Does Vertical Integration Effect Market Power? Evidence from U.S. Food Manufacturing Industries

Journal of Agricultural and Applied Economics, Apr 2005 by Bhuyan, Sanjib

Data and Empirical Procedure

The focus of this study is on the U.S. food manufacturing sector at the census, four-digit, industry group or SIC level. There were 49 food manufacturing industries at the four-digit SIC level in 1992, the year for which all the necessary data for this study were publicly available (data available upon request). Required data on the vertical integration index (FVI) and the market power index (L) were obtained from Bhuyan and Bhuyan and Lopez, respectively. The vertical integration index was not reported for the six industries (for SICs 2043, 2062, 2068, 2076, 2085, and 2097), and the market power index was not reported for five industries (for SICs 2037, 2038, 2046, 2091, and 2099). Thus, these 11 food manufacturing industries were dropped from this study, and the remaining 38 were used.

The four-firm concentration ratio (CR4) is the most accepted and commonly used measure of market concentration (Rogers), and it is used to represent market concentration in this study. Data on CR4 were obtained from the /992 Census of Manufacturers, Industry Series reports (U.S. Department of Commerce 1997). Data on regional concentration (represented by the dummy variable, REG) advertising intensity (ADVT) and capital intensity (KINT), the later two expressed as a percent of 1992 sales, were kindly provided by Professor Richard T. Rogers of the University of Massachusetts at Amherst. The Industrial Productivity Database, which is publicly available courtesy of the National Bureau of Economic Research (NBER, www.nber.org), was the source for industry sales data for 1982 (adjusted to the 1987 SIC definition), 1987, and 1992 to compute the demand fluctuations (DEMFLUC) variable. The variable import intensity (IMPORT), defined as a percentage of imported, processed food to 1992 sales, was obtained using trade data from the NBER's trade data bank. The /992 Benchmark Input-Output Accounts of the United States (U.S. Department of Commerce 1998) data were used to construct the INTRA variable as a percentage of 1992 sales to control for aggregation bias.

The level of aggregation used here has masked the more interesting, firm-level variations within an industry and across industries. However, as Perry noted, such firm-level analysis is impossible because of the lack of firm-specific data. Additionally, as in any empirical study, this study also suffers from data that do not perfectly represent the theoretical variables. But it is believe that in the absence of conceptually desirable data, use of available data should not diminish the importance of the linkage between market power and vertical integration. Given such data constraints, readers are urged to use their good judgment when perusing the results presented in a later section.

Because the dependent variable L is bounded between O and 1, it was transformed into the log-odds functional form, ln[L/(l - L)] for estimation in Equation (2).6 Preliminary screening showed evidence of heteroskedasticity (Glejser test, χ^sup 2^ 9 d.f. = 18.293). To obtain heteroskedasticity-consistent estimates of standard errors, White's method was employed (an alternative technique, the weighted least-square method, yields similar results). Summary statistics and a correlation matrix of variables used in Equation (2) are presented in Appendix Tables 1 and 2, respectively.