Competition and the relative productivity of large and small firms

Abstract

Using a comprehensive dataset on the incidence of price-fixing across British manufacturing industries in the 1950s, I compare collusive and competitive industries and find evidence of a negative relationship between collusion and the labour productivity of larger firms relative to smaller firms. In particular, collusion is associated with a reduction or even a reversal of the productivity gap between larger and smaller firms. This result is robust to controlling for the potential endogeneity of collusion.

Acknowledgements

I would like to thank two anonymous referees for very helpful comments and suggestions.

Notes

¹ An earlier study of the links between collusion and productivity in the UK, using cross-section data before the abolition of the cartels, is Broadberry and Crafts (1996).

² The answer to this question is unlikely to be ‘the very small and inefficient firms that survived under the cartel umbrella’ since these firms did not affect the aggregate industry productivity very much.

³ On the other hand, there is a large empirical literature on the effects of cartels on prices (see Connor and Bolotova, 2006; Levenstein and Suslow, 2006 for recent surveys).

⁴ The effectiveness of outside competition was limited in many industries because the cartels tended to contain most or all of the largest and best-known domestic firms; because practices intended to limit outside competition, such as aggregated rebates and collective exclusive dealing, were common; and because competition from imports was often limited as a result of tariffs and quantitative controls, differing technical standards, transport costs or international restrictive agreements.

⁵ Since comprehensive collusion data for the 1940s are not available, one cannot rule out the possibility that a few of the cartels became effective after 1951. This would, if anything, tend to bias toward zero the estimated coefficient of collusion on relative productivity in the Ordinary Least Squares (OLS) regressions of Section III.

⁶ The proportion of an industry's total sales subject to significant restrictions is for 1951, the same year as the productivity data. This proportion may change over time but rarely is this change so large during the 1950s as to cause an industry to move above or below the relevant cut-off point (and even then, it does not move much above or below). Furthermore, the results are robust to using different cut-off points, as will be shown later.

⁷ The sample excludes industries with significant government participation and includes two nonmanufacturing industries – these do not significantly affect the results.

⁸ Gross output is defined as the total value of sales and work done during the year adjusted for changes in the value of stocks. Net output is gross output minus the cost of materials and fuel, payments for work given out and transport payments.

⁹ The correlation coefficient between CONC and COLL is a modest 0.17. I also experimented with other available measures of market structure, such as the average size of the largest firms divided by the average size of the smaller firms or the total number of firms, but these were not statistically significant. It is not surprising that CONC performs better; unlike the other two variables, it is not affected by the number or the size of very small firms in an industry. A measure of minimum efficient scale based on the median plant size would be a better measure of scale economies but could not be constructed because of data limitations.

¹⁰ 10 sectors are distinguished: food and drink; coal products and chemicals; basic metals; mechanical engineering and vehicles; instruments and electrical engineering; metal products; textiles, leather and clothing; building materials, pottery, glass and wood products; paper products; and other manufacturing (the benchmark in Equation 1).

¹¹ Note that although several of these variables may directly affect labour productivity, there is no reason to expect that they might affect relative labour productivity. I check this later both by reporting the results of overidentification tests for the 2SLS regressions and by running OLS regressions of RELPROD with the entire set of instruments used as regressors. Either way, there is no evidence of any direct effect of the instruments on RELPROD.

¹² The data on capital stock are estimates at the three-digit level of aggregation rather than primary data and were taken from O’Mahoney and Oulton (1990). They were not available for 1951, so 1954 estimates were used instead and combined with employment data from the 1954 Census of Production.

¹³ While the advertising–sales ratio and the R&D–sales ratio are endogenous, it is generally exogenous industry characteristics that will determine whether these ratios are above or below 1% (or 2%). Thus, in an industry below the 1% cut-off point, advertising is not very effective in raising consumers’ willingness to pay or there is little scope for technological innovation from within the industry. On the other hand, in an industry above the 1% cut-off point, advertising/R&D ‘works’. Of course, whether such an industry has an advertising–sales ratio or R&D–sales ratio of 5% or 10%, say, may be largely determined endogenously. But my binary variables ADV and RD are not very sensitive to endogenous factors that affect advertising and R&D intensity. The procedure for constructing RD and ADV involved combining information from various official and market research sources; see Symeonidis (2003) for details and a list of the sources used.

¹⁴ The group of industries with high protection contains the engineering industries, instruments, vehicles, finished metal goods, some chemicals, paper and paper products, furniture, pottery and glass, most finished textile goods, rubber products and various other finished manufactures. The low-protection group contains most food and drink industries, some chemicals, basic metals, clothing and footwear, wood products, publishing, leather and most textile semi-manufactures and building materials.

¹⁵ The difference in n between the various columns of Table 2 (or Table 4) is due to missing data for some of the additional variables used.

¹⁶ When either CONC or the sector dummies are omitted, the coefficient on COLL is still everywhere negative but smaller (and statistically significant at the 10% level at best). This is not surprising; both CONC and the sector dummies are correlated with COLL as well as having a direct effect on RELPROD, so their omission causes a bias in the estimated coefficient on COLL.

¹⁷ The abolition of cartels and the resulting intensification of competition caused concentration to rise in previously collusive industries. On the other hand, there is no evidence of concentration facilitating collusion after controlling for capital intensity (Symeonidis, 2002, 2003).