Concentration history and market power in US manufacturing industries

Abstract

For a 1963–1992 panel of US manufacturing industries, the relationship between seller concentration and both price-cost margins (PCMs) and prices is investigated for industries divided by whether concentration has recently increased or decreased. Regressions of PCM in levels and first differences, and price equations in first differences, establish that the positive effect of concentration on prices and profits is always weaker in industries where concentration has recently increased and always stronger in industries where concentration has recently decreased. These results are attributed to the different endogeneity biases in the two samples. Increasing concentration industries are more likely the ones where leading firms have lowered prices to gain share, while decreasing concentration industries are more likely the ones where smaller firms have lowered concentration by lowering prices. An additional conclusion is that the cost-reducing effects of changes in concentration are greater for increasing concentration industries, meaning that increasing concentration industries have lower price increases compared to decreasing concentration industries.

Notes

¹ Demsetz (1973) introduced the efficiency explanation for positive concentration coefficients. On endogeneity and measurement problems, see Schmalensee (1989).

² Running separate regressions for each increasing and decreasing concentration industry subsample does not change the results.

³ For example, Domowitz et al. (1986); Coate (1991); Dickson (2005) all find a significant negative effect for the capital-output variable using fixed-effects regressions on panels of US manufacturing. Crandall and Winston (2003) also found, using least squares on a panel covering 20 industries from 1984 to 1996, a significant negative effect for the capital-output ratio.

⁴ For example, 1-year first differences do not produce significant estimates of the concentration coefficient. This is probably because, with a relatively slow moving variable like concentration, short differencing intervals are dominated by measurement error, which biases the coefficient to zero.

⁵ With the smaller SEs in Table 3, the concentration coefficients are only different from one another at the 28% level of significance for the 5-year interval, and at the 2% level for the 10-year interval. However, as will be shown, once price and cost effects are separated, it is clear that there are significant differences in market power effects between increasing and decreasing concentration samples.

⁶ The input price indices are from the NBER database. The wage index is constructed as payroll divided by employees. A Hausman test rejected the least squares estimator at the 5% level, although using AVC, instead of the constructed instrument, did not affect the conclusions.

⁷ For the 5- and 10-year intervals, the concentration coefficients for increasing and decreasing concentration samples are significantly different from one another at the 2 and 1% level, respectively. For the 20-year interval the difference is less clear-cut, with the difference being significant at the 16% level.

⁸ This same kind of endogeneity bias is cited (p. 300) by Salinger (1990) who writes about dominant firms who engage in price wars to shrink the fringe so that concentration is rising and prices are falling.

⁹ Regressions were also performed that included changes in input prices (energy, materials and labour) as independent variables in place of changes in average cost. The concentration coefficients were not significantly affected.

¹⁰ Another coefficient affected by the omission of AVC is the coefficient for the capital-output ratio which becomes positive and significant at the one% level. Capital intensity is positively correlated with AVC largely because the labour-output and energy-output ratios are correlated with the capital-output ratio.

¹¹ Gisser (1984) also finds stronger efficiency effects for US manufacturing when concentration rises rather than falls.