|
Article Excerpt
This paper uses data on U.S. metropolitan areas from 1970 to 1999 to examine the extent to which recent increases in earnings are attributable to agglomeration economies. We decompose the total change in earnings over the 1970-1999 period into components attributable to changes in relative growth, industry mix and interaction effects. We find strong evidence that relative growth matters more than industry mix. In addition, we find that relative growth effects are related to both localization and urbanization economies.
**********
For at least the last three decades, a lively debate has taken place, partly in academic circles, but even more among policy makers, planners and their critics, about whether urban growth is caused by what cities do or how cities do it. The "what-cities-do" advocates assert that increasing urban growth is largely a matter of (1) moving into sectors that have high value-added per worker, that is, sectors like manufacturing (Bluestone and Harrison 1982, Cohen and Zysman 1987, Fingleton 1999), selected services (especially producer services; Hansen 1990, Coffey 2000) and, of course, high tech in general (DeVol et al. 1999) and particular areas, such as computer hardware and software or biotechnology (Blakely and Nishikawa 1992, Pollack 2002), and (2) maintaining a position in these industries that is superior to that of one's competitors. In contrast, the "how-cities-do-it" adherents assert that the real source of urban growth is agglomeration economies in production, particularly localization economies, and policies to support this, such as infrastructure investment, effective education and appropriate taxation (see, e.g., Henderson 1988, Moomaw 1988, Glaeser et al. 1992, 1995, Quigley 1998, Duranton and Puga 2000 and Rosenthal and Strange 2001, among many others).
One method to test for the extent and nature of agglomeration economies is to estimate augmented plant-level production functions, augmented by variables representing the size of the city in which the plant is located (urbanization economies) or some measure of linkages to related nearby firms (localization economies). (1) Data requirements for this approach are severe (see, e.g., Sveikauskas 1975 and the critique by Moomaw 1981). A second approach is to study compensating variations in wages and/or rents (Rosen 1986, Gyourko, Kahn and Tracy 1999, Glaeser and Mare 2001, Gabriel and Rosenthal 2004). Some papers examine the births and deaths of firms in different locations (Reynolds, Miller and Mahi 1995, Zucker, Darby and Brewer 1998, Rosenthal and Strange 2001).
We proceed as follows. First, we adapt a well-known descriptive technique of urban and regional growth analysis, shift-share analysis (see Hoover and Giarratini 1984 or Loveridge and Selting 1998 for elaboration of this technique). The goal of shift-share analysis, as traditionally defined, is to provide insight into comparative change by separating the rate of growth in earnings over a specific time period into two components: an industry growth effect and relative growth. (2) The industry growth effect is that portion of metropolitan growth that would be expected if all industries within the urban area were growing at some average rate nationally. The relative growth effect is the remaining growth in excess of the rate of growth in a particular industry. These two growth components are typically measured at the two-digit SIC code level, and then aggregated up to the metropolitan level by using actual industry weightings. But this last step is problematic. Because the data are aggregated up to the metropolitan level by using actual industry weightings, part of the cross-sectional variation in the relative growth component is undeniably due to market-wide variation in the industry weighting. This, in turn, creates a clear bias toward finding that industry mix matters. The way out of this dilemma is to weight the remaining rate of growth in excess of the rate of growth in a particular industry by the concentration of firms nationwide in that industry. In doing so, we also need to add an interaction term or cross-product effect that combines the two pure effects of industry mix and the relative growth. The addition of this interaction term or cross-product effect is a consequence of the nonlinearity of the growth rate differentials as a function of the industry mix and relative growth effects. (3)
Second, we apply our modified shift-share analysis to disaggregated real earnings growth data at the metropolitan level. Specifically, we calculate real earnings growth (deflated by the GDP deflator) by SIC code, by MSA, over the time period from 1970 through 1999. (4) We then examine whether the cross-sectional variation in metropolitan growth rates is due to differences in relative growth, industry mix or the interaction by performing a variance decomposition.
Third, we examine how metropolitan growth rates vary with fundamental determinants. Our analysis also employs the components of metropolitan growth as the dependent variable. The explanatory variables are metropolitan characteristics, including size and an industry concentration ratio as proxies for localization and urbanization economies.
Among other results, we show that approximately three-fourths of the cross-sectional variance in metropolitan growth rates can be explained by local conditions unrelated to industry mix. The remaining cross-sectional variance in metropolitan growth rates is attributed to differences in industry mix, interaction between industry mix and local effects. We find that when a city has a higher growth due to the local relative growth effect, it also tends to have a low growth term due to industry mix and vice versa.
In addition to demonstrating that industry mix matters somewhat, but that local factors matter more, we confirm the results of other studies that demonstrate that local growth effects are related to both urbanization and localization economies. With respect to the interaction effect, we find that it mainly increases only with respect to localization economies.
Another of our findings is a negative relationship between our relative growth term and earnings per capita at the beginning of the sample period. This result suggests that metropolitan areas with low initial earnings per capita will exhibit higher returns on the margin, drawing investment and raising growth rates until economies converge. Similarly, metropolitan areas with high initial earnings per capita will exhibit lower returns on the margin. These results are consistent with several previous studies such as Barro and Sala-i-Martin (1991) and Carlino and Mills (1993).
Taken together, these results tell us that the growth of local economies is no random walk, but is partly forecastable. Thus there are states of the world in which investments in particular markets are more or less favorable, at least probabilistically. More specific policy implications include the following. First, despite the near ubiquitousness of efforts by local officials to peg one's fortunes to one "hot" industry after another, (5) the results suggest it is tempting, but wrong, to credit the overall patterns of metropolitan economic development largely to industry mix. (6) While we find some effect of industry mix and some interaction with local factors, it is largely the latter that determines metropolitan growth.
In addition to public policy, our results have obvious implications for business policy, in particular for real estate investment decisions. Too many studies to list exhaustively document the links between the health of the local economy and the local real estate market. See, for example, Pollakowski, Wachter and Lynford (1992), Wheaton and Torto (1994), Grenadier (1995). Capozza, Kazarian and Thomson (1997) and Eppli, Shilling and Vandell (1998). A recent paper by Mourouzi-Sivitanidou (2002) presents a nice review and particularly interesting results regarding the effects of local employment on equilibrium office rents and the time path of adjustment to their equilibrium levels. Several studies have made explicit links between agglomeration and the health of the real estate market, including Corgel and Gay (1987), McNulty (1995) and Mun and Hutchison (1995). In general, these papers take growth in local employment and earnings/income as exogenous; this paper emphasizes that metropolitan growth rates are forecastable and hence offer different ex ante investment opportunities.
The remainder of the paper is organized as follows. The second section provides background on agglomeration economies and on previous work in this area. The third section describes the updated shift-share formula used to decompose metropolitan growth rates into components attributable to changes in relative growth, industry mix and interaction effects. We apply our updated shift-share formula to data from about 300 U.S. metropolitan areas between 1970 and 1999. The fourth section describes these data. The fifth section examines the cross-sectional variation in metropolitan growth rates and shows what fraction of this variation is due to relative growth and what fraction is due to industry mix and the interaction term. The sixth section looks at agglomeration economies in relation to other fundamental determinants of metropolitan growth and shows evidence consistent with the idea that cross-sectional differences in metropolitan growth rates are explained by a set of fundamentals, including both urbanization and localization economies. The last section concludes.
Selected Literature
The literature on agglomeration economies in cities is enormous. (7) The literature on regional models (shift-share and related models) related to our model is also very large (see Loveridge and Selting 1998). Thus we make no attempt to be comprehensive in this brief review, but rather wish to highlight a few key points.
Cities exist largely because of economies of scale, specialization and trade. Economies of scale can be subdivided into the "usual kind" (i.e., the kind found in Principles textbooks that give rise to U-shaped cost curves) that are internal to the firm and agglomeration economies, or economies of scale external to the firm, that shift cost curves down. Put another way, agglomeration economies are economies of scale from a spatial or locational externality. Thus (as the Soviets found out to their cost), the return to an investment often depends on where it is made and what other investments are made nearby.
Agglomeration economies are commonly attributed to Marshall (1890) and were much discussed by urban economists for many decades. The concept of agglomeration has achieved greater visibility in the last decade because nonspecialist economists and management gurus like Paul Krugman (1995) and Michael Porter (1998) have adopted and elucidated the concept.
While much recent literature highlights agglomeration, it should be noted...
|
