Okun's law in panels of countries and states

Abstract

This article contributes to the empirical literature of Okun's law in three respects. First, in contrast to the limited data used in the existing studies, we employ two extensive (across countries and across states, i.e. within a country) panel data sets to investigate the validity of Okun's law. Second, the use of the Pooled Mean Group (PMG) estimator permits us not to pre-filter the data as often done in the current literature, and can take into account the possibility of cointegration between unemployment and output. Third, in addition to the short-run relationship or cyclical components between unemployment and output, we also estimate the long-run linkage between these two important variables. Empirical results show that unemployment and output are long-run cointegrated, irrespective of using country- or state-level data. Moreover, the unemployment-output linkages are found to be negative and highly significant both in the short- and long-run. Our results not only confirm the validity of Okun's law (in the short-run) but also point out that a similar tradeoff exists in the long run.

Keywords:

• Okun's law
• panel data
• pooled mean group estimator

JEL Classification:

• C23
• E24

Acknowledgements

The title is partially borrowed from Barro (1991) and Glaeser et al. (1995). The authors are grateful to Edward F. Blackburne and Mark W. Frank (2007) for making publicly available the Stata (xtpmg) code used in this article. The very constructive comments and suggestions from an anonymous referee are also highly appreciated. Any remaining errors are our own responsibility.

Notes

¹By allowing for both features, Huang and Chang (2005) implement a structural change with threshold approach to Canada data, and provide evidence supporting that the Okun's relationship is not only unstable but also nonlinear.

²Additional advantage of the PMG approach is that we do not need to pre-test the existence of (panel) unit roots. The methodology is equally applicable irrespective of stationary or nonstationary variables being used.

³However, for some countries, there are some missing data (gaps) between different time periods. In this case, we use the interpolation method to fill the ‘gap’. In the robustness check, we drop these countries to see if the main results are altered.

⁴ For brevity, the results of Hausman tests and the outcomes obtained by the MG estimators are omitted, but available upon request.

⁵ Note that the Hausman test statistics appear to be negative for Far West and Mideast.