The Birth, Death, and Persistence of Firms: Creative Destruction and the Spatial Distribution of U.S. Manufacturing Establishments, 2000–2006

Abstract

This article deals with the dynamics of the U.S. manufacturing sector, analyzing the birth, death, and ongoing existence of firms in the beginning of the twenty-first century. Schumpeter’s notion of creative destruction is hypothesized to explain the spatiotemporal dynamics of the distribution of manufacturing establishments. We implemented a partial adjustment model that accounts for spillover effects between counties, unknown forms of heteroskedasticity, and spatial autocorrelation. The steady-state equilibrium birth and death rates converged to 6.8 percent and 6.1 percent per year, respectively, during the 2000–06 period. We found that firm birth and death were not decisively affected by a creative destruction process during that period, but firm birth and death positively affect the survival (or persistence) rate of single-unit manufacturing firms.

• location
• location factors
• manufacturing
• partial adjustment model
• L60
• R11
• R12

Acknowledgments

The views expressed here are those of the authors and may not be attributed to the Federal Reserve Bank of Kansas City, the University of Tennessee, Purdue University, VU University Amsterdam, or the Tinbergen Institute. Portions of this research were made possible through the United States Department of Agriculture Hatch funding, project NE-1029.

Notes

The terms entry and exit are synonymous with birth and death in the theoretical and empirical literature on industrial organization. See, for example, Hopenhayn (), Ericson and Pakes (), Pakes and Ericson (), Baldwin and Gorecki (), Johnson and Parker (), Reynolds, Miller, and Maki (), Love (), and Fotopoulos and Spence ().

The derivation of the reduced-form equations is provided in Appendix A.

Persistence is defined as year-to-year survival. See http://www.census.gov/econ/susb/methodology.html for more details.

As one reviewer noted, our model does not restrict the predicted values of the rates to be bounded between 0 and 1. However, an investigation of the predicted values for the birth-and-death rate equations revealed that the values land outside the interval only 1/3078 (0.03 percent) and 5/3078 (0.16 percent) of the time for birth and death, respectively. Similarly, the predicted value for the persistent rate is above the [0,1] interval 3/3078 (0.1 percent) times across all observations. These occurrences are probably due to chance alone and do not appear to be systematic. For example, a Type I error rate of α = 1 percent is the expected probability that the predicated value of an event would fall outside the [0,1] interval = 1 percent x 3,078 = 30 occurrences. If we had observed at least 30 observations being predicted outside they interval, then there would be reason to be more concerned about using the linear specification to model the percentage changes. Future work in this area could incorporate an estimation framework for a (fractional) logit specification to attend to the limited dependent nature of birth, persistence, and death rates.

The modifiable areal unit problem would also be a much more persistent issue of concern if counts of births and deaths were used in place of rates (Openshaw and Taylor ). However, as one reviewer pointed out, using rates may moderate potential problems with some types of heteroskedasticity, but it can also introduce additional problems that are associated with variance in the estimates of the rates if the denominator varies significantly across units. We attended to the issue of heteroskedasticity by implementing a robust estimator of the variance-covariance matrix.

We thank one of the anonymous reviewers for bringing this argument to our attention.

The employment share of the creative class was constructed using Wojan and McGranahan’s () results and is available at http://www.ers.usda.gov/Data/CreativeClassCodes/.

Regional price differences are important but also hard to measure given the available data. Net flow wages per commuter, local spending, income, and wages in part pick up some aspects of regional price differences, but we lack more ideal, direct measures (i.e., county-level price indexes for the prices of consumer goods and producers).

There are 3,078 U.S. counties included in this analysis of which 1,061 are metropolitan, 665 are micropolitan, and 1,352 are noncore based on the OMB classification from 2003.

Derivation and results from the determinational equation are shown in Appendix B. The dominant root indicates that the system is stable and this is confirmed by the phase diagrams in Figure 1.