit SIC industry level: (1) the sold to each manufacturing industry; and (2) the share of manufactured inputs that from each manufacturing industry. For each measure, we calculate the overlap between an incumbent and the MDP industry by taking the product of those vectors. We then estimate equation (9), interacting the MDP effect with each measure of industrial overlap. There is evidence of differential spillovers based on overlap defined with inputs, but not with outputs. We suspect that overlap in plant output consumed by the are received firm's industry 29 Column of Table 7 reports estimates for 1 Columns basis of comparison. incumbent plants in the MDP's and (2) (3) report estimates 2-digit industry and of Table (4) Table in 5, as a from the baseline specification for other industries, respectively. all columns are from the same regression. As in these from Column industries all the 5, numbers The entries square brackets in convert the estimated percent changes into millions of 2006 dollars. The estimated changes are substantially larger in the example, the estimated increase 17% significant in Model TFP in for plants in the same 33% and a poorly determined 1 MDP's own 2-digit industry at x = in Model sample size, which was 1. The TFP 2-digit in the confirm this visual impression. As trend in Model in MDP MDP in 1 and industry and other industry estimates are noisy due to the small also evident in the statistical results. Importantly, there evidence of differential trends downward 2. In contrast, 2. Figures 3 and 4 graph annual changes in TFP, providing 2-digit industry analogues to Figure 3.3% For a statistically is Model 5 in estimates for plants in other industries are a statistically insignificant marginally significant 8.9% 2-digit industry. years before the Figure in losing counties 1, MDP's not any is opening, and statistical tests the estimated impact reflects the continuation of a and a cessation of the downward trend in winning counties. To probe the role of economic distance further, Dollar Plant" openings that were in the MDP the original 82 MDP retail we identified an additional ten "Million and wholesale trade sectors. openings, but are not included in the main sample of 47 manufacturing openings. Estimating equation (8) for these ten trade sector changes of -2.2% (2.9%) in Model year observations in 31 counties). generate similar TFP These plants are part of increases, 1 It and 4.9% (6.5%) in MDP Model 2 (on a openings, we find TFP sample of 12,105 plant- appears that the non-manufacturing sector openings did not though the estimates in Model 2 are too imprecise to reject their equality with the baseline estimates. These findings provide further evidence that the spillovers are concentrated among manufacturing sector only plant inputs These inputs is is plants that are economically close to the new 30 plant. a poor reflection of overall industrial overlap, and are not confident that overlap results may also provide a test of causing plants to move closer to whether the estimated spillovers are due their production possibility frontier. to increased Specifically, competition for these new non- manufacturing plants increase competition for land, labor, and other local inputs. The resulting increase prices may cause all in input plants (regardless of industry) to search for opportunities to increase productivity. In such a situation, all local plants this paper's in more persuasive. would exhibit increased TFP. These primary findings. 30 results suggest that this mechanism does not explain By Continuous Measures of Economic (ii) economic proximity more worker flows, proximity, and flows input-output economic proximity variables are standardized interpretation, these standard deviation of one. In investigate the role of by using several measures of economic proximity directly technological We now Distance. (Table that capture To ease 2). have a mean of zero and to cases, a positive value indicates a "closer' relationship all the between the industries. we Specifically, (9) \n(Ypijt ) = ft + estimate the following equation: \n(L pijt > k1(t 0) yt + 7t 2 (1(t > the MDP between the 1(t > 0)) p;£ ln(Mpi/t proximity dummy ^(KW'inner) + + n 3 (l(Winner) pj x for a with variables is 713, l(Winner) pj, which 0) i} x Proximity^) ) + ap winning county, the is new economically distant from the A zero plant, relative to new dummy (1 (Winner) and 0), for after the MDP x opening, and the in "closer" industries experience Table 8 reports estimates of 713 for six columns include the proximity measures one one standard deviation increase MDP's is same comparison among incumbents that the estimated productivity spillover the incumbents in a county, regardless of their spillover. This finding > adds incumbents that are geographically close but plant (relative to the means coefficient and the 1(t that 1 after the economically close to the plants' industry > x Proximity MDP opening. A positive coefficient means that the estimated larger after the MDP opening for incumbents that are geographically and TFP productivity spillover all 1(t pj the coefficient on the triple interaction is measure of proximity. This coefficient assesses whether plants loser counties). + 81(Winner) pj simply an augmented version of Model is coefficient of interest a greater increase in ) Epijt This equation of the The + B^KWinner) x ft a measure of economic proximity between the incumbent plant industry and industry. interactions lf>0)pjt. is ln(^;t ) + ft 0) x Proximityij) + nit + Ap + where Proximityij + ft ln(^% ) + ) economic proximity industry is new the at a time. For example, Column same for plant. measures of economic proximity. The CPS Worker in the to the is in first six (1) reports that a Transitions variable between incumbent associated with a 7 percentage point increase in the consistent with the theory that spillovers occur through the flow of workers across firms. One possibility production or information on new is that new workers share ideas on how to organize technologies that they learned with their previous employer. This measure tends to be especially high within 2-digit industries, so 31 this finding was foreshadowed by the In results in Columns (2), (3), and is 2-digit industry. the measures of intellectual or technological linkages (4), indicate meaningful increases in the spillover. shared own Table 7 based on the plant's The mechanism by which these ideas precise unclear, although both the flow of workers across firms and the mythical exchange of ideas over beers between workers from different firms are possibilities. Notably, there variation in these measures within 2-digit industries than in the Columns (5) and (6) provide an auto manufacturer encourages (or even forces) all plants owned by the its MDP's finding does not rule out this channel within firms. technology flows is CPS labor transitions measure. fail to support the types suppliers to adopt more of stories where efficient production firm are dropped from the analysis, so this The finding on the importance of labor and consistent with the results in Ellison, Glaeser and Kerr (2009) and and Glaeser (2002), while the finding on input and output flows stands Ellison, more is support for the flow of goods and services in little determining the magnitude of spillovers. Thus, the data techniques. Recall, are Dumais, in contrast with these papers' findings. In the Column simultaneously. The (7) specification, we include all the measures of economic proximity labor flow, the citation pattern, and the technology input interactions remain positive but are statistically insignificant. The customer and supplier all interactions are negative and statistically insignificant. Overall, this analysis provides support for the notion that spillovers occur between firms that share workers evidence workers is and between firms that use similar technologies. In terms of Section IC, consistent with intellectual externalities, to the extent that they are who move from firm to firm, and to the extent that they occur technologies that are reasonably similar. The estimates in Table 8 seem among embodied less consistent all in firms that use with the hypothesis that agglomeration occurs because of proximity to customers and suppliers. caution against definitive conclusions, because the utilized measures are this We imperfect proxies for the potential channels. Further, the possibility of better matches between workers and firms could not be directly tested with these data. C. Firm Entry and Labor Costs as Indirect Tests of Spillovers Baseline estimates found economically substantial productivity gains for incumbent establishments following the opening of the new MDP. 32 In the presence of positive spillovers, the model makes two empirical predictions and increased local input costs. productivity spillovers are larger than short-run increases in the cost of local First, if inputs, the that are explored in this subsection: increased firm entry MDP county should experience entry by new firms (relative to The 9 tests this prediction at the county level. entries in Panel data from the Census of Manufactures, which manufacturing output (Column 2) counties, and owned by plants all come from 1 county. in the MDP regressions that use conducted every five years. The dependent number of establishments (Column log of the variables are the is losing counties). Table The sample is 1) and the log of total winning and losing restricted to firms are excluded from both dependent variables. The covariates include county fixed effects, year fixed effects, case fixed effects, and an indicator for whether the observation is from MDP after the opening. The parameter of interest is associated with the interaction of indicators for an observation from a winning county and from after the MDP opening, so Column it is (1) reports that the winning counties in all a difference-in-difference estimator of the impact of the MDP after the plants are of an equal size. because it treats an increase number of manufacturing in reports that the opening of a The total A limitation of this value of output is associated with a measure new attracted new economic assumes it plant equally. 14.5% increase that in total Column (2) output in the this is not estimated precisely. Overall, these results are consistent with estimated increases in MDP 1 by roughly 12.5% that is •3 economically more meaningful, is output at an existing plant and a MDP manufacturing sector, although opening. plants increased MDP opening. activity to the winning counties TFP, as it appears that the (relative to losing counties) in the manufacturing sector. Presumably, these new manufacturing establishments decided to locate the winning counties to gain access to the productivity advantages generated by in the spillover effect. The second theoretical prediction is that, if spillovers are positive, the prices of local inputs will increase as firms compete for these factors of production. supplied input for manufacturing plants wage 31 is labor. This prediction is The most important locally tested using individual-level data for winning and losing counties from the 1970, 1980, 1990, and 2000 Censuses of Because data years are 1 - is available every 5 years, depending on the Census year relative to the associated with one earlier MDP opening, the sample opening and 4-8 years after the MDP opening. Thus, each date and one later date. Models are weighted by the number of plants 5 years before the years -6 to -10 and column 4 is MDP weighted by the county's total 33 manufacturing output in MDP in the years -6 to -10. opening county is in Population. 32 These data are preferable Manufactures (i.e., the aggregate wage to the measure of labor costs reported bill for we year, education US We and case fixed effects. from a winning county, occurs indicators. 33 This interaction is the Model version of equation 1 Column (3) in Panel 2 counties after the MDP wage million after the MDP MDP opening, and the interaction of these two an adjusted difference-inis analogous to (8). of Table 9 reports that wages increase by 2.7% wage and is winning in marginally statistically significant. Multiplying the by the average labor earnings bill for MDP opening. This who is opening on wages. This equation employers finding is in winning counties implies that the by roughly $151 in all industries increased consistent with positive spillovers and an sloping labor supply curve, as in the model in Section Moretti (2004), is opening, after adjusting for observable individual heterogeneity. This increase quality-adjusted ^annual of for interactions and year, sex and race and Hispanic and the focus of the regression and effect appears quantitatively sizable estimated 2.7% dummies also include indicators for whether the observation after the difference estimator of the impact of the education and experience). (e.g., estimate changes in log wages, controlling for worker age and year, age-squared and citizenship, Census of production and non-production workers), which does not provide information on the quality of the labor force Specifically, in the I. This finding finds significant productivity spillovers is upward also consistent with and increases in wages of similar magnitude. It is possible to use the estimated increase in calculations of the MDP's impact on incumbent plants' Table 5 indicated an increase in TFP not possible to estimate a version of assume that wages to make some back of the envelope profits. Recall, the Model of approximately 4.8% (we focus on Model Model 2 with 1 1 result in because it is the decennial population Census data). If workers are homogenous or that high and low substitutable in production, then the labor market-wide increase in workers are perfectly skill wages applies throughout manufacturing sector. In our sample, labor accounts for roughly we 23% 32 of total costs, the so the The sample is limited to individuals who worked last year, worked more than 26 weeks, usually work more than 20 hours per week, are not in school, are at work, and who work for wages in the private sector. One important limitation of the Census data is that they lack exact county identifiers for counties with populations below 100,000. Instead, it is possible to identify PUMAs in the Census, which in rural areas can include several counties. This introduces significant measurement error, which is partly responsible for the imprecision of the estimate. The pre-period is defined as the most recent census before the MDP opening. The post-period is defined as the most recent census 3 or more years after the MDP opening. Thus, the sample years are - 10 years before the MDP opening and 3 - 12 years after the MDP opening. 1 34 estimated 2.7% increase in skill adjusted wages implies by that manufacturers' costs increased 13% approximately 0.62%. The increased production costs due to higher wages are therefore of the gain in TFP. These calculations demonstrate due TFP do not translate directly to profits higher costs of local inputs. Since the prices and quality of other inputs are not to the observable, not possible to determine the total increase is it wage expect the that the gains in plants enter or increase to be larger for plants and industries that experience expand and compete workers with the for we production costs. Further, in TFP increases as skills relevant for these sectors. these reasons, this back of the envelope calculation should be interpreted as a lower For bound of the increase in input costs. In the long run, an equilibrium requires that the total impact on profits is zero. VI. Validity and Robustness Our main empirical finding substantial average increase in incumbent plants research design of many is in counties that in V Section that is TFP among incumbent MDP openings are associated with a plants in those counties, relative to narrowly missed receiving the supported by the similarity of pre-trends in new TFP plants. (Figure The 1) validity of this and the balancing ex-ante observable characteristics of winning and losing counties and their incumbent plants (Table 3). Nevertheless, the possibility remains that the paper's identifying assumption invalid and that incumbent plants productivity shocks coincident to the Consequently, specifications, this section winning counties experienced unobserved positive in new plant's opening. explores the robustness (i) unobserved mismeasurement; and, (i) estimates to various the role of functional form assumptions, unobserved industry and regional shocks, and weighting; (iii) of the and investigates several possible alternative interpretations of the estimated spillover effects. Specifically, this section analyzes: inputs; is (vi) changes changes in inputs; in the price (ii) attrition; (iv) the general endogeneity of plant (v) declining plant TFP and of plant output. Functional Form, Industry and Regional Shocks, and Weighting. Table 10 reports estimates from a series of specification checks. the results from the preferred specification in We As a basis for comparison, Column (4) of Table Column (1) reports 5. begin by generalizing our assumption on plants' production technology. Estimates 35 in Table 5 assume a Cobb-Douglas technology. In Column (2) of Table Column the translog functional form. of each production input to effect (3) is 10, inputs are modeled with based on a Cobb-Douglas technology but allows the differ at the 2-digit SIC level. This model accounts for possible differences in technology across industries, as well as for possible differences in the quality of inputs used by different industries. For example, some similar across different manufacturers, Column the (4) allows the effect of the inputs it is possible that even if technology were industries use more labor than others. skilled to differ in winning/losing counties and before/after MDP opening. Columns (5) and add census division by year fixed effects and census division by year (6) by 2-digit industry fixed effects. These specifications aim to purge the spillover unobserved region-wide shocks or region by industry shocks correlated with the probability of winning a Until this point, we have of effects might be to productivity that MDP (e.g., a declining Rust Belt). presented results based on specifications weight that observations by the square root of the plant's total value of shipments eight years prior to the MDP opening. As discussed above, the resulting estimates measure the change in productivity per average dollar of output, which reflects the Column (7) reports the results 2, the estimated change is economic impact of the from unweighted regressions productivity for the average plant. For Model full Model 1, concentrated why among smaller plants Taken the largest plants. fail to 34 benefit from the is set A is the in None of the within one standard error of the baseline estimate to is to explore The weighted estimates appear estimates are smaller than the baseline ones, the magnitude of the decline fail be plant's presence. contradict the findings from the baseline specification in Table 5. Although Overall, these results may both models, which insensitive to the specific functional form of the production function. all for aside, the results indicate that the spillovers are together, the results in Table 10 are striking. example, they are in 1.46% (1.07%); promising avenue for future research new change findings should be interpreted cautiously because the building capital coefficient becomes slightly negative a sign of misspecification. If this concern reveal that the estimated change 0.65% (2.81%). These plant. Nevertheless, undermine the conclusion from Table 34 in is specifications many of the modest. For both Models 5 that the to be 1 opening of a and 2. MDP unweighted regressions, the estimated effect among incumbent plants in the largest decile (8 years prior to the is: 2.90% (3.12%) higher than the average effect of 1.16% (0.98%) for Model l;and 16.7%(11.0%) higher than the average effect of -1.1 6% (3.04%) for Model 2. In MDP opening) 36 TFP among incumbent leads to a substantial increase in theories of spillovers. (ii) and plants, with this is consistent 35 General Endogeneity of Inputs. An important conceptual concern labor inputs should be treated as endogenous, because the same is that capital and forces that determine output also determine a firm's optimal choice of inputs (Griliches and Mairesse 1995). Unlike the usual estimation of production functions, parameters, results in 9i and 62, our aim is the consistent estimation so the endogeneity of capital and labor of the spillover only relevant to the extent that is it biased estimates of these parameters. This subsection employs the productivity literature's techniques to control for the endogeneity of capital and labor to assess this issue's relevance in this paper's setting. We First, in employ Columns three (2) main approaches, and the and (3), we calculate TFP results are collected in for each plant by Appendix Table fixing the parameters 1. on the inputs at the relevant input's share of total costs (van Biesebroeck 2004; Syverson 2004a; Foster, Haltiwanger, and Syverson 2008). This method may mitigate any bias in the estimation of the parameters on the inputs associated with unobserved demand shocks. In these two columns, the cost shares are calculated at the plant level and the 3-digit SIC industry level over the full sample, respectively. Second, Columns (4) through (6) present estimates based on methodologies that build on work by Olley and Pakes (1996). These methods are based on the result that, under certain conditions, adjustment for investment or intermediate inputs (e.g., materials) will correlation between input levels and unobserved shocks to output. Column remove the (4) controls for 4th degree polynomial functions of log capital and log investment, and the interaction of both functions (separately for both types of capital). Column (4), Column (5) includes the same controls as but replaces investment with materials; an alternative proposed by Levinsohn and Petrin (2003). Building on Column log materials, as collinearity may (5), Column (6) includes interactions between log labor and complicate the estimation of the labor coefficient (Ackerberg, Caves, and Frazer 2006). Third, 3 We Column (7) presents estimates that instrument for current input levels with also tested whether the results are sensitive to the choice of the date of the year that the plant is first observed in the SSEL as the MDP MDP opening date. Model 1 opening. lagged When we estimates a change of use the 5.23% (2.39%) and Model 2 estimates a change of 11.2% (5.57%). When we use the year of the MDP article for the opening date, Model 1 estimates a change of 4.58% (2.45%) and Model 2 estimates a change of 4.88% (4.21%). 37 changes from = -2 t to Bond a technique proposed by Blundell and in inputs; t = -1 may unobserved output shocks in Column expected sign, and = 0. = t 0, change change due to potential (7) reports the weak instrument Appendix Table 1 The may but 2SLS Of course, results. "it that is we have used only the The 72% summed of the 0.86% increase endogeneity typical in variable inputs (labor) and statistically significant, but not 1% and the 36 and coefficients for labor overall production function has mild decreasing returns to scale, with a leading to a way of downward biased and loaded onto is labor. In the baseline specification, all inputs are positive an expected lagged first also reports coefficients from the production function, as a fixed inputs (capital), so the estimated effect of capital is a strong assumption that bias. unobserved productivity shocks lead to changes labor coefficient not be correlated with is assessing the effectiveness of the production function estimation. concern increase in each input Indeed, the estimated first-stage results (not shown) have the not otherwise correlated with output, and this lagged is t predict input levels at (1998). capital. increase in all The inputs in output. Overall, the estimated changes in TFP are consistent with the findings from the baseline specification. This exercise fails to suggest that the possible endogeneity of labor and capital is the source of the estimated productivity spillovers. Unobserved Changes (iii) comprehensive of all in The Inputs. input inputs that affect plant output. measures the in ASM Further, the available data are not may not adequately measure the degree of input usage or the quality of inputs. Consequently, it is possible that the estimated spillovers reflect changes in unobserved inputs, unobserved usage, and/or input quality. This subsection explores these possibilities. State and local governments frequently offer substantial subsidies plants to locate within their jurisdictions. to new manufacturing These incentives can include tax breaks, worker training funds, the construction of roads, and other infrastructure investments. these investments benefits firms other than the road intended for a MDP may (Chandra and Thompson 2000). investment, then 36 As it is and 1 .84% If the productivity gains inappropriate to interpret for building capital, For example, the construction of a new also benefit the productivity of a basis of comparison, the weighted cost shares are capital, MDP. among them some of the incumbent firms we have documented are due to public as evidence of spillovers. 70.9% for materials, plants the year before the 38 possible that It is 23.2% for labor, MDP opening. 3.98% for machinery To we investigate this possibility, total capital estimated the effect of MDP openings on government expenditures and government construction expenditures with data from the Survey of Governments. In models similar to equation (8), we find that the opening of a Annual MDP is associated with statistically insignificant increases in capital and construction expenditures. In most specifications the estimated impact of a insignificant. Even MDP opening is negative and statistically produce positive insignificant estimates, there in the specifications that is no plausible rate of return that could generate a meaningful portion of the productivity gains in winning counties. Based on these measures of public investment, it seems reasonable to conclude that public investment cannot explain the paper's results. Incumbent plants may respond capital usage. If MDP to the opening by increasing the intensity of their winning counties had been depressed and the capacity, then incumbent plants might increase production simply closer to capacity. As an indirect test of this possibility, we models and insignificant changes find small capacity utilization The is is unclear. If the incumbent plants or the opening may upgrade of equation (8) used by unobserved changes would lead (iv) Sample may decline. If the attracts higher quality Plants. If the attrition though the MDP receives bad workers from workers to the county, then incumbent TFP MDP TFP change in for for their quality, incumbent plants. increases competition for inputs and raises in quality-adjusted wages, this might to close. Indeed, for a variety sample of incumbent plants in of the true by the estimated changes encourage plants with declining differential we poaches good workers from incumbent plants, the to an underestimate or overestimate local input prices, as suggested measured 6, the quality of their workforce. Since the specifications control for the Attrition of attrition in the Table in labor quality, number of hours worked by production and non-production workers, but not this in measure. This finding suggests that greater capital MDP quality of the workforce in existing plants plants MDP opening unlikely to be the source of estimated productivity spillovers. results could also be influenced direction of this bias their capital stock increasing in the use of the capital is identical to the version in this by operating was used below estimated whether the affected the ratio of the dollar value of energy usage (which stock) to the capital stock. In capital stock of reasons, differential winning and losing counties could contribute to the productivity trends among could either result from plants shutting survivors after the down MDP opening. This operations or from plants continuing operations but dropping out of the group of plants that are surveyed with certainty as part of the 39 ASM. The available evidence suggests that differential attrition winning counties. Similar numbers of winning and losing plants remained of spillovers in sample end: at i = at its 72% 5 as a fraction counties is unlikely to explain the finding is winning counties and in of the number of plants 68% at x - in losing counties 0). The (i.e., slightly larger attrition rate in losing that the is MDP opening allowed some winning county plants to remain open that would have otherwise closed. Thus to the extent that a weakening plants operating, the baseline analysis Within aggregate MDP plants in some MDP such attrition numbers, the location. This may plants all we might industries, then MDP, 71% sample attrition increases input prices for each the will underestimate the overall MDP TFP increase. 37 If and disproportionally increases productivity for expect to see increased agglomeration of those plants in of incumbents (v) Declining Plant in winning counties and 5 years). 69% Within the same 2-digit SIC as in losing counties remained in the 5. TFP and Mismeasurement. incumbent plant TFP was declining prior counties, and in losing counties after the to the MDP MDP of the results The decline in and conclude TFP appear continuously in the is it in opening. This finding we is will it affects the TFP in the overall of large and aging manufacturing plants that in the eight years prior to the and sample that striking, as productivity explore whether MDP specifications include both plant and industry-by-year fixed effects. this specification show does not. restricted to a set ASM 1 both winning and losing not necessarily inconsistent with rising is economy. Recall, the sample that Table 4 and Figure opening generally increases over time in the overall economy. Here, " The opening keeps might change the mix of existing firms. only occur in the long-run (more than = interpretation MDP an intriguing hypothesis, but difficult to test in this setting because is at t TFP from number of plants consistent with the paper's primary result. Specifically, one seemingly reasonable interpretation of this result the the in the opening. Further, the The estimated decline in miss the process of creative destruction where less winning and losing counties prior to the MDP TFP trend in losing counties was -0.0052 (0.0080). Further, the estimation of equation (8) on the sample of plants that is present for all years from -7 to +5 yields results that are qualitatively similar to those from the full sample. null hypothesis of equal trends opening cannot be rejected; the TFP in TFP among attriting plants in trend in winning counties minus the 40 no productive plants are replaced by more productive plants. The estimated downward trend appears of plants in all analysis of the U.S. counties, and MDP be a general phenomenon for similar samples TFP among plants declined annually similar plants in openings. Specifically, we From our not limited to winning and losing counties. randomly selected manufacturing plant openings TFP of incumbent declining is to the U.S. (Table 5, in by 0.5% prior Column [5]), we find to the opening. Further, U.S. counties over the 6-year periods following the all created a sample of all U.S. plants that appear in the ASM for fourteen straight years, deflated output and materials by the CPI, and regressed log output on log capital stocks (building and machinery, created using the same permanent inventory method), log labor hours, log materials, plant fixed effects, industry fixed effects, year fixed effects, and weighted the regression by plant output. The estimated year effects report average changes TFP over each year, and we correspond to the periods following declined in all words, the MDP openings. U.S. counties by an average of 39 4.7% (with pre-MDP opening TFP changes among over 6-year periods that Over these periods, we find that TFP a standard error of 0.4%). In other large and aging plants in our winning and losing counties (and post-MDP opening changes TFP TFP calculated average changes in in sample of in losing counties) are similar to changes among large and aging plants throughout the U.S. An reflects alternative explanation measurement is that measured declines error, particularly in the construction practice in the existing literature, we in TFP are a statistical artifact that of capital stocks. Following standard construct capital stocks based on depreciating plants' past inputs and adding deflated investments in 40 new capital. This procedure uses standard NBER depreciation rates, but if these rates are too low for firms in our sample then aging firms will begin to have more measured capital than they have firms' TFP appear to decline in firm age. TFP changes and plant fixed effects, appear labor or materials if firms' affect the relative Mechanically, this will Because the regressions control became unobservably worse to affect our comparison of firm 38 TFP in make for industry-by-year are estimated solely on aging plants. Similar biases Such measurement problems are unlikely need not in reality. would as plants aged. main results of interest, as this bias winning and losing counties. For it to Moreover, we note that our sample overlaps the late 1970s and early 1980s, which was a period of poor economic performance and low productivity. Foster, Haltiwanger, and Krizan (2000) have documented that within-plant productivity growth is cyclical, and is particularly low during downturns. 39 For example, if there was one MDP opening in 1987, the period 1987 to 1993 received a weight of 1/47. 40 This is necessary in the later portions of the sample, when book values for current capital are no longer reported. 41 affect our estimates, measurement of capital stocks (or other inputs) would need to be systematically biased in winning counties after the checks provide some reassurance on Column vary by industry and this issue: MDP from Table opening. The previous robustness 10, (4) allows input effects to Column (3) allows input effects to MDP vary after the opening or winning counties (but not the interaction of those two). Further, the specifications Table would be affected 1 differently increases in winning counties after the Changes (vi) is Output. Another concern the quantity of output. of the productivity virtually all literature, the However, due dependent variable output or price multiplied by quantity. Consequently, do not expect this to show TFP all it is that the theoretically is to the data limitations faced in our models is the value of possible that the estimated spillover of higher productivity. effect reflects higher output prices, instead We error in capital stocks, but Appendix in MDP opening. in the Price of Plant correct dependent variable by by measurement in be a major factor in our context. The sample comprised of is manufacturing establishments that generally produce goods traded outside the county. 41 In the extreme case of a perfectly competitive industry that produces a nationally traded good, output prices would not increase disproportionally To explore this possibility further, in industries that are equation (8) incumbents' that more regional or interacts production and consumption. industry concentration. more local or more concentrated. measure We , 1(t>0) j£ of average , the productivity change We and estimate a industries; in fact, there is by traveled also estimate this regression with a Model x 1(t > (1 (Winner) distance These specifications do not find more concentrated incumbent plants To 43 a county that experienced increased demand. we examine whether l(Winner) pj industry-specific in 1 is 42 larger version of 0)) p;t output , with between measure of incumbents' that estimated changes are larger some evidence for larger effects in on that ship their products further. we use data by detailed good between production and consumption (Weiss 1972). Across all sample plants, the 10th centile is 239 miles, 25th centile is 355 miles, median is 466 miles, 75th centile is 602 miles, and 90th centile is 722 miles. This suggests that most establishments in our sample produce goods that are widely traded outside the county. Across all industries in the Weiss data, distance varies between 52 and ,337 miles, with a mean of 498. Examples of regional industries are: hydraulic cement, iron and steel products, metal scrap and waste tailings, ice cream and related frozen desserts, and prefabricated wooden buildings. give an indication of the tradability of goods produced by firms in a given industry industry code on the average distance travelled by a 1 42 Similarly, input (labor) spillovers goods (Black 43 et al. may be less pronounced for incumbent plants that produce nationally traded 2005). The information on distance is from Weiss (1972). The information on industry concentration of Census ("Concentration Ratios" 2002). 42 is from the Bureau were not Earlier estimates found that the estimated spillovers industries that tend to ship products to the might be largest. Further, the on demand incumbent for MDP MDP plants' output, particularly since these to similar estimated TFP These exercises suggest increases. effects might have similar effects non-manufacturing MDPs and wholesale trade sectors. However, these non-manufacturing in the retail incumbent which output price industry, a context in opening of a non-manufacturing larger for MDPs were did not lead that output price increases are not the source of the estimated spillover effects. VII. Discussion and Implications for Policy A. Discussion The preferred increased by a 5% 12% increase. Model 2 estimates suggest The 12% TFP interpret this large effect fraction incumbent plants' total factor productivity Model following the opening of a Million Dollar Plant, while increase implies an additional manufacturing output five years after the To that and what it MDP $430 million opening. In this Section, opening. There is TFP discuss amount of that is 56% that plants located in counties at the top we 4 level A 12% TFP increase in TFP is higher than the county th at the 10 equivalent to moving from MDP TFP percentile, indicating productive that similar all production the 10th percentile of the county- it deviation increase in the distribution of county TFP. Attracting a we calculate the percentile of the th 56% more of the distribution are distribution to the 27th percentile; alternatively, counties, and to cross-sectional variation in productivity across U.S. plants located in counties at the bottom of the distribution, holding constant inputs. how implies for the spatial distribution of economic activity. counties in the manufacturing sector. For example, the county at the 90 distribution has average annual county average manufacturing productivity explained by the in a tremendous in we put the magnitude of the estimated spillover effect in perspective, of overall variation estimates find 1 is equivalent to a 0.6 standard MDP find this implied shift in the relative standing is a major event for these of counties large but not unrealistic. Our estimates have 44 Specifically, these numbers interesting implications for the distribution are of economic activity obtained using cross-sectional plant-level data from the 1987 Census of Manufactures (the mid-point of our sample period). We regress log output on log inputs (log building capital, log machinery capital, log materials, log labor) and a full set of county fixed effects. We then look at the distribution of the county fixed effects, which represent the average TFP among 43 all manufacturing firms in a given county. across locations. Along with substantial increases in suggests that profits increased, at least TFP in TFP, increased firm entry and expansion However, the documented increase the short-run. does not translate necessarily as a similarly large increase Increased economic activity generated by the MDP in profits for in incumbent firms. leads to firms bidding up local factor prices like labor and land. Difference-in-difference estimates found that wage rates increased by 2.7%, compared to TFP 4.8% increases of in the difference-in-difference wages because we have estimate a trend break model for skill-adjusted Census data). Since this receive a spillover is may become plants in each more new location. This agglomeration, we important, because of the it spatial distribution open In interpreting the it is would fully activity. It in winning counties and the same TFP 2-digit SIC staying in other industries. TFP other new reflects the plants impact of the plant opening and opened in the county following the output increased (Table 10). Consequently, the TFP all new plants and Second, the effect of a MDP MDP opening and overall manufacturing estimates should be interpreted as a general expanded output from incumbent opening is economy. other associated changes. For example, equilibrium reduced-form effect that combines both the direct impact of the impacts of subsequent First, as the partial equilibrium impact of the "Million Dollar Plant" opening, holding constant everything else in the county's it increases, magnitude of our estimates, three points need to be highlighted. inappropriate to interpret the estimated increase in Instead, seems agglomerate around industries. Indeed, despite experiencing substantially larger sample more than winning incumbent plants in the of economic a great deal of documented cross-sectional agglomeration and co- no evidence of incumbent plants is helps explain the existence of expect that local wages rise by more than 2.7% for the particular workers demanded by such there is on decennial might be long-run agglomeration of similar unlikely, however, that industries receiving larger spillovers there to rely to incumbents and the productivity gains are larger for plant, there is industrial clusters, a pervasive feature MDPs. While (we are unable less profitable. similar to the MDP 1 a county-wide increase in labor costs, incumbent firms that do not If labor costs increase equally for all plants that are Model MDP and the plants. not representative of the typical plant opening. The MDPs differ size. MDPs are significantly larger than the average from the average manufacturing plant new selected sample. Unlike most manufacturing plants, the 44 in several respects, most importantly plant in the U.S. Moreover, they are a MDPs generated bidding from local governments, presumably because there was an ex ante expectation of substantial positive spillovers. If spillovers vary in the by industry, then it may be important to note that MDPs tend to be automotive, chemical, computer, and electronics industries (relative to the average manufacturing plant opening). Third, the counties bidding for plants new manufacturing plant opening. We may be those that would particularly benefit from a do not expect that winners and losers were great counties that almost attracted special plants; rather, they were counties willing industrial stimulus. In considering potential locations, the MDPs to provide tax subsidies for might be attracted to a declining manufacturing sector and the expectation of lower future wages. These estimates of agglomeration spillovers come from a selected counties for which we bound (and perhaps set of plants and set of expect large spillovers, which implies that our estimates are a likely upper substantially so). This is an issue of external validity, rather than the consistency of our estimates. However, our estimates are representative of the benefits generated by large plants bid on by these policy. From local governments, which is a research standpoint, finding spillovers from a population of interest for public MDPs appears to be a necessary condition for agglomeration spillovers from a broader set of plants (rather than a sufficient condition) and a call for further research. B. Implications for Local Economic Development Policies The presence of new significant agglomeration externalities implies that the attraction plant to a locality generates external productivity benefits for existing firms. question whether is efficiency-enhancing. in this From agglomeration significant externalities and may context publicly-financed subsidies to attract the point of view externalities of an individual indicates increase efficiency in some that cases. providing locality, subsidies An new of a important plants are presence of the can internalize However, from the aggregate point of view, the efficiency of policy depends on whether the benefits of attracting a new plant for the receiving county are homogeneous. Consider the case where agglomeration spillovers are homogeneous. This could happen, for example, if the functional form for agglomeration economies is linear and productivity spillovers do not depend on economic distance between existing plants and the new Assuming that the new plant will locate somewhere 45 in the plant. U.S. irrespective of the provision of of the subsidies, providing subsidies for plants to locate in a particular city, county, or region country 2009). is socially wasteful from a national perspective (Glaeser 2008; Glaeser and Gottlieb Further, even from a local perspective, the bidding for plants game where all conventional wisdom of the benefits are bid away. that the provision likely to is sum be a zero This special case forms the intuition for the of incentives for firms to locate in particular locations is wasteful. However, our results indicate that productivity spillovers distance between the industry of the the county in advance of when new its profits are high but the spillovers are 8). This heterogeneity is important because plant are heterogeneous, the socially efficient where the sum of cannot capture these spillovers on spillovers plant and the industrial composition of plants located in opening (see Table its the benefits of attracting a for a plant to locate new vary based on the economic its profits own and and the spillovers are greatest. outcome The new plant consequently might choose a location where minimal. In this case, payments is its to plants that generate can increase national welfare because they can cause plants to internalize the externalities in rnaking their location decision. Further, from the point of government, heterogeneous spillovers imply local governments may view of the not bid away all local of the benefits (Greenstone and Moretti 2004). There are at least two issues localities to plants are a one-for-one supply of land is inelastic. If the of incidence that bear noting. First, the exchange of land rents from the former supply of land payments from to the latter if the elastic, the increase in land rents is not substantial variability in the spillovers and is necessarily one-for-one (see Moretti 2010). Second, Figure 2 demonstrated that there this is is could affect the provision and/or magnitude of subsidies. For example, the estimated impact negative in 40% of the cases. Consequently, risk-adverse local governments to provide tax incentives with this distribution may be of outcomes or only willing to bid unwilling less than the average spillover. VIII. Conclusions This paper makes three main contributions. increases in is TFP among incumbent First, the estimates document substantial plants following the opening of "Million Dollar Plants". This consistent with firms agglomerating in certain localities, at least in part because they are 46 more productive from being close to other firms. Second, the estimates shed light on the channels that underlie the estimated spillovers. Estimated spillovers are larger between plants that share labor pools and similar technologies. This consistent with intellectual externalities, to the extent that they occur is use similar technologies or are embodied in workers finding is consistent with higher rates of TFP due matches. Clearly, our evidence on the mechanisms to understand in more detail the sources who move between to among firms. Additionally, this improved efficiencies of worker-firm not conclusive. Further research is firms that is needed of agglomeration economies. Third, firms appear to pay higher costs in order to receive these productivity spillovers. Spatial equilibrium requires that increases in TFP are prices, so that firms are indifferent across locations. adjusted labor increased is consistent with this prediction. demand to locate in the accompanied by increases The in local input finding of higher prices for quality- The increased levels of economic activity reflect winning county, which leads to higher local prices and a new equilibrium. This paper has demonstrated that directly measuring TFP. These of co-agglomeration rates that spirit, sector. it is tests may tests for the can serve as an important complement to the measurement reflect spillovers, cost shifters, or natural advantages. In this important to determine whether impacts on Further, presence of spillovers can be conducted by TFP are evident outside the manufacturing the significant heterogeneity in estimated spillovers across cases and the variation in estimates across industries underscore that there structural source of these spillovers. 47 is still much to learn about the References Acemoglu, Daron. 1996. "A Micro foundation for Social Increasing Returns in Human Capital Accumulation." Quarterly Journal of Economics 11 (August): 779-804. Ackerberg, Daniel A., Kevin Caves, and Garth Frazer. 2006. "Structural Identification of Production Functions." Manuscript, Department of Economics,*UCLA. Audretsch, David, and Maryann Feldman. 1996. "R&D Spillovers and the Geography of Innovation and Production." American Economic Review 86 (June): 630-640. Audretsch, David, and Maryann Feldman. 2004. "Knowledge Spillovers and the Geography of Innovation." In Handbook of Urban and Regional Economics, Vol 4, edited by J. Vernon Henderson and Jacques-Francois Thisse. Amsterdam: North-Holland. Becker, Randy, John Haltiwanger, Ron Jarmin, Shawn Klimek, and Daniel Wilson. 2005. "Micro and Macro Data Integration: The Case of Capital." Working Paper no. 05-02, Center for Economic Studies, Bureau of the Census. Black, Dan, Terra McKinnish, and Seth Sanders. 2005. "The Economic Impact of the Coal Boom and Bust." The Economic Journal 115 (April): 449-476. Bloom, Nick, Mark Schankerman, and John Van Reenen. 2007. "Identifying Technology Spillovers and Product Market Rivalry." Working Paper no. 13060 (April), NBER, Cambridge, MA. Blundell, Richard, and Stephen Bond. 1998. "Initial Conditions and Dynamic Panel Data Models." Journal of Econometrics 87: 1 Moment Restrictions in 15-143. Card, David, Kevin F. Hallock, and Enrico Moretti. 2007. "The Geography of Giving: The Effect of Corporate Headquarters on Local Charities." Manuscript, Department of Economics, University of California, Berkeley. Chandra, Amitabh, and Eric Thompson. 2000. "Does Public Infrastructure Affect Economic Activity? Evidence from the Rural Interstate Highway System." Regional Science and Urban Economics 30: 457-490. By Doing, Worker Turnover, and Productivity Dynamics." Working Paper no. 593, Econometric Society 2004 Far Eastern Meeting. Davis, James, and J. Vernon Henderson. 2008. "The Agglomeration of Headquarters." Regional Science and Urban Economics 38 (September): 445-460 Davis, Steven J., John C. Haltiwanger, and Scott Schuh. 1996. Job Creation and Destruction. Chiang, Hyowook. 2004. "Learning Cambridge, MA: MIT Press. Davis, Donald and David E. Weinstein. 2002. "Bones, Bombs, and Break Points: The Geography of Economic Activity." American Economic Review 92 (December): 1269-89. Dumais, Guy, Glenn Ellison, and Edward L. Glaeser. 2002. "Geographic Concentration as a Dynamic Process." Review of Economics and Statistics 84 (May): 193-204. Duranton, Giles, and Diego Puga. 2004. "Micro-Foundations of Urban Agglomeration Economies." In Handbook of Urban and Regional Economics, Vol 4, edited by J. Vernon Henderson and Jacques-Francois Thisse. Amsterdam: North-Holland. Ellison, Glenn, and Edward L. Glaeser. 1997. "Geographic Concentration in U.S. Manufacturing Industries: A Dartboard Approach." Journal of Political Economy 105 (October): 889927. and Edward L. Glaeser. 1999. "The Geographic Concentration of Industry: Does Natural Advantage Explain Agglomeration?" American Economic Review 89 (May): 311-316. Ellison, Glenn, Ellison, Glenn, Edward L. Glaeser, and William 48 Kerr. 2009. "What Causes Industry . Agglomeration? Evidence from Coagglomeration Patterns." American Economic Review forthcoming. Foster, Lucia, John C. Haltiwanger, and C. J. Krizan. 2000. "Aggregate Productivity Growth: Lessons from Microeconomic Evidence." Manuscript, Bureau of the Census. John C. Haltiwanger, and Chad Syverson. 2008. "Reallocation, Firm Turnover, and Efficiency: Selection on Productivity or Profitability?" American Economic Review 98 (March): 394-425. Glaeser, Edward L. 1999. "Learning in Cities." Journal of Urban Economics 46 (September): Foster, Lucia, 254-77. Edward "The Economics of Location-Based Tax Incentives." Working Paper no. 1932, Harvard Institute of Economic Research. Glaeser, Edward L., and Joshua D. Gottlieb. 2008. "The Economics of Place-Making Policies." Working Paper no. 14373 (October), NBER, Cambridge, MA. Glaeser, Edward L., and Janet E. Kohlhase. 2003. "Cities, Regions and the Decline of Transport Costs." Papers in Regional Science 83 (January): 197-228. Glaeser, Edward L., and Joshua D. Gottlieb. 2008. "The Wealth of Cities" Journal of Economic Glaeser, Literature L. 2001. XL VII 4. Greenstone, Michael, and Enrico Moretti. 2004. "Bidding for Industrial Plants: Does Winning a 'Million Dollar Plant' Cambridge, Increase Welfare?" Working Paper no. 9844 (July), NBER, MA. Griliches, Zvi. 1958. "Research Cost and Social Returns: Hybrid Journal of Political Economy 66 (October): 4 1 9-43 1 Griliches, Zvi, and Jacques Mairesse. 1995. "Production Corn and Related Innovations." Functions: the Search for Working Paper no. 5067 (March), NBER, Cambridge, MA. Grossman, Gene M., and Elhanan Helpman. 1991. Innovation and Growth in the Global Economy. Cambridge, MA: MIT Press. Henderson, J. Vernon. 2001. "Urban Scale Economies." In Handbook of Urban Studies, edited by Ronan Paddison. London: SAGE Publications. Henderson, J. Vernon. 2003. "Marshall's Scale Economies." Journal of Urban Economics 55 Identification." (January): 1-28. Henderson, J. Vernon, and Duncan Black. 1999. Economy, 107 (April): 252-284 "A Theory of Urban Growth." Journal of Political Jaffe, Adam B. 1986. "Technological Opportunities and Spillovers of R&D: Evidence from Firm's Patents, Profit and Market Value." American Economic Review 76 (December): 984-1001. Adam Manuel Trajtenberg and Rebecca M. Henderson. 1993. "Geographic Knowledge Spillovers as Evidenced by Patent Citation." Quarterly Journal of Economics 108 (August): 577- 598. Jovanovic, Boyan, and Rafael Rob. 1989. "The Growth and Diffusion of Knowledge." Review of Economic Studies 56 (October): 569-582. Krugman, Paul. 1991a. Geography and Trade. Cambridge, MA: MIT Press. Krugman, Paul. 1991b. "Increasing Returns and Economic Geography." Journal of Political Jaffe, B., Localization of Economy 99 (June): 483-499. Levinsohn, James, and Amil Petrin. 2003. "Estimating Production Functions Using Inputs to Control for Unobservables." Review of Economic Studies 70 (April): 317-342. Jr. 1988. "On the Mechanics of Economic Development." Journal of Monetary Lucas, Robert E. 49 Economics, 22 (July): 3-42. Marshall, Alfred. 1890. Principles of Economics. New York: Macmillan and Co. Moretti, Enrico. 2004a. "Estimating the external return to higher education: Evidence from cross-sectional and longitudinal data." Journal of Econometrics 120 (July-August): 175- 212. Moretti, Enrico. 2004b. "Workers' Education, Spillovers and Productivity: Evidence from Plant- Level Production Functions." American Economic Review 94 (June): 656-690. Moretti, Enrico. 2004c. "Human Regional Economics, Vol Handbook of Urban and Vernon Henderson and Jacques-Francois Thisse. Capital Externalities in Cities." In 4, edited by J. Amsterdam: North-Holland. Moretti, Enrico (forthcoming) "Local Labor Markets", in Handbook of Labor Economics, Ashenfelter and Card Eds. Mowery, David C, and Arvids A. Ziedonis. 2001. "The Geographic Reach of Market and Non- Channels of Technology Transfer: Comparing Citations and Licenses of Working Paper no. 8568 (October), NBER, Cambridge, MA. Olley, Steven G., and Ariel Pakes. 1996. "The Dynamics of Productivity in the Telecommunications Equipment Industry." Econometrica 64 (November): 1263-98. Ottaviano, Gianmarco, and Jacques-Francois Thisse. 2004. "Agglomeration and Economic Geography." In Handbook of Urban and Regional Economics, Vol. 4, edited by J. Vernon Henderson and Jacques-Francois Thisse. Amsterdam: North-Holland. Petrongolo, Barbara, and Christopher A. Pissarides. 2005. "Scale Effects in Markets with Search." The Economic Journal 1 16 (January): 21-44. market University Patents." . Roback, Jennifer. 1982. "Wages, Rents and the Quality of Life." Journal of Political Economy 90 (December): 1257-1278. Rosenbaum, Paul R., and Donald B. Rubin. 1983. "The Central Role of the Propensity Score in Observational Studies for Causal Effects." Biometrika 70 (April): 41-55. Rosenthal, Stuart S., and William C. Strange. 2001. "The Determinants of Agglomeration." Journal of Urban Economics 50 (September): 191-229. Rosenthal, Stuart S., and William C. Strange. 2004. "Evidence on the Nature and Sources of Agglomeration." In Handbook of Urban and Regional Economics, Vol. 4, edited by J. Vernon Henderson and Jacques-Francois Thisse. Amsterdam: North-Holland. Anna Lee. 1994. Regional Advantage: Culture and Competition in Silicon Valley and Route 128. Cambridge, MA: Harvard University Press. Syverson, Chad. 2004a. "Market Structure and Productivity: A Concrete Example." Journal of Political Economy 6 (December): 1 181-1222. Syverson, Chad. 2004b. "Product Substitutability and Productivity Dispersion." Review of Economics and Statistics 2 (May): 534-550. van Biesebroeck, Johannes. 2004. "Robustness of Productivity Estimates." Working Paper no. 10303 (February), NBER, Cambridge, MA. Weiss, Leonard W. 1972. "The Geographic Size of Markets in Manufacturing." The Review of Saxenian, Economics and Statistics 54 (August): 245-257. Zucker, Lynne G., Michael R. Darby, and Marilynn B. Brewer. 1998. "Intellectual Capital and the Birth of U.S. Biotechnology Enterprises." (March): 290-306. 50 Human American Economic Review 88 Table The "Million Dollar Plant" Sample 1. 01 Sample MDP Openings: 1 Across All Industries 47 Within Same 2-digit SIC 16 Across All Industries: Number of Loser Counties per Winner County: 31 1 2+ 16 Reported Year - Matched Year: -2 to 2 20 -1 15 12 to 3 1 Reported Year of MDP Location: 1981-1985 1986-1989 1990-1993 MDP 11 18 18 Characteristics, 5 years after opening: 3 452801 Output ($1000) Output, relative to county output 1 year prior (901690) 0.086 (0.109) Hours of Labor (1000) 2986 (6789) 'M illion Dollar Plant openings that were matched to the Census data and for which there were incumbent plants in both winning and losing counties that are observed in each of the eight years prior to the opening date (the opening date is defined as the earliest of the magazine reported year and the year observed restricted to include matches for which there were incumbent plants in the in the SSEL). This sample Million Dollar Plant's 2-digit SIC is then in both locations. 2 Only a few of these differences are 3. Census confidentiality rules prevent being more specific. original 47 cases, these statistics represent 28 cases. A few very large outlier plants were dropped so that the mean would be more representative of the entire distribution (those dropped had output greater than half of their county's previous output and sometimes much more). Of the remaining cases: most SSEL matches were found in the ASM or CM but not exactly 5 years after the opening date; a couple of SSEL matches in the 2xxx-3xxx SICs were never found in the ASM or CM; and a couple of SSEL matches not found were in the 4xxx SICs. The MDP characteristics are similar for cases identifying the effect within same 2-digit SIC. Standard deviations are reported in parentheses. All monetary amounts are in 2006 U.S. do liars. 3 Of the 51 Table 2. Summary Statistics for Measures of Industry Linkages Mean Only Measure of Industry Linkage Description st 1 Only 4 m Standard All Plants Quartile Quartile Deviation 0.119 0.002 0.317 0.249 0.022 0.001 0.057 0.033 0.022 0.000 0.106 0.084 0.011 0.000 0.042 0.035 0.017 0.000 0.075 0.061 0.042 0.000 0.163 0.139 Labor Market Pooling: CPS Worker Proportion of workers leaving a job Transitions in this industry that MDP move to the industry (15 months later) Technology Spillovers: Citation pattern Percentage of manufactured industry patents that cite patents manufactured Intellectual or Technology in MDP industry R&D flows from MDP Input percentage of all private sector industry, as a technological expenditures Technology Output R&D flows to MDP industry, as a percentage of all original research expenditures Customers and Suppliers: Manufacturing Industry inputs from Proximity to Input as a percentage MDP industry, of its manufacturing inputs Industry output used by Output industry, as a percentage of to Notes: CPS MDP Manufacturing its output manufacturers CPS Worker Transitions was calculated from the frequency of worker industry movements in the rotating by Census Industry codes, matched to 2-digit SIC. The last 5 measures of cross-industry relationships were provided by Ellison, Glaeser, and Kerr (2009). These measures are defined in a 3digit SIC by 3-digit SIC matrix, though much of the variation is at the 2-digit level. In all cases, more positive values indicate a closer relationship between industries. Column reports the mean value of the measure for all incumbent plants matched to their respective MDP. Column 2 reports the mean for the lowest 25% and Column 3 reports the mean for the highest 25%. Column 4 reports the standard deviation across all observations. The sample of plants is all incumbent plants, as described for Table 1, for which each industry linkage measure is available for the incumbent plant and its associated MDP. These statistics are calculated when weighting by the incumbent plant's survey groups. This variation is 1 total value of shipments eight years prior to the MDP opening. 52 s , 30 /— r^ **" c ^ I. . ON NO — o NO — ON CI oo — o — nO r- 00 , VI d d fij ! ~ r«1 2 d *- 53 .ts u I =5 ^- -5 i — * d 5f — M CI oo rsi P-. _;—:>*> d -— — ea i rft eN no "* o" <-n 1 m — o d . J 2 1 '3 ' d d c -o u c C/l D 3 < u oo* - ~ ND o d q d CN O — q p <n <= ,_o u XOS Vl WS 8-S O in o (N oo t-- filial « W3 ^u 30 o cN c s o u m q d rn o d o (N c~. o m 00 CI ON NO in (N ON 00 c ^^ rS-* o 3 o Q t» cN tN o d d d -3 ^ E =j — = <d ea 3 o * q d s u S c I 8 -a i a 1^ d 2 CO „'S8 u e >M > s E u M to o co U 3 I) u fi '§ O 3 3 fe co co '-' eu VI ea 5.^ (N o oo „ ca ea o t~- S C = 3 q d o tN ° s > u o q d in ON r~ in NO .2 in rNi o NO d t-~ NO ON CN t o d o O in On t> d CN *" o NO oo NO o d m t~ S! ! £ 2 2 71 O. _ en ON t^ vi U. t^ NO — f> m "*' fS es f) Cn »t ei co d CU s IT) O u c U _ « > 'S & ?P xJ E .S g a. I c *j co u " u a. u u X o x> q ri oo d NO — NO q «n — *3- -J — o d oo NO d CN in -< E2 H c Si «i I 1 U £ ON in cN _u t/3 cm C u 00 CN +3 'J d" - U IN 1 U u ea ± §1O m CO c C : >H 1/5 " T3 rV m •* rr "1 rs ea eu u eu s e o- .2 OB e "e CO NO o d NO On O d cn 00 NO t^ oo t--~ -3"3- oo .2 c ° C 3 r~ s > u o d (N m q d NO d m CN d >• o q d d " co X o tN 00 q d d CN OO o d o d 5 -5 on G CU 3 U Q ^ T3 n^_ C — o o On o oo NO -^ —" o in ! 'S-S8 r°^i C u q d CN IN 53 x iS o. u i> 2 co .2 fi x ca > S o <u i« "2 a >. Cu u co —J |r. g. > * u .£ T3 x <o £fiw u ,-. u, M T3ox!" 3 k. .2 en as 00 ea s * « S a •J s 3 o CO 1 U C3 U O C B o ' O -2 —o o ccl o ?> — S3 DO [— NO oo c c o x U > o O oo c oo ea J3 3 ft O ft U _o "S. E w o o o NO •u e c <u cO Q. CO U NO X) 3 C3 [jj 5 u X! _ J3 (« ca ,2 ea ta ,eB ea H co - co cN ea c= — ft ^° 3 co o 3 ca o a rn ir> i INT. in CI in u 5)3 o d O d ~ 60 C « o 'c u co ^ "3 3 O U i X a . c- O Do o ° u u =c CO •^ in »— en o d m CN O d I ^o 3 _o " d 1 U 'u s- — -a c J2 o & 53 co c co C/3 S = |a. -ao E u 00 % R .E Xw X o a X>, u 00 On c oo nS P7 .5 c 3 U oo c cU J3 u S "C -S o O ° q lis 151 „ & to Cm o C 3 «o >n -a > o u 00 c o -a I —>»£ X O0 a" ej x r" c co * E u u 5a 3 § 00- o q o c 2 73 : c , 3 O ca au c o u JC, O E S '-E cc 00 a c/i 3 o o ft j. g © o -2 ft S =o co a u .S fi 3 X * •J js a Table Incumbent Plant Productivity, Relative 4. Event Year T = -7 T = -6 T T T T = Winning In Q) (2) (3) 0.067 0.040 0.027 (0.058) (0.053) (0.032) 0.047 0.028 0.018 (0.044) (0.046) (0.023) 0.041 0.021 0.020 (0.036) (0.040) (0.025) -0.003 0.012 -0.015 (0.030) (0.030) (0.024) = -3 = -l T = T= 0.011 -0.013 0.024 (0.022) (0.022) (0.021) -0.003 0.001 -0.005 (0.027) (0.011) (0.028) -2 0.013 -0.010 0.023 (0.018) (0.011) (0.019) 0.023 -0.028 0.051* (0.026) (0.024) (0.023) 1 T=2 T T x = =4 = 0.004 -0.046 0.050 (0.036) (0.046) (0.033) 0.003 -0.073 0.076+ (0.047) (0.057) (0.043) 3 0.004 -0.072 0.076* (0.053) (0.062) (0.033) ' -0.023 -0.100 0.077* (0.069) (0.067) (0.035) 5 R-squared 0.9861 Observations 28732 Notes: Standard errors are clustered regression: machinery The natural log of output capital, materials), dummy variables When a plant is at of a MDP Opening Columns the county level. 1 and 2 report coefficients from the same (all worker hours, building capital, regressed on the natural log of inputs is year x 2-digit SIC fixed effects, plant fixed effects, case fixed effects, and the reported whether the plant for Ye ar (l)-(2) Counties -5 T to the Difference Counties = -4 = Losing In a winner or loser is in a winning or losing county more than once, it receives a in each year relative to the dummy observations are weighted by the plant's total value of shipments eight years prior to the plants in all cases is MDP incumbent plants at is the 5% level; opening. MDP opening. Data on were method from early book values and subsequent investment. The sample of Columns 1 - 2 of Table 3. ** denotes significance at 1% level; * denotes only available 8 years prior to the calculated using the permanent inventory significance MDP variable for each incident. Plant-year same as in + denotes significance at 1 0% level. 54 opening and 5 years after. Capital stocks Table Changes 5. in Incumbent Plant Productivity Following mdpw:inners MDP Losers (1) Model Mean a MDP Opening MDP Counties MDPW:inners MDP Losers All Counties (2) All Counties Random Winners (4) (3) (5) 1: 0.0442+ Shift 0.0435+ (0.0233) (0.0235) 0.0524* 0.0477* -0.0496** (0.0225) (0.0231) (0.0174) [$170m] R-squared 0.9811 0.9812 0.9812 0.9860 -0.98 Observations 418064 418064 50842 28732 -400000 (plant x year) Model 2: 0.1301* 0.1324* (0.0533) (0.0529) Effect after 5 years 0.1355** (0.0477) 0.1203* -0.0296 (0.0517) (0.0434) [$429m] Level Change 0.0277 0.0251 0.0255 0.0290 0.0073 (0.0241) (0.0221) (0.0186) (0.0210) (0.0223) 0.0171 + 0.0179* 0.0183* (0.0091) (0.0088) (0.0078) Trend Break Pre-trend - 0.0057 - 0.0062 (0.0063) 0.0058 -0.0048 - 0.0044 -0.0048 (0.0046) (0.0046) (0.0044) (0.0040) - (0.0046) 0.0152+ (0.0079) R-squared 0.9811 0.9812 0.9813 0.9861 -0.98 Observations 418064 418064 50842 28732 -400000 YES YES YES YES YES YES YES YES N/A (plant x year) & Ind-Year FEs Case FEs Plant NO Years Included All All Notes: The table reports results from fitting several -7 All < x < All 5 versions of equation (8). Specifically, entries are from a regression of the natural log of output on the natural log of inputs, year x 2-digit SIC fixed effects, plant fixed effects, and case fixed effects. In Model 1, dummy two additional MDP variables are included for whether the plant is in a winning county 7 to 1 The reported mean shift indicates the difference in these two coefficients, i.e:, in TFP following the opening. In Model 2, the same two dummy variables are included along with pre- and post-trend variables. The shift in level and trend are reported, along with the pre-trend and the total effect evaluated after 5 years. In Columns (1), (2), and (5), the sample is composed of all manufacturing plants in the ASM that report data for 14 consecutive years, excluding all plants owned by the MDP firm. In these models, additional control variables are included for the event years outside the range from x = -7 through x = 5 (i.e., -20 to -8 opening or the average change years before the and 6 to 17). Columns (3) Column and (2) (4), the to 5 years after. adds the case fixed effects that equal sample is 1 during the period that restricted to include only plants in counties that industry-year fixed effects to be estimated solely from plants in these counties. For won x ranges from -7 through or lost a Column MDP. (4), the 5. In This forces the sample is restricted further to include only plant-year observations within the period of interest (where x ranges from -7 to 5). This forces the industry-year fixed effects to be estimated solely on plant-by-year observations that identify the parameters of interest. In Column (5), a set industries as the of 47 plant openings MDP those estimates). For openings all (this in the entire country were randomly chosen from the ASM in the same years and procedure was run 1,000 times and reported are the mean and standard deviation of regressions, plant-year observations are weighted by the plant's total value of shipments eight years prior to the opening. Plants not two uncommon in a winning or losing county are weighted by their total value of shipments in that SIC values were excluded so that estimated clustered variance-covariance matrices would always be positive definite. In brackets is the value in 2006 U.S. dollars from the estimated increase in productivity: The percent increase is multiplied by the total value of output for the affected incumbent plants in the winning counties. Reported in parentheses are standard errors clustered at the county level. ** denotes significance at 1% year. All plants from level; * denotes significance at 2-digit 5% level; + denotes significance at 55 10% level. Table Model 6. 1 : Changes Mean in Incumbent Plant Output and Inputs Following a Output Worker Hours Machinery Building Capital Capital (1) (2) (3) (4) 0.1200** Shift (0.0354) Model 2: After 5 years (in logs). 0.0401 (0.0348) 0.0826+ 0.0562 (0.0478) (0.0469) Notes: The table reports results from outcome variables 0.0789* (0.0357) See the fitting text for at 10% 0.0911** 0.0089 (0.0691) - (0.0302) 0.0077 0.0509 (0.0541) (0.0375) versions of equation (8) for each of the indicated more Standard errors clustered details. are reported in parentheses. ** denotes significance at denotes significance - (5) 0.1327+ (0.0300) MDP O pening Materials 1% level; * level. 56 at the county level denotes significance at 5% level; + : 7. Changes Incumbent Plants Table in in Incumbent Plant Productivity Following a MDP Opening the MDP's 2-Digit Industry and All Other Ind ustries All Industries (1) Model Mean MDP's All Other 2-digit Industry 2-Digit Industries (2) 01 1 0.0477* 0.1700* (0.0231) (0.0743) (0.0253) [$170m] [$102m] [$104m] Shift 0.0326 R- squared 0.9860 0.9861 Observations 28732 28732 Model 2: 0.3289 (0.2684) (0.0504) [$429m] [$197m] [$283m] Level Change 0.2814** 0.0290 (0.0210) Trend Break 0.0152+ 0.0044 -0.0174 (0.0265) R-squared 0.9861 Observations 28732 Notes: The table reports results from 0.0147+ 0.0079 (0.0081) (0.0344) (0.0044) - 0.0004 (0.0171) (0.0895) (0.0079) Pre-trend 0.0889+ 0.1203* (0.0517) Effect after 5 years - all industries 0.0026 (0.0036) 0.9862 28732 fitting versions of equation (8). estimates from the baseline specification for incumbent plants plants in for As a basis for comparison, in all industries Column (baseline estimates for 1 reports incumbent 5). Co lumns 2 and 3 report estimates from a single regression, which and pre/post variables with indicators for whether the incumbent plant is in the same or a different industry. Reported in parentheses are standard errors clustered at the (Column 4 of Table fully interacts the winner/loser 2-digit industry as the county level. MDP The numbers in brackets are the value (2006 U.S.S ) from the estimated increase in productivity: The percent increase is multiplied by the total value of output for the affected incumbent plants in the winning counties. ** denotes significance at 1% level; * denotes significance at 5% level; + denotes significance at 10% level. 57 . . i ^O -TJ U u -* _I "* c c E — "^ u I— aj a .2 j= O 1* a T3 ^| 9 3 CN 1- o nc frn (N © q d d no 00 <n tN (N o q q d © -* ^ CN o •* m oq d d O VI X) C3 M C S— u ca C ^f -t ro ** o o q q d d ts> >» O O 1— fl H a '— O " "0 J= > OJ in rt £ u 3 73 TO s 73 TT + O O C3 £ § o -5 u T3 O O 3 3 C 73 IT, >Q on J* ON CO 1- 3- l* •n ON t2 n. ro -a SS ON r>-> u -C iN a. g .o z 8 no o in on o © d d eft U ii u a c O u c CL >> . o a 'it J3 B-S -3- l_ D. ,0 U c X ^o '— '— 2u o C 3 -C C C JS IS U O i„ o E 3 -J O c c 00 O -3 5 ?3 "J C w u U "u TO i_ ca — 3 3 1 3 3 H CO O NO (N — O o o d d to C * a «n on f> 2? On r«i o ™ m o on On m © <N CN NO NO os — o o d d 00 + — o — * rn rN r^ ro "I ON <^ s? On ro <N q q d d O in (N "* ON >n on en 2? on o o d d © m tN <N ron f> 2? on m © m 3 3 D. a. c 1 >-. a. o c C a DO u H B- 3 O 00 c 3o u t9 ^-i C3 3 u U D. 00 "3 o 3 C Bl rs 3 o" T Od x) O 2 E Number Table 9. Changes in Counties' Following a MDP Opening of Plants, Total Output, and Skill-Adjusted Panel Panel 2 (Census of Population) 1 (Census of Manufactures) Dependent Variable: Dependent Variable: Dependent Variable: Log(Plants) Log(Total Output) Log(Wage) (JJ (2) (3) 0.1255* Difference-in-Difference (0.0900) 0.9984 0.9931 0.3623 209 209 1057999 Observations Notes: The table reports results from number of establishments and 0.0268+ 0.1454 (0.0550) R-squared fitting three regressions. In Wages Panel (0.0139) 1, of Census of the dependent variables are the log the log of total manufacturing output in the county, based on data from the Manufactures. Controls include county, year, and case fixed effects. Reported are the county-level difference-in- MDP opening. Because data are available every 5 years, depending on the Census MDP opening, the sample years are defined to be - 5 years before the MDP opening and 4-8 years after the MDP opening. Thus, each MDP opening is associated with one earlier date and one later date. The difference estimates for receiving a year relative to the Column (1) model 1 is weighted by the number of plants weighted by the county's total manufacturing output In Panel 2, the dependent variable is log in in the county in years -6 to -10 and the wage and controls include education*year, sex*race*Hispanic*citizen, and case fixed effects. Reported difference estimate for receiving a defined to be opening is - 1 dummies is for in the private sector. individual individual the significance at may 1, The sample in all individuals in a county group were more than one FIPS and observations are parentheses are standard errors clustered level; * denotes significance at who worked more not in school, are at work, and who unique individuals - some IPUMS is restricted to work more than 20 hours per week, are The number of observations reported refers to then appear in same weight. Reported 1% is MDP opening. Because data are available every 10 years, the sample years are MDP opening and 3-12 years after the MDP opening. As in Panel each MDP county groups include more than one FIPS, so The same model for age*year, age-squared *year, 10 years before the associated with one earlier date and one later date. wages (2) the county-level difference-in- than 26 weeks in the previous year, usually work Column years -6 to -10. 5% level; + denotes 59 individuals matched weighted at to each potential to give FIPS. each unique the county level. ** denotes significance at 10% level. 5 ^ '' . CD J2 mo +-» .. — H Cm 5 3 ca te a o u. a g; 73 .- C fO ja iw * — o |£ S IS C CO 5 g 3 — ^ CO r- ^j [fl 5! -T3 =; Hrt co =^^ S3 ^t ,-i S3 u -a ca co 22 Si -a 2»oS JS '3 fc 1 Q. ' 5/5 CD S u u sop p 73 .« N— £ a 5 3 vo >* o q q © o t-~ — O o (N o P 2in ca (D o. ^ CO 5 IS u P » a h S o -** J] iio s -** bfl 3 o — + ON VI \o on 3 w- — m o o CM °' T3 2- a ^^ * o o 5 *jM.2c d d^ c ^ .2 OO "5t cd Si It^ 3 tl| co ^ o + cn s a * ti. '5 u .2 £P in S c. CS >- O c H ^^ « at •* CN o o <=2 * CO 1 i- o a..—. g-.s o © 3 a a •— <-, ~^ in p- -aU-1 (N q q d d a + £> .2 ^o Ij £ -21 *> ca ^^ ^ o O o o d 2, <N Tt CN "** o c •a .2 £ s u a HH _C * 0£>"c3 ha B W — in d d o5 _ CJ\, C a .2 2 S- ^- * 'co 2 o a- c S <D U o « o + M oo m 1 " -2 .2 — in — o 1 o g 2 d d S 1i1 r^. r< 73 < S w -C * c ^^ S r*- -S. * 7: .SP CD 3 oo .9 ora.Sg q d dq s 1 § E CD ca i- c u .2 c 3 .S "3 ,2 w cd 'q ."til m u — ,— + ^ r- tN * o o CN d d in rn —— * m o cm dq d * * , i> r~ — o dd ^ 0. > . g, CD CO l-l u 'Ja in CD § co 1 jg S ,3 CD "- ^3 t~- CD CD J g „ '5 E 73 is .0 o — — o 1 g ° 2 t" B 3 73c to <C SS e S c ji U * a|« X * 2 = c C r C CD __ £3 • to U CD •1 •- s C u co n 5 „ "r^.2 g " u c PL, a «J o CD CO fe ^ \C g-.S s u s o o C ^ .£ T3 cd ca § « S 8 -s c 3 -8 x 13 ;. | fe c CuO a » 00 O BO 3 CD H .£ "3 O O S. 5 w c c S O T g JS U o S ~ S3 H (N 73 jj 3 < o 2 ao 2 .. 3 .5 ? c c -a cn H » £ CD 3 "3 " JJ S >. >, -a CD —g £ C O C ?? o i ' ' i i o a .3 2 u C «— o 3 a —c U s c oj to m Q\ — — r~ m rCN o\ — oc r^ O r) — o S o o O c r- 'J- if) cr- OO 0000(NO-*Or<i do d<6d>d>d><6dd>™ H L~ a J3 ^- > J U ' in —>n Ocn'ONDO-^fcncncON VO N CN t^ 00 CI — "* CN OO oo r^- rj\ t^ oc oo 00 (-^ •— -1o> CN -I- CI * m O * P oo oooooooo CN r-i (N C XJ o fa. OJD ^o.x>9 9 « rt «H 3 e. 3 o oo * o o o o 00 O o o d d - a " o\ ITS *sO PI CM oo "~ u D. c — —3 C3 o o o d d cd L- 5 B ^ BX "B O 3 ca oj l> id *-» — I— 3 O U CN *- O d 3) ON O no * CN On \n oo r- u-i CN CM 00 "3" 5 a- CO u J2 NOtOoo d d d d — NO (i"> u £> 3 •— — > 4^ cj 5 i- en 3 ea 83 s W IS <M u OJ *J to u. o ca 3 D. Im OJ Oj CO a CJ Si u -" < 2 H _o cd 3 O '3 3. ir. CJ u Oil en Uq J3 5. o "^ 'J o o o + •• ^u ^ •j c c ,_3 u CJ N ca NO U. T3 U 3 o 3 CJ ca •S X XO ca cj c 2 ^3 OJ > CJ oil CO CJ t_ OJ U > OJ s^ — p -a OJ M tf^ CJ o u X au —m w 13u o o '3 £ 00 2 x " iS 3 3 U O.TJ c X 3 3 J a u — Cl ^ 5 z u d x-l I -- 'I CJ cj : OJ '" >r> oj i. — * CN Mi^r^ — -«ONO-^rn OOOOCNO^Ooo P P d d dddododoM — Ov o On in -t O — cn o fOn s0 o~ <* rj O0 — — r~ — o^ rl in n m — >n O q q q o q CN q d d d d d d d d d . C*l "3. 3 .£ OJ 'fcH U i -3- Q 00 3 n. ca oj o U 3 O (U d> 3 OJ r 3 o 3 O C ° -9 a. ca n. ^ ca Q. r3 D. IS _ , <c IS 00 u c 3 T3 S3 22 ca u U 00 c is '3 m U a. tyi C c u s D. "3 a- W (/I 1— o .2 o "ca (U (N ^ u OJ XI O 2 < C3 U oo c 2 '3 CQ Q. cd U -~ cj u S 8 H _cS "H — W o o x .E O ca -J ?o *I 3 o OJ 'x: CJ IS Sa ca OJ w oo o OJ X) CO OJ c u e '3 a- - x> s oi i_ o T3 i o « : > oj >>~ ' oo VI a w o o CO > On vd t> CO >o X y ~ c .= is CQ 3 E 3 00 ca u oj oj m o o d d _c j= ft- CO ' ca ^- oo no cn cn E. 3 a" U a -t— CO 3 8> 3 3 _J — O o" 3 E* 3 J5 btf 1 = o 3 cj c« 3 or o 3 3 Cfl oo OJ .2 vx •™ U •— — m CN -* m o CN o d d d o 3 © *— 3 CO ci U CN ca CO 3 3 +* cj Bcd ~C- X r. BO < ft, n O oj _£3 CO 3 .3 ca <" oj "2 ™ x> o CO q '55 73 CQ a =3 & r - O ^o CN oo o OO o — m cn 00 o o d d -a « 1 ca > co CJ -2 =3 cu "o o tK <M bx S Q. E ca i= u —O 3 1— c TO .s a cS 3 S S * -i GO s 'S J3 _o 2.2 £ ca u 3 C ,0 J c + ^~. — (N On CN m o o cn d d § i> 3 cd C/} o •xi c -O J3 C*l o £ a. "a CN 00 O o * — o d d On f» oo cn <N C3 l. s & c C a .2 1.9 O Bc s o * J u cc SO U u ^ '3 00 S 3 *s 8 o ca oo JN On i-s S a o U bo — o o d d ON VD On C3 ^- r3 co ^ u 3 ^ ^ ex ^ -*. '3 ft. -; + *$ 3 O Xu U 3 CL o w C C/5 _S '3 a. »» U c S 2 c3 a, c 3 3 ^—^ :.:. —B cj as Z oo yi CO U -^o -au c 3 c o Q. oo E a <u U S" o b) c u bfl cd J3 x: > CI " I u oj 3 CO 00 o u c 33 a to x: - o " oj5 ON ea ft. ^§1o . CU s o u 3 "3 O - S X a oj >. -a c CN ca 3 ca da "2 'E — u cj "ca cS ,S E oj x ^j d T3 OJ CJ to 1^" 16 .3 CJ Incumbent Plants' Productivity the Year of a MDP Opening Figure 1. All All Industries: Difference: in Winning Winners vs. vs. Losing Counties, Relative to Losers Winners - Losers 0.1 -K 0.05 - -7 -0.05 -6 J Year, relative to opening Notes: These figures accompany Table 4. 62 X Figure 2. Distribution of Case-Specific Mean Shift Estimates, Following a MDP Opening 0.5 0.4 0.3 0.2 9-1 W ww ffM^ \u \\\\\- ] - n 45 -0.2 -0.3 -0.4 -0.5 Notes: The figure reports results from a version of Model cases. T he figure reports only 1 that estimates the 45 estimates because two cases were excluded 63 parameter for 6] for each of the 47 MDP Census confidentiality reasons. Figure 3. Incumbent Plants' Productivity in the MDP's 2-Digit Industry, Winning Losing Counties, Relative to the Year of a MDP Opening 2-digit 0.2 - 0.1 - vs. MDP Industry: Winners Vs. Losers A \><r - -7 -0.1 - -0.2 - -6 -5 ,•'--..._, a... '"'•. •' ^"^•\$ /~'i- -2 / A\ .-• A-- "'•£•'" 1 Year, relative to opening —— Winning Counties ---a--- Losing Counties Difference (Winners - Losers) Notes: These figures accompany Table ''•.. 7, Column 2 (MDP's 64 2-digit Industry). 2 3 4 ^\§ Figure 4. Incumbent Plants' Productivity in Other Industries (not the MDP's 2-Digit Winning vs. Losing Counties, Relative to the Year of a MDP Opening Industry), Other Industries: Winners Vs. Losers Year, relative to opening Winning Counties - a- • • Losing Counties Difference (Winners - Losers) 12 3 4 -0.05 Year, relative to opening Notes: These figures accompany Table 7, Column 3 (All 2-digit Industries, 65 except the MDP's 2-digit Industry). 5