Institutional determinants of public–private sector wages’ linkages

Abstract

We estimate the probability of public sector wage leadership – defined as Granger causality from public to private sector wages – in a pool of 15 Organization for Economic Cooperation and Development (OECD) countries as a function of countries’ institutional features, and notably wage-setting institutions. Government's involvement in collective bargaining and collective bargaining centralization are positively correlated with the probability of finding public wage leadership. Among the factors that reduce its probability, we can underline the impact of globalization and the degree of unionization of the economy.

Keywords:

• labour market institutions
• government wages
• private sector wages
• causality

JEL Classification:

• J30
• J51
• J52
• E62
• E63
• H50
• H6

Acknowledgements

We thank Ludger Schuknecht for helpful comments and discussions.

Notes

¹ The measure of wages in this article is compensation per employee. Compensation per employee is computed using compensation of employees and employment data. Compensation of private sector employees is defined as total economy compensation of employees minus compensation of government employees. Compensation per private employee is defined as private compensation of employees divided by total employees minus government employment minus self-employment.

² Lindquist and Vilhelmsson (2004) and Lamo et al. (2012) explore a different concept of public sector leadership, based on vector error correction models (VECM).

³ The original wage data, that are then detrended in Lamo et al. (2013a, b) is a standard OECD data set that includes 18 countries. In this article, we focus on the 15 countries for which institutional information is available, namely, Germany, France, Italy, Spain, Netherlands, Austria, Belgium, Ireland, Portugal, Finland, Sweden, Denmark, United States, United Kingdom and Japan.

⁴ These authors show that a standard Wald test can be used to test linear constraints in this framework by just adding an extra lag in estimating the parameters of the process. This approach is quite appealing because the least-squared estimation may be applied to the levels of the model. To carry out the causality test, it is not necessary to perform a VEC reparameterization of the process to account for cointegration, because the least-square estimators of the relevant matrices do not change due to the reparameterization.