Does Size Matter in Firm Performance? Evidence from US Public Firms

Abstract

This paper reexamines the determinants of firm performance and, in particular, the role that firm size plays in profitability. A fixed‐effects dynamic panel data model for over 7,000 US publicly‐held firms during the period 1987–2006 provides evidence that profit rates are positively correlated with firm size in a non‐linear manner, holding an array of firm‐ and industry‐specific characteristics constant. In addition, industry‐specific fixed effects play a negligible role in the presence of firm‐specific fixed effects.

Key Words:

• Profit Rates
• Panel Data
• Firm Size
• Industry Effects
• Firm Effects

JEL classifications:

• L25
• D21
• C61

Notes

1. In order to control for the effects of inflation over time, the nominal values of assets are deflated by the producer price index. All other variables are expressed in ratios instead of levels so that no corresponding adjustments are needed. Because many firms are multinationals and export to other countries so that using their total sales and assets to measure market power is misleading. From this perspective, we subtract the amounts of assets and sales associated with foreign operations to obtain firm data associated with operations only in the US. The foreign‐dependent data are obtained from Research Insight’s Geographic Segment database.

2. The Bureau of Census does not publish concentration data for the construction and agricultural sectors, so that we calculate the four‐firm concentration ratios for these two industries as the ratio of the total market shares of the largest four firms in the sample over the total industry receipts reported by the Census. Since the data are available only every five years, we interpolate the data using a weighted average method for intermittent years such that, for example, CR4₁₉₈₈ (concentration ratio for 1988) equals 0.75×CR4₁₉₈₇+0.25×CR4₁₉₉₂. Data beginning 2002 equal the values for 2002.

3. We have also looked for possible multicollinearity by examining the partial correlation matrix of the explanatory variables. Most estimates in absolute terms are below 0.5, suggesting no severe problem of multicollinearity among the regressors.

4. To avoid the singularity problem, the last industry on the list (‘unclassified’) does not have a dummy variable.

5. While the four‐digit level of NAICS industry definition is arbitrary, using the five‐digit level definition requires substantially more dummy variables, thus dramatically reducing the degrees of freedom in estimation.

6. See Greene (2000, chapter 16) for detailed discussions of panel data regression.

7. The likelihood ratio test statistic is computed as two times the difference of the log‐likelihood values. A χ ² test on coefficient restrictions also confirms that the industry dummy variables together provide no marginally significant explanatory power in the model for profit rates.