The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Federal Reserve Bank of Chicago Evolving Agglomeration in the U.S. auto supplier industry Thomas Klier and Daniel P. McMillen WP 2006-20 1 Evolving Agglomeration in the U.S. auto supplier industry Thomas Klier Federal Reserve Bank of Chicago Research Department 230 S. LaSalle St. Chicago, IL 60604 312-322-5762 tklier@frbchi.org Daniel P. McMillen Department of Economics (MC 144) University of Illinois at Chicago 601 S. Morgan St. Chicago, IL 60607 312-413-2100 mcmillen@uic.edu JEL codes: R30, R15, L62 Key words: Spatial econometrics, non-parametric statistics, agglomeration, plant location, auto supplier industry Abstract Using nonparametric descriptive tools developed by Duranton and Overman (2005), we show that both new and old auto supplier plants are highly concentrated in the eastern United States. Conditional logit models imply that much of this concentration can be explained parametrically by distance from Detroit, proximity to assembly plants, and access to the interstate highway system. New plants are more likely to be located in zip codes that are close to existing supplier plants. However, the degree of clustering observed is still greater than implied by the logit estimates. The authors thank Cole Bolton and Paul Ma for excellent research assistance. 2 1. Introduction The North American automobile industry has been remarkably concentrated since its inception. Assembly operations are characterized by significant scale economies in production. Only a small number of assembly plants are required to serve the entire continent, and these plants tend to be located in the center of the country. Although prominent exceptions to this rule were once operating on both the east and west coasts, many of these outlying plants have been closed in recent years as the industry has retrenched toward the middle of the U.S. and lower Ontario. These trends, which are documented in Rubenstein (1992) and Klier and McMillen (2006), have been accompanied by changes in the geographic distribution of auto supplier plants. Though supplier plants are often part of comparatively small firms, their operations are also subject to internal scale economies. A supplier plant may serve several assembly plants. Moreover, the rise of just-in-time inventory practices has increased the incentive for suppliers to locate close to assemblers. Supplier plants thus tend to cluster near assemblers, and suppliers too have re-trenched toward the center of the country in recent years. Maps of assembly and supplier operations show a growing concentration of auto suppliers along an axis running southward from Detroit. Whereas the industry once was concentrated in a corridor running from Chicago to New York, it now has a north-south orientation. In this paper, we use both parametric and nonparametric techniques to document the changing geographic structure of the American auto supplier industry. We focus on suppliers rather than assemblers because their much larger number makes them more amenable to statistical analysis. Of the 2,627 supplier plants in our dataset, 431 opened 3 after 1990. Using a nonparametric approach developed by Duranton and Overman (2005), we begin by documenting the degree of localization exhibited by this industry. Both new and old supplier plants are far more concentrated than would be expected by pure randomness, and this result holds whether we define randomness as an equal chance that a plant might locate in any zip code in the eastern U.S or we weight the probability by the level of employment in the zip code. This descriptive analysis suggests that the geographical distribution of new and old plants is remarkably similar given the amount of change undergone by the auto industry during this time. The next step in our analysis is a parametric investigation of the determinants of auto supplier locations. Using zip codes as the underlying geographic unit, we present conditional logit estimates of the location decisions of new and old plants. We find that both new and old plants are more likely to be located in zip codes that are near assembly plants, close to Detroit, and are served by interstate highways. In addition, we find that new plants are more likely to be located in zip codes that are close to existing supplier plant locations. The changing geographic orientation of the industry is evident in the conditional logit estimates: new plants are more likely than existing plants to locate in the East South Central region. However, the similarities are more striking than the differences. As new plants open in the southern United States, they tend to follow a location pattern similar to the plants that have preceded them in the region. In the final step of our analysis, we use the predicted probabilities from the conditional logit models as the base for the Duranton and Overman (2005) measure of concentration. We find that actual plant locations are even more concentrated than implied by the conditional logit estimates. However, new plant locations are not more 4 concentrated than would be implied by a simple random choice from existing plant locations. This result reinforces our finding that new plants follow a location pattern similar to existing plants. We also find that Duranton and Overman’s (2005) nonparametric procedure is useful as a diagnostic tool: the conditional logit models, while apparently fitting the data well, fail to account adequately for the degree of clustering exhibited in practice. Whether the focus is on new or old plants, our results portray a highly clustered auto supply industry. Plants opening after 1990 are more likely than older plants to locate along an axis running south from Detroit. But both new and old plants are highly concentrated, locating close to assembly plants, near highways, and near other supplier plants. Although the geographic orientation has moved south, the industry is neither more nor less concentrated now than prior to 1990. 5 2. The U.S. Auto Industry In the 1890s, during the beginnings of the U.S. auto industry, more than half of the producers of automobiles were located in the northeast between Philadelphia and Boston. 1 Soon afterwards, during the first decade of the twentieth century, southeastern Michigan emerged as the hub of auto production in the U.S. It attracted or retained the most successful motor vehicle producers because many of the industries from which automotive technology is derived, such as the production of engines and carriages, were already thriving in the region. 2 Subsequently, automakers and suppliers could tap into a rich pool of skilled mechanics and engineers. According to the 1904 Census of manufacturers, 42% of all cars were made in Michigan, as the industry’s leading producers and their major facilities were based in Michigan by then. Over time the location of auto assembly and auto parts plants evolved differently. The Ford Motor Company developed a system of branch assembly plants which was quickly copied by the other major producers of vehicles. It was based on the fact that auto assembly is a classic weight-gaining industry: it is cheaper to produce finished vehicles near the centers of population than to ship finished vehicles from a central location to many destinations across the country. Motor vehicle parts, on the other hand, continued to be produced in the Midwest and then shipped to the various assembly plants located across the country. A quickly growing industry was well-suited for a branch assembly plant system as production runs for the best-selling vehicles were large enough to support 1 This section draws heavily on Rubenstein (1992) Bicycle manufacturing, the third major contributor to the early development of the automobile, proved the exception as the country’s largest bicycle manufacturers were located in the Northeast. According to Rubenstein (1992), bicycle manufacturers contributed to the emergence of southeastern Michigan as the industry’s hub by failing to recognize the automobile’s potential and thereby losing their early technological lead in the face of rapid technological innovation. 2 6 more than one assembly plant. This combination of decentralized assembly plants combined with the concentration of parts production in Michigan and its neighboring states of Indiana and Ohio continued until the 1980s. The forces leading to a restructuring of the auto industry geography began during the 1960s. In response to increased sales of smaller cars by foreign producers, U.S. producers introduced a number of smaller platforms over the years, e.g., “compact,” “intermediate,” and “subcompact” cars. As a result the growth of product variety outpaced the growth of overall demand, leading to substantially smaller production runs, even for the best-selling models. Subsequently no individual model sold enough to justify production at more than one, or at most two, assembly facilities. This development led to a re-concentration in the geography of auto production. In conjunction with the recessions induced by the 1970s oil crisis and an increase in motor vehicle imports, domestic auto producers reduced capacity and shut down some of their production facilities. Specifically, assembly plants located on the coasts were increasingly abandoned in favor of locations in the center of the country. As a result the location of assembly plants began to re-concentrate in the Midwest. Starting in the early 1980s, foreign producers began producing vehicles in the United States. 3 They strongly preferred locations in the interior of the country. Yet the foreign producers extended the auto region to the south by opening plants in Kentucky and Tennessee, and most recently as far south as Mississippi and Alabama (see Klier and McMillen 2006). 3 The exception is Volkswagen, which started producing cars in Westmoreland, PA, in 1978. The company’s spell of producing cars in the U.S. did not last very long. That plant closed in 1989. 7 The auto industry has experienced a significant southward extension even as the Midwest re-emerged as the center of vehicle production after the demise of the branch plant system in vehicle assembly. This movement southward has been driven primarily by the location of foreign-owned assembly plants during the 1980s and 1990s. Incidentally, most of these plants are located at greenfield sites, some distance from traditional manufacturing locations. Today the preferred locations for motor vehicle assembly are defined by a north-south region that is often referred to as the I65 – I75 corridor, as it is rather well defined by two of the major north-south interstate highways, extending south from Michigan to Tennessee and beyond. Thus, North American auto supplier plants have been remarkably concentrated for a long time (Klier and McMillen 2006). When the industry got its start just over 100 years ago, raw materials and worker skills available in the upper Midwest, between Chicago and Buffalo, furthered the development of this industry. Auto suppliers remained concentrated in the upper Midwest during the branch (assembly) plant era, as it was cheaper to ship parts than finished vehicles from a central location. During the early 1980s the U.S. auto industry was shaped by the arrival of foreign producers who brought with them the Just-in-time production system as well as a substantial number of foreign suppliers. The 1980s also witnessed the emergence of the auto corridor, a region extending south from Detroit into Kentucky and Tennessee, with fingers reaching into Mexico and Canada. During this time new parts plants showed a tendency to locate farther south, reinforcing the north-south orientation of the auto region. 8 This brief overview of the geography of the U.S. auto industry shows a longclustered industry that now remains highly clustered after a recent major re-orientation southward. 4 The spatial concentration of today’s industry (Ellison and Glaeser 1997) is reinforced by tightly linked supply chains that require most suppliers to be within a day’s shipping distance of their assembly plant customers. Figures 1 and 2 illustrate the changing geography of auto supplier plants. Both maps are based on the data used later in our statistical models. The maps focus on the eastern half of the U.S., where the vast majority of plants producing auto parts destined for vehicle assembly are located. Figure 1 shows the distribution of “old” auto supplier plants. The most densely populated zip codes define a north-south auto region, with southern Michigan and Indianan and Ohio as its hub. Yet the industry covers a much larger area as its plants are well represented in almost every state on the map. Though Figure 2 is based on a much smaller number of plants (1/5 of the number of plants represented in Figure 1), it clearly illustrates the formation of a rather well-defined auto region that extends south from Michigan to northern Alabama and Georgia and reaching into the Carolinas. These maps clearly show that auto supplier plants that opened between 1991 and 2003 re-trenched toward the center of the country. 4 Klier and McMillen (2006) trace in some detail the re-orientation of the auto industry geography by comparing location choices for assembly and supplier plants during the 1980s and 1990s. They also compare the location patterns of domestic and foreign plants. Woodward (1992) and Smith and Florida (1994) find evidence that vertical linkages as well as the presence of highway infrastructure influence plant location decision of Japanese plants in the United States. 9 3. Data A Michigan-based vendor, ELM International, provided the primary data for our analysis. Though the ELM database covers the entire North American auto industry, we limit our analysis to the eastern United States. We include states that border the western bank of the Mississippi River in order to include large concentrations of plants in places such as St. Louis, Dallas – Fort Worth, and Minneapolis – St. Paul. Very few plants exist between this line of states and the West Coast. The 31 states represented in our definition of the eastern United States form a reasonably compact and integrated economic area. The ELM database includes data at the plant and company level. However, plants that produce machine tools or raw materials and those that produce primarily for the aftermarket are not part of the database. The data include information on “captive” supplier plants, which are parts operations that assemblers own and operate themselves, such as engine and stamping facilities. The database includes information on a plant’s address, products, employment, parts produced, customer(s), union status, as well as square footage. Records were crosschecked with state manufacturing directories to obtain information on the plant’s age, and information on captive plants was obtained from Harbour (2003). We then geocoded the data to the zip code level, and verified the accuracy of the data whenever possible by checking individual company’s websites and through phone calls. The dataset includes data for 3,319 supplier plants in the eastern United States. Of these plants, 431 are “new”, which we define as having opened since 1991. We dropped 692 observations with missing data on plant age. We refer to the remaining 2,196 observations, which began operation before 1991, as “existing” or “old” plants. 10 Since the dataset is cross-sectional in nature, the age variable applies only to surviving establishments. Although this focus on survivors may lead us to understate the extent to which “old” plants are concentrated near Detroit, it provides an accurate view of the geographic distribution of new plants and it allows us to test whether the distribution differs from that of surviving older plants. Using 1991 as the starting date for new plants allows us to determine whether the major changes undergone by the American auto industry in the 1980s fundamentally altered the geographic distribution of the industry. A further advantage of focusing on plant openings from after 1990 is that it allows us to match the plant openings with explanatory variables from the 1990 U.S. Census. Moving the date forward by one year from the time of the census ensures that these explanatory variables can be taken as exogenous. Table 1 presents descriptive statistics for the variables used in our analysis. Separate sets of statistics are presented for the new and old plant samples. In addition, we present descriptive statistics for samples of randomly chosen alternative locations. These alternative locations comprise the rejected alternatives for our conditional logit models. To identify these alternatives, we match each plant with five randomly chosen zip codes that (1) are different from the plant’s actual zip code, and (2) are different from each other. Candidate alternatives include any zip code in the eastern United States, including those with neither a new nor old supplier plant. The alternative for one plant may include a zip code that already has another plant. Table 1 includes descriptive statistics for explanatory variables for the conditional logit models. Having an interstate highway run through a zip code increases the 11 likelihood of having a plant, and the effect is stronger for new plants. Zip codes with either new or old plants are more likely than randomly chosen alternatives to be near assembler plants. Variables drawn from the 1990 U.S. Census include population density, the proportion of the zip code’s white population, the proportion who have graduated from high school, and the proportion who work in manufacturing jobs. We also include regional dummy variables and a variable indicating whether the zip code is located in a metropolitan area. Finally, we include a variable measuring the distance in from Detroit. Plants are much more likely to be located close to Detroit and in the base region, the East North Central region. 4. The Geographic Distribution of Supplier Plants In this section, we use the methodology developed by Duranton and Overman (2005) to compare the geographic concentration of existing and new auto supplier plants in the eastern United States. Our dataset is geocoded down to the zip code level. Using the geographic coordinates, we begin by calculating the distance between every pair of plants. With n plants, there are n(n-1) distance pairs. Using a standard kernel density function (Silverman, 1986), we can calculate the density of bilateral distances at any target distance d as: K (d ) = 2 n(n − 1)h n −1 n ⎛ d − d i, j f⎜ ⎜ h i =1 j = i +1 ⎝ ∑ ∑ ⎞ ⎟ ⎟ ⎠ (1) where di,j is the distance between observations i and j, h is the bandwidth, and f is the kernel function. As in Duranton and Overman (2006), we use a standard Gaussian kernel 12 with an optimal bandwidth. 5 All distances are measured in straight-line miles. Following Duranton and Overman, we refer to the estimated functions as K-densities. We calculate separate K-densities for new and existing plants. We calculate equation (1) at 40 evenly spaced target points between d=0 and d=800. The results are shown in Figure 3. The striking feature of Figure 3 is the similarity between the estimated densities. Both density functions have twin peaks at distances of about 135 miles and 250 miles. The densities rise rapidly to the first peak and trail off slowly at distances beyond 250 miles. The most common distances between plants are in the range of about 100-300 miles. Given the size of the eastern United States, these distances are not small. Most importantly, the distribution of distances between plants has not changed significantly since 1991. Plants are not substantially closer to one another now than they were before 1991. Although Figure 3 shows that the K-densities are similar for new and old plants, it does not show directly whether the auto supplier industry is heavily concentrated. Measuring geographic concentration requires a base model of possible locations. To measure concentration, Duranton and Overman (2006) compare actual K-densities to the density that would be expected if plants were located randomly across space. Using a different but related approach, Ellison and Glaeser (1997) compare actual locations to the expectation if plants were assigned to locations based on the an area’s share of total manufacturing employment. In this section, we use three base models of possible locations to measure geographic concentration. In the first model, the probability that a zip code is chosen as a 5 To calculate the optimal bandwidth, we first calculate the standard deviation (s) of the n(n-1) bilateral distances. Following Silverman (1986), the optimal bandwidth for a Gaussian kernel is 1.06sn-.2. 13 plant location is pi = 1/nz , where nz is the number of zip codes. In the second model, the probability for zip code i is pi = Ei/(ΣiEi), where Ei represents total employment in zip code i. Analogously to Ellison and Glaeser (1997), the probabilities in the third model are based on the share of total manufacturing employment – pi = Emi/(ΣiEmi), where Emi represents manufacturing employment in zip code i. Of the 28,036 zip codes in the eastern United States, 19,506 have some employment while 19,151 have some manufacturing employment. After assigning a probability of pi to each zip code, we make n draws with replacement from the set of zip codes to construct our base model set of locations. We then calculate the distance between every actual plant location and the randomly drawn set of replacement locations. We then re-calculate the K-density as: K (d ) = 1 n2h n n ⎛ d − d i, j ⎞ ⎟ ⎟ h ⎠ i =1 j =1 ∑ ∑ f⎜ ⎜ ⎝ (2) where di,j denotes the distance between the actual plant location i and the randomly drawn location j. There are now n2 distances to calculate – n base plant locations and n zip code draws. However, we maintain h at the value used in equation (1) to keep the level of smoothing at the same level as before. Following Duranton and Overman (2006), we draw 1,000 set of plant locations and re-calculate the equation (2) K-density. We then order the K-densities from smallest to largest at each distance d, and calculate the implied 95% bootstrap confidence interval by choosing the 25th and 975th largest values. Figure 4 shows the actual new supplier plant K-density (the solid line) and the 95% bootstrap confidence interval for the K-density based on the uniform probabilities pi = 1/nz. Given the large area covered by the eastern United States, simply assigning 431 14 new plants randomly across space would lead to a density function with a peak at a distance of roughly 500 miles. Instead, the actual K-density function has twin peaks at about 135 miles and 250 miles. The actual K-density function is well above the 95% confidence interval from distances of zero to 350 miles. Figure 4 provides clear visual evidence that new supplier plants are highly concentrated geographically. Figures 5 and 6 show comparable confidence intervals for the new-plant Kdensity function based on the total employment probabilities Ei/(ΣiEi) and the manufacturing employment probabilities Emi/(ΣiEmi). The 95% confidence intervals are virtually identical because the two sets of probabilities are highly correlated. 6 The only difference between these figures and Figure 4 is that the area where the actual K-density function is above the 95% confidence interval extends a bit farther – to 390 miles rather than 350. Whether we use uniform probabilities or weight the probabilities by employment shares, the K-densities imply a highly concentrated distribution of auto supplier plants in the eastern United States. 6 Across all 28,036 zip codes, the correlation between the total and manufacturing employment shares is 0.87. The correlation is 0.86 for the 19,151 zip codes that have some manufacturing employment. 15 5. Conditional Logit Models of Plant Locations In this section, we present conditional logit models explaining the probability that an auto supplier plant is located in a zip code. The primary question is whether we can explain the geographic concentration of supplier plants with such key explanatory variables as distance from Detroit, the presence of a highway, and proximity to assembly plants. Our analysis is not the first attempt to model the location decision of auto supplier plants. Woodward (1992) and Smith and Florida (1992) use county-level data to establish the importance of highway transportation as a determinant of plant location. However, our analysis is unique in the level of geographic detail and the use of a conditional logit approach in place of simple multinomial logit. With 28,036 zip codes, 2,627 plants, and plant openings as recent as 2003, our dataset is unusually detailed. The existing literature uses county-level data and multinomial logit models to determine the effect of county characteristics on the probability of plant location. In contrast, the conditional logit model operates at a more micro level. For each plant, we know the characteristics of rejected zip codes as well as the characteristics of the chosen location. The conditional logit model combines the chosen and rejected locations to produce a much more efficient set of coefficient estimates. Implicitly, each plant faces 28,036 potential location choices. However, we follow Ben-Akiva and Lerman (1985) and randomly choose five rejected alternatives when estimating the model. Since the rejected alternatives are chosen randomly, the resulting coefficients estimates are consistent and more efficient than a simple county-level multinomial logit model. The resulting data set has 6n observations. The dependent variable equals 1 for the first observation for each plant and the explanatory variables include the 16 characteristics for the chosen plant location. The dependent variable equals zero for the next five observations for each plant and the explanatory variables include the characteristics for the randomly chosen rejected locations. The standard errors are adjusted for the clustering that is implicit in having six observations for each plant. The results are shown in Table 2. For existing plants – those that opened prior to 1991 – the results imply that a zip code is more likely to be chosen as a plant location if an interstate highway runs through it, assemblers are nearby, it is in a metropolitan area, and it is in a right to work state. 7 The probability of an existing plant is higher when the zip code is near Detroit, has a high population density and a high proportion of manufacturing in employment. All of the regional dummy variables are significant except East South Central, whose negative coefficient is not significantly different from zero. In the second column of results in Table 2, the specification for new plant locations is similar to the model of existing plant locations. However, we add as an explanatory variable the number of existing supplier plants within 100 miles. This variable can reasonably be taken as exogenous for the new-plant model. Unfortunately, the number of assemblers and the number of suppliers are very highly correlated: the correlation between these two variables is 0.91 in the zip codes with new plants. This multicollinearity makes it difficult to separate the effects of proximity to assemblers and to existing supplier plants. In the last column of results, we present the results when the model is re-estimated after keeping only the more influential variable, the number of existing supplier plants. Making allowances for the smaller sample size, the results for 17 new plants are very similar to the results for existing plants. A zip code is more likely to have a new plant if it is served by an interstate highway, is close to an assembler, has a high proportion of employment in manufacturing, and is close to Detroit. We also find that new plants are more likely to choose zip codes that are within 100 miles of existing industry plants, both assembly and supplier plants. The model also suggests that, for new supplier plant locations, proximity to the nearest assembly plant matters instead of the number of assembly plants that are within 100 miles. This variation in the way existing assembly plant locations affect the choices of supplier plants is consistent with evidence of tighter linkages between assemblers and suppliers during the 1990s. An increasing number of logistics and supplier functions must be performed in very close proximity to the assembly location. In a number of cases, this tendency has led to the construction of a supplier park immediately adjacent to an assembly plant. With pseudo-R2s in the range of 0.35-0.41, the models fit the data well by the standards of discrete choice models. The models suggest the roots of geographic concentration lie in (1) highway access, (2) the desire to locate near assembly plants, and (3) the strong influence of Detroit on location decisions in the auto industry. In addition, we find that existing supplier plants appear to have some influence on the location of new plants. However, the models are not able to determine whether existing plants exert a causal influence on new plants due to direct agglomerative forces or if existing plant locations are serving as a proxy for missing variables that influence both new and existing plant location choices. 7 In states with right to work laws, a worker does not have to join a union as a condition for working in a unionized plant. Since nearly all right to work states are in the South, it is sometimes difficult to 18 6. K-Densities Based on Logit Probability Estimates The estimated probability estimates from the conditional logit models can be used as the basis for K-density confidence intervals. The implied base model asks a different question than before: are actual plant locations more concentrated geographically than implied by the estimated logit models? If we base the analysis of new-plant K-densities on the new-plant logit probabilities, we have what amounts to a specification test of our conditional logit model. If the model adequately accounts for the determinants of new plant locations, then the actual K-densities for new plants should lie within the 95% confidence interval implied by the estimated probabilities. The question is somewhat different if we base the confidence intervals for new-plant K-densities on the estimated probabilities produced by the logit model of existing plant locations. In this case, the question is whether new plants effectively follow the decision rule that is implied by the existing plant model. New plants may seek out locations that have a high probability of having an existing plant even if a plant has not yet located there. In this case, the Kdensities for new plants may lie within the 95% confidence interval implied by the existing plant logit model even if it differs from the distribution of actual existing plant distance densities. To calculate the 95% confidence interval for the new-plant K-densities based on the estimated logit models, we again draw randomly with replacement from the set of actual zip codes. The probabilities are based on the estimated conditional logit models. Thus, equation (2) again forms the basis for the bootstrap K-density. Unlike multinomial logit, the conditional logit model does not produce an intercept. Instead, separate intercepts are implied for each plant. Many zip codes are not represented in either the set distinguish the effects of this variable from regional indicators. 19 of actual plant locations or the randomly drawn alternatives. To construct probabilities for every zip code in the eastern United States, we take the sample of 6n observations and re-estimate the model using simple multinomial logit. In the re-estimated model, the dependent variable equals one for n observations and zero for the remaining 5n observations. The resulting coefficient estimates are consistent but not as efficient as the models that take into account the clustering by plant group. However, the re-estimated multinomial logit model includes an intercept, and the coefficients can be used to calculate probabilities for every zip code. 8 Figure 7 shows the actual K-density for new supplier plants and the bootstrap confidence interval implied by the probabilities estimated using the model of new plant locations. 9 Comparing the confidence intervals across Figures 4-7, we see that a much lower degree of concentration is implied by comparing actual densities to the densities implied by the new plant logit model. The actual K-density function is still above the 95% confidence interval in Figure 7, but it is much closer than was the case when the confidence intervals were based on zip code employment levels or uniform draws from all zip codes. Thus, the logit model has succeeded in explaining much of the tendency toward geographic concentration. Explaining the degree of concentration further would require more explanatory variables or a model that explicitly takes account of spatial autocorrelation. 10 8 Apart from the intercept, the coefficients of the conditional logit model and the multinomial re-estimated model are nearly identical. 9 The probabilities are based on the model without the variable indicating the number of assemblers within 100 miles. 10 The literature on discrete choice models with spatial autocorrelation is still largely undeveloped. Relevant models include those proposed by Beron and Vijverberg (2004), Case (1992), LeSage (2000), McMillen (1992), and Pinkse and Slade (1998). Currently, the models are only practicable for relatively small datasets because they involve inverting large weight matrices. 20 As shown in Figure 8, calculating confidence intervals for the K-density function based on the existing plant logit probabilities produces a diagram that is virtually identical to Figure 7. This result is not surprising since the correlation between the two estimated sets of probabilities is 0.90. Figure 9 shows the confidence intervals when we replace the estimated old-plant logit probabilities with actual old-plant locations. To construct these confidence intervals, we randomly draw samples of 431 locations from the 2196 actual old-plant locations. We then measure the distance of the 431 actual new plants to the randomly drawn sample of locations. Aside from minor differences, the resulting 95% confidence interval contains the new-plant density function. In other words, the distribution of new-plant distances is nearly the same as what would be expected if new plants locations were simply drawn randomly from the sites of old plants. This result does not imply, of course, that new plants actually locate in the same sites as old plants. The importance of the result is that new plants show no additional tendency to cluster beyond the level of concentration of old plants. As the auto industry changed its orientation southward, the overall level of concentration remained essentially the same as before. 21 7. Conclusion For the past century, the U.S. auto industry has been characterized by a small number of assembly plants and a large number of clustered supplier plants. Detroit remains the hub of the industry even as foreign plants have become more prominent. As American companies closed plants on the coasts and re-trenched toward the middle of the company, the industry has spread southward. The geographic distribution of auto supplier plants now displays a north-south orientation, with a concentration of plants along a corridor running from Detroit southward through Ohio, Kentucky, Tennessee, and into Alabama. In this paper, we use a combination of nonparametric and parametric techniques to characterize the geographic distribution of auto supplier plants in the eastern United States. Using a nonparametric procedure developed by Duranton and Overman (2005), we find that auto supplier plants are much more concentrated than would be implied by random location choice. We then investigate the roots of this geographic concentration using parametric conditional logit models. We find that the location choices of U.S. auto supplier plants are well explained by a small set of variables: the probability that a zip code has a plant is higher if the zip code has good highway access, is close to Detroit, and is near assembly plants. We also find that new supplier plants – those that have opened since 1991 – are more likely to locate in zip codes that are near existing concentrations of supplier plants. Despite the recent change in the geographic orientation of the industry, both the nonparametric and parametric procedures suggest that the distribution of plants has not changed significantly over time. Although plant openings have been 22 concentrated in the area south of Detroit, the new location pattern mimics the distribution of existing plants in the area. Our results also suggest the usefulness of Duranton and Overman’s (2005) procedure as a specification test for the conditional logit models. Although the logit models fit the data well, we find that plant locations are more concentrated geographically than is implied by the predicted logit probabilities. This result calls for the development of discrete choice models that explicitly take account of spatial clustering. 23 References Ben-Akiva, Moshe and Steven Lerman, Discrete Choice Analysis: Theory and Application to Travel Demand, MIT Press, Cambridge MA (1985). Beron, K. J. and Wim P. M. Vijverberg, “Probit in a Spatial Context: A Monte Carlo Analysis,” in Luc Anselin, Raymond J. G. M. Florax, and Sergio J. Rey (eds.), Advances in Spatial Econometrics: Methodology, Tools and Applications, Springer, New York (2004), 169-195. Case, Anne C. “Neighborhood Influence and Technological Change,” Regional Science and Urban Economics 22 (1992), 491-508. Duranton, Gilles and Henry G. Overman, “Testing for Localisation using Microgeographic Data”, Review of Economic Studies 72 (2005), 1077-1106. Duranton, Gilles and Henry G. Overman, “Exploring the Detailed Location Patterns of UK Manufacturing Industries using Microgeographic Data, manuscript (2006). Ellison, Glenn and Edward L. Glaeser, “Geographic Concentration in U.S. Manufacturing Industries: A Dartboard Approach,” Journal of Political Economy 105 (1997), 889-927. Harbour Consulting, Harbour Report, 2004 (2003). Klier, Thomas and Daniel P. McMillen, “The Geographic Evolution of the U.S. Auto Industry,” Economic Perspectives 30 (2006), 2-13. LeSage, James P., “Bayesian Estimation of Limited Dependent Variable Spatial Autoregressive Models,” Geographical Analysis 32 (2000), 19-35. McMillen, Daniel P., “Probit with Spatial Autocorrelation,” Journal of Regional Science 32 (1992), 335-348. Pinkse, Joris and Slade, Margaret E., “Contracting in Space: An Application of Spatial Statistics to Discrete-Choice Models,” Journal of Econometrics 85 (1998), 125154. Rubenstein, James M., The Changing U.S. Auto Industry – A Geographical Analysis, Routledge, London (1992). Silverman, A. W., Density Estimation for Statistics and Data Analysis, Chapman and Hall, New York (1986). 24 Smith, Donald and Richard Florida, “Agglomeration and Industrial location: An Econometric Analysis of Japanese-Affiliated Manufacturing Establishments in Automotive-Related Industries,” Journal of Urban Economics 36 (1994), 23-41. Woodward, Douglas, “Locational Determinants of Japanese Manufacturing Start-ups in the United States,” Southern Economic Journal 58 (1992), 690-708. 25 Table 1 Descriptive Statistics Interstate highway Distance to nearest assembler (100 miles) Number of assemblers within 100 miles Population density (1000s per sq. mile) Proportion white Proportion high school graduates Proportion manufacturing Metropolitan New England Middle Atlantic West North Central South Atlantic East South Central West South Central Right to Work state Distance from Detroit (100 miles) Number of observations Existing Plants Plant Zip Random Code Alternative 0.5168 0.3186 (0.4998) (0.4659) 0.6316 1.1662 (0.6550) (0.9962) 7.0724 1.9405 (8.4403) (3.6457) 1.4534 1.6382 (2.6110) (7.2710) 0.9027 0.8773 (0.1741) (0.2035) 0.7355 0.7126 (0.1233) (0.1347) 0.2713 0.2009 (0.0932) (0.1075) 0.6699 0.5932 0.0291 0.0813 0.0515 0.1802 0.0360 0.1117 0.1015 0.2586 0.1120 0.1138 0.0109 0.0479 0.1890 0.3557 2.6959 4.7679 (2.0829) (2.2097) 2196 10980 New Plants Plant Zip Random Code Alternative 0.6125 0.3197 (0.4877) (0.4665) 0.5958 1.1587 (0.5566) (1.0176) 7.0325 2.1276 (8.6411) (4.0782) 1.1125 1.6635 (2.1005) (7.4259) 0.8941 0.8799 (0.1758) (0.2029) 0.7200 0.7144 (0.1238) (0.1344) 0.2684 0.2006 (0.0880) (0.1070) 0.6845 0.5810 0.0093 0.0701 0.0302 0.1712 0.0302 0.1077 0.1276 0.2488 0.2483 0.1118 0.0070 0.0501 0.2645 0.3480 2.8686 4.6955 (2.0634) (2.2894) 431 2155 Note. Standard deviations are in parentheses for the continuous variables. 26 Table 2 Conditional Logit Models Interstate highway Distance to nearest assembler (100 miles) Number of assemblers within 100 miles Population density (1000s per sq. mile) Proportion white Proportion high school graduates Proportion manufacturing Metropolitan New England Middle Atlantic West North Central South Atlantic East South Central West South Central Right to Work state Distance from Detroit (100 miles) Number of existing supplier plants within 100 miles Pseudo-R2 Number of observations Existing Plants 0.6957* (0.0609) -0.0114 (0.0604) 0.0340* (0.0060) 0.0115* (0.0051) -0.2627 (0.1815) 1.6900* (0.2928) 5.4731* (0.3194) 0.2154* (0.0681) -0.7605* (0.1739) -1.5705* (0.1160) -0.7301* (0.1552) -0.9349* (0.1337) -0.1737 (0.1363) -0.6324* (0.2714) 0.4255* (0.1249) -0.2782* (0.0309) 0.3532 13176 New Plants 1.2154* (0.1450) -0.4203* (0.1470) -0.0068 (0.0281) -0.0037 (0.0219) 0.2647 (0.4076) 0.9358 (0.6804) 5.7246* (0.7976) 0.3009 (0.1611) -0.6020 (0.5987) -1.2945* (0.3577) 0.0557 (0.4063) 0.4681 (0.3053) 1.4789* (0.2755) -0.0625 (0.7443) 0.2678 (0.2801) -0.2770* (0.0891) 0.0022 (0.0014) 0.4076 2586 New Plants 1.2154* (0.1450) -0.4126* (0.1435) -0.0037 (0.0220) 0.2625 (0.4078) 0.9576 (0.6749) 5.7663* (0.7790) 0.2996 (0.1610) -0.6151 (0.5960) -1.3213* (0.3400) 0.0326 (0.3944) 0.4489 (0.2943) 1.4686* (0.2719) -0.0759 (0.7415) 0.2712 (0.2796) -0.2818* (0.0868) 0.0019* (0.0007) 0.4076 2586 Notes. Standard errors are in parentheses. An asterisk indicates significance at the 5% level. 27 Figure1 Distribution of Old Auto Supplier Plants Source: ELM International, state manufacturing directories, supplier company websites, and Harbour Consulting (2003) 28 Figure 2 Distribution of New Auto Supplier Plants Source: ELM International, state manufacturing directories, supplier company websites, and Harbour Consulting (2003) 0.00000 0.00025 0.00050 0.00075 0.00100 0.00125 0.00150 0.00175 0.00200 0 200 New Plants Existing Plants 400 Densities for Existing and New Supplier Plants Figure 3 600 800 29 0.00000 0.00025 0.00050 0.00075 0.00100 0.00125 0.00150 0.00175 0.00200 0 200 400 New Supplier Plant Density and Confidence Interval: Random Choice of Any Zip Code Figure 4 600 800 30 0.00000 0.00025 0.00050 0.00075 0.00100 0.00125 0.00150 0.00175 0.00200 0 200 400 New Supplier Plant Density and Confidence Interval: Total Employment Figure 5 600 800 31 0.00000 0.00025 0.00050 0.00075 0.00100 0.00125 0.00150 0.00175 0.00200 0 200 400 New Supplier Plant Density and Confidence Interval: Manufacturing Employment Figure 6 600 800 32 0.00000 0.00025 0.00050 0.00075 0.00100 0.00125 0.00150 0.00175 0.00200 0 200 400 New Supplier Plant Density and Confidence Interval: Estimated Probability of a New Plant Figure 7 600 800 33 0.00000 0.00025 0.00050 0.00075 0.00100 0.00125 0.00150 0.00175 0.00200 0 200 400 New Supplier Plant Density and Confidence Interval: Estimated Probability of an Existing Plant Figure 8 600 800 34 0.00000 0.00025 0.00050 0.00075 0.00100 0.00125 0.00150 0.00175 0.00200 0 200 400 New Supplier Plant Density and Confidence Interval: Actual Locations of Existing Plants Figure 9 600 800 35 Working Paper Series A series of research studies on regional economic issues relating to the Seventh Federal Reserve District, and on financial and economic topics. A Proposal for Efficiently Resolving Out-of-the-Money Swap Positions at Large Insolvent Banks George G. Kaufman WP-03-01 Depositor Liquidity and Loss-Sharing in Bank Failure Resolutions George G. Kaufman WP-03-02 Subordinated Debt and Prompt Corrective Regulatory Action Douglas D. Evanoff and Larry D. Wall WP-03-03 When is Inter-Transaction Time Informative? Craig Furfine WP-03-04 Tenure Choice with Location Selection: The Case of Hispanic Neighborhoods in Chicago Maude Toussaint-Comeau and Sherrie L.W. Rhine WP-03-05 Distinguishing Limited Commitment from Moral Hazard in Models of Growth with Inequality* Anna L. Paulson and Robert Townsend WP-03-06 Resolving Large Complex Financial Organizations Robert R. Bliss WP-03-07 The Case of the Missing Productivity Growth: Or, Does information technology explain why productivity accelerated in the United States but not the United Kingdom? Susanto Basu, John G. Fernald, Nicholas Oulton and Sylaja Srinivasan WP-03-08 Inside-Outside Money Competition Ramon Marimon, Juan Pablo Nicolini and Pedro Teles WP-03-09 The Importance of Check-Cashing Businesses to the Unbanked: Racial/Ethnic Differences William H. Greene, Sherrie L.W. Rhine and Maude Toussaint-Comeau WP-03-10 A Firm’s First Year Jaap H. Abbring and Jeffrey R. Campbell WP-03-11 Market Size Matters Jeffrey R. Campbell and Hugo A. Hopenhayn WP-03-12 The Cost of Business Cycles under Endogenous Growth Gadi Barlevy WP-03-13 The Past, Present, and Probable Future for Community Banks Robert DeYoung, William C. Hunter and Gregory F. Udell WP-03-14 1 Working Paper Series (continued) Measuring Productivity Growth in Asia: Do Market Imperfections Matter? John Fernald and Brent Neiman WP-03-15 Revised Estimates of Intergenerational Income Mobility in the United States Bhashkar Mazumder WP-03-16 Product Market Evidence on the Employment Effects of the Minimum Wage Daniel Aaronson and Eric French WP-03-17 Estimating Models of On-the-Job Search using Record Statistics Gadi Barlevy WP-03-18 Banking Market Conditions and Deposit Interest Rates Richard J. Rosen WP-03-19 Creating a National State Rainy Day Fund: A Modest Proposal to Improve Future State Fiscal Performance Richard Mattoon WP-03-20 Managerial Incentive and Financial Contagion Sujit Chakravorti and Subir Lall WP-03-21 Women and the Phillips Curve: Do Women’s and Men’s Labor Market Outcomes Differentially Affect Real Wage Growth and Inflation? Katharine Anderson, Lisa Barrow and Kristin F. Butcher WP-03-22 Evaluating the Calvo Model of Sticky Prices Martin Eichenbaum and Jonas D.M. Fisher WP-03-23 The Growing Importance of Family and Community: An Analysis of Changes in the Sibling Correlation in Earnings Bhashkar Mazumder and David I. Levine WP-03-24 Should We Teach Old Dogs New Tricks? The Impact of Community College Retraining on Older Displaced Workers Louis Jacobson, Robert J. LaLonde and Daniel Sullivan WP-03-25 Trade Deflection and Trade Depression Chad P. Brown and Meredith A. Crowley WP-03-26 China and Emerging Asia: Comrades or Competitors? Alan G. Ahearne, John G. Fernald, Prakash Loungani and John W. Schindler WP-03-27 International Business Cycles Under Fixed and Flexible Exchange Rate Regimes Michael A. Kouparitsas WP-03-28 Firing Costs and Business Cycle Fluctuations Marcelo Veracierto WP-03-29 Spatial Organization of Firms Yukako Ono WP-03-30 Government Equity and Money: John Law’s System in 1720 France François R. Velde WP-03-31 2 Working Paper Series (continued) Deregulation and the Relationship Between Bank CEO Compensation and Risk-Taking Elijah Brewer III, William Curt Hunter and William E. Jackson III WP-03-32 Compatibility and Pricing with Indirect Network Effects: Evidence from ATMs Christopher R. Knittel and Victor Stango WP-03-33 Self-Employment as an Alternative to Unemployment Ellen R. Rissman WP-03-34 Where the Headquarters are – Evidence from Large Public Companies 1990-2000 Tyler Diacon and Thomas H. Klier WP-03-35 Standing Facilities and Interbank Borrowing: Evidence from the Federal Reserve’s New Discount Window Craig Furfine WP-04-01 Netting, Financial Contracts, and Banks: The Economic Implications William J. Bergman, Robert R. Bliss, Christian A. Johnson and George G. Kaufman WP-04-02 Real Effects of Bank Competition Nicola Cetorelli WP-04-03 Finance as a Barrier To Entry: Bank Competition and Industry Structure in Local U.S. Markets? Nicola Cetorelli and Philip E. Strahan WP-04-04 The Dynamics of Work and Debt Jeffrey R. Campbell and Zvi Hercowitz WP-04-05 Fiscal Policy in the Aftermath of 9/11 Jonas Fisher and Martin Eichenbaum WP-04-06 Merger Momentum and Investor Sentiment: The Stock Market Reaction To Merger Announcements Richard J. Rosen WP-04-07 Earnings Inequality and the Business Cycle Gadi Barlevy and Daniel Tsiddon WP-04-08 Platform Competition in Two-Sided Markets: The Case of Payment Networks Sujit Chakravorti and Roberto Roson WP-04-09 Nominal Debt as a Burden on Monetary Policy Javier Díaz-Giménez, Giorgia Giovannetti, Ramon Marimon, and Pedro Teles WP-04-10 On the Timing of Innovation in Stochastic Schumpeterian Growth Models Gadi Barlevy WP-04-11 Policy Externalities: How US Antidumping Affects Japanese Exports to the EU Chad P. Bown and Meredith A. Crowley WP-04-12 Sibling Similarities, Differences and Economic Inequality Bhashkar Mazumder WP-04-13 3 Working Paper Series (continued) Determinants of Business Cycle Comovement: A Robust Analysis Marianne Baxter and Michael A. Kouparitsas WP-04-14 The Occupational Assimilation of Hispanics in the U.S.: Evidence from Panel Data Maude Toussaint-Comeau WP-04-15 Reading, Writing, and Raisinets1: Are School Finances Contributing to Children’s Obesity? Patricia M. Anderson and Kristin F. Butcher WP-04-16 Learning by Observing: Information Spillovers in the Execution and Valuation of Commercial Bank M&As Gayle DeLong and Robert DeYoung WP-04-17 Prospects for Immigrant-Native Wealth Assimilation: Evidence from Financial Market Participation Una Okonkwo Osili and Anna Paulson WP-04-18 Individuals and Institutions: Evidence from International Migrants in the U.S. Una Okonkwo Osili and Anna Paulson WP-04-19 Are Technology Improvements Contractionary? Susanto Basu, John Fernald and Miles Kimball WP-04-20 The Minimum Wage, Restaurant Prices and Labor Market Structure Daniel Aaronson, Eric French and James MacDonald WP-04-21 Betcha can’t acquire just one: merger programs and compensation Richard J. Rosen WP-04-22 Not Working: Demographic Changes, Policy Changes, and the Distribution of Weeks (Not) Worked Lisa Barrow and Kristin F. Butcher WP-04-23 The Role of Collateralized Household Debt in Macroeconomic Stabilization Jeffrey R. Campbell and Zvi Hercowitz WP-04-24 Advertising and Pricing at Multiple-Output Firms: Evidence from U.S. Thrift Institutions Robert DeYoung and Evren Örs WP-04-25 Monetary Policy with State Contingent Interest Rates Bernardino Adão, Isabel Correia and Pedro Teles WP-04-26 Comparing location decisions of domestic and foreign auto supplier plants Thomas Klier, Paul Ma and Daniel P. McMillen WP-04-27 China’s export growth and US trade policy Chad P. Bown and Meredith A. Crowley WP-04-28 Where do manufacturing firms locate their Headquarters? J. Vernon Henderson and Yukako Ono WP-04-29 Monetary Policy with Single Instrument Feedback Rules Bernardino Adão, Isabel Correia and Pedro Teles WP-04-30 4 Working Paper Series (continued) Firm-Specific Capital, Nominal Rigidities and the Business Cycle David Altig, Lawrence J. Christiano, Martin Eichenbaum and Jesper Linde WP-05-01 Do Returns to Schooling Differ by Race and Ethnicity? Lisa Barrow and Cecilia Elena Rouse WP-05-02 Derivatives and Systemic Risk: Netting, Collateral, and Closeout Robert R. Bliss and George G. Kaufman WP-05-03 Risk Overhang and Loan Portfolio Decisions Robert DeYoung, Anne Gron and Andrew Winton WP-05-04 Characterizations in a random record model with a non-identically distributed initial record Gadi Barlevy and H. N. Nagaraja WP-05-05 Price discovery in a market under stress: the U.S. Treasury market in fall 1998 Craig H. Furfine and Eli M. Remolona WP-05-06 Politics and Efficiency of Separating Capital and Ordinary Government Budgets Marco Bassetto with Thomas J. Sargent WP-05-07 Rigid Prices: Evidence from U.S. Scanner Data Jeffrey R. Campbell and Benjamin Eden WP-05-08 Entrepreneurship, Frictions, and Wealth Marco Cagetti and Mariacristina De Nardi WP-05-09 Wealth inequality: data and models Marco Cagetti and Mariacristina De Nardi WP-05-10 What Determines Bilateral Trade Flows? Marianne Baxter and Michael A. Kouparitsas WP-05-11 Intergenerational Economic Mobility in the U.S., 1940 to 2000 Daniel Aaronson and Bhashkar Mazumder WP-05-12 Differential Mortality, Uncertain Medical Expenses, and the Saving of Elderly Singles Mariacristina De Nardi, Eric French, and John Bailey Jones WP-05-13 Fixed Term Employment Contracts in an Equilibrium Search Model Fernando Alvarez and Marcelo Veracierto WP-05-14 Causality, Causality, Causality: The View of Education Inputs and Outputs from Economics Lisa Barrow and Cecilia Elena Rouse WP-05-15 5 Working Paper Series (continued) Competition in Large Markets Jeffrey R. Campbell WP-05-16 Why Do Firms Go Public? Evidence from the Banking Industry Richard J. Rosen, Scott B. Smart and Chad J. Zutter WP-05-17 Clustering of Auto Supplier Plants in the U.S.: GMM Spatial Logit for Large Samples Thomas Klier and Daniel P. McMillen WP-05-18 Why are Immigrants’ Incarceration Rates So Low? Evidence on Selective Immigration, Deterrence, and Deportation Kristin F. Butcher and Anne Morrison Piehl WP-05-19 Constructing the Chicago Fed Income Based Economic Index – Consumer Price Index: Inflation Experiences by Demographic Group: 1983-2005 Leslie McGranahan and Anna Paulson WP-05-20 Universal Access, Cost Recovery, and Payment Services Sujit Chakravorti, Jeffery W. Gunther, and Robert R. Moore WP-05-21 Supplier Switching and Outsourcing Yukako Ono and Victor Stango WP-05-22 Do Enclaves Matter in Immigrants’ Self-Employment Decision? Maude Toussaint-Comeau WP-05-23 The Changing Pattern of Wage Growth for Low Skilled Workers Eric French, Bhashkar Mazumder and Christopher Taber WP-05-24 U.S. Corporate and Bank Insolvency Regimes: An Economic Comparison and Evaluation Robert R. Bliss and George G. Kaufman WP-06-01 Redistribution, Taxes, and the Median Voter Marco Bassetto and Jess Benhabib WP-06-02 Identification of Search Models with Initial Condition Problems Gadi Barlevy and H. N. Nagaraja WP-06-03 Tax Riots Marco Bassetto and Christopher Phelan WP-06-04 The Tradeoff between Mortgage Prepayments and Tax-Deferred Retirement Savings Gene Amromin, Jennifer Huang,and Clemens Sialm WP-06-05 Why are safeguards needed in a trade agreement? Meredith A. Crowley WP-06-06 6 Working Paper Series (continued) Taxation, Entrepreneurship, and Wealth Marco Cagetti and Mariacristina De Nardi WP-06-07 A New Social Compact: How University Engagement Can Fuel Innovation Laura Melle, Larry Isaak, and Richard Mattoon WP-06-08 Mergers and Risk Craig H. Furfine and Richard J. Rosen WP-06-09 Two Flaws in Business Cycle Accounting Lawrence J. Christiano and Joshua M. Davis WP-06-10 Do Consumers Choose the Right Credit Contracts? Sumit Agarwal, Souphala Chomsisengphet, Chunlin Liu, and Nicholas S. Souleles WP-06-11 Chronicles of a Deflation Unforetold François R. Velde WP-06-12 Female Offenders Use of Social Welfare Programs Before and After Jail and Prison: Does Prison Cause Welfare Dependency? Kristin F. Butcher and Robert J. LaLonde Eat or Be Eaten: A Theory of Mergers and Firm Size Gary Gorton, Matthias Kahl, and Richard Rosen Do Bonds Span Volatility Risk in the U.S. Treasury Market? A Specification Test for Affine Term Structure Models Torben G. Andersen and Luca Benzoni WP-06-13 WP-06-14 WP-06-15 Transforming Payment Choices by Doubling Fees on the Illinois Tollway Gene Amromin, Carrie Jankowski, and Richard D. Porter WP-06-16 How Did the 2003 Dividend Tax Cut Affect Stock Prices? Gene Amromin, Paul Harrison, and Steven Sharpe WP-06-17 Will Writing and Bequest Motives: Early 20th Century Irish Evidence Leslie McGranahan WP-06-18 How Professional Forecasters View Shocks to GDP Spencer D. Krane WP-06-19 Evolving Agglomeration in the U.S. auto supplier industry Thomas Klier and Daniel P. McMillen WP-06-20 7
@FedFRASER