Financial markets and firm size: The role of employment protection laws and barriers to entrepreneurship

ABSTRACT

This article provides evidence on the institutional determinants of firm size for the period 1980–1998. Using a comprehensive longitudinal database across 29 industrial sectors in 15 Organisation for Economic Co-operation and Development (OECD) countries, we study how labour regulations and barriers to entrepreneurship (BE) affect industrial organization in the presence of capital market frictions. We show that strict employment protection laws (EPL) and high BE negatively affect firm size in sectors that are more dependent on external funds. Our findings demonstrate that the interaction between market regulations and financial market imperfections help to explain some of the differences in firm structure across countries.

KEYWORDS:

• Institutions
• firm size
• financial imperfections
• labour costs
• entrepreneurship costs

JEL CLASSIFICATION:

• J21
• G20
• C33

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ A number of theoretical papers are also devoted to the optimal firm size (Gibrat’s study, 1931; Williamson 1967; Lucas 1978; Cabral and Mata 2003).

² Moreover, the literature on financial development data is more ambiguous with respect to the financial crisis of 2008. Therefore, we consider that it is clearer to investigate employment protection laws (EPL) and barriers to entrepreneurship (BE) interactions with financial development before 2000 and leave other questions with respect to institutions and financial markets in different economic contexts for further research.

³ We study 15 countries and 29 sectors, based on U.S. classification.

⁴ It is plausible, however, that some amounts of measurement error remains. In this case, we would obtain larger standard errors of the estimated coefficients, implying that our conclusions are conservative.

⁵ This measure is also used by Kumar, Rajan, and Zingales (2001).

⁶ Our main conclusions are robust if we use the traditional natural logarithm of the ratio between the number of employees by the number of establishments in the sector. In the article, we interchangeably use firms and establishments although we always measure the latter.

⁷ See Claessens and Laeven (2003).

⁸ See Rajan and Zingales (1998) to find various explanations of the causes of these different correlations.

⁹ For more detail of how measures are computed, see Botero et al. (2004).

¹⁰ Fonseca, Lopez-Garcia, and Pissarides (2001) and Nicoletti, Scarpetta, and Boylaud (2000) used Logotech data. Here, the start-up index is taken from Fonseca, Lopez-Garcia, and Pissarides (2001); Start-up index = [no. of weeks + no. of procedures/(average procedures per *week)/2].

¹¹ In practice, we use between 8 and 10 lags as instruments to minimize the risk of weak instruments. We test the adequacy of including only one lag of average firm size in the conditional mean by looking at the second-order serial correlation test, as is customary. Furthermore, since average firm size is likely to be quite persistent, we augment the set of moments with moments that impose mean stationarity on these outcomes. Blundell and Bond (1998) show how this improves on the efficiency of the ‘crude’ Arellano and Bond estimator.

¹² We keep the hypothesis of predetermined variables even if coefficients do not differ too much considering institutions exogenous or predetermined.

¹³ When the autoregressive component estimate is bounded from above by the generalized least squares (GLS) estimate and from below by the fixed effect estimate, it is likely that the finite-sample bias in generalized method of moments (GMM) estimators is small (Bond 2002).

¹⁴ In the GLS specification, we still include both country- and sector-specific fixed effects.

¹⁵ For market size, the difference in coefficients (βfixed effects – βrandom effects) is −0.18 with a standard error of 0.01148 using the assumption that the GLS estimator is efficient under the null. The difference for financial development is 0.069 and −0.076 for the interaction with EPL. In both cases, standard errors are low (0.012 and 0.013, respectively) such that we can reject the null of random effects. The within estimates imply relatively large changes in average size of firms following changes in market size. The estimated elasticity is close to unity.

¹⁶ In the GLS specification, this will lead to an upward bias in the estimate of ${\alpha }_S$. In the within case inclusion of the lagged average, firm size should bias downward the estimate of ${\alpha }_S$. Both situations could also affect other estimates.

¹⁷ Current firm size does not affect current financial development and labour regulations, but could affect future values of financial development and labour regulations.

¹⁸ The effect of market size is much smaller than in any of the specification in Table 5, but it is statistically significant. It implies a much lower elasticity of approximately 0.07.

¹⁹ The effect is, however, much lower than those estimated in Table 5.

²⁰ These measures, although widely used in the literature, are invariant over time. We have tried other institutional variables and robustness that other studies use without any large differences in the results. Nickell et al. (2003) collect EPL variables that vary over the period; however, they do not vary much. Further, the correlation between them is very high. For BE data, we have also tested robustness with Djankov et al. (2002) variables with similar results (available upon request).

Additional information

Funding

This work was supported by the Spanish Ministry of Education and Science [ECO2013-48496-C4-4-R, ECO2015-67999-R], the Regional Government of Aragón, the European Social Fund [S125 project: Compete], and the Centro Universitario de la Defensa Zaragoza [2013-08 project].