Testing for Granger causality in heterogeneous panels with cross-sectional dependence

Abstract

This paper proposes a panel Granger causality approach for heterogeneous panels with cross-sectional dependence. We define a panel VAR model with unobserved common factors and apply the PANIC procedure to obtain the de-factored data. We then estimate the lag augmented (LA)-VAR model for each cross section and construct the panel statistics based on the meta-analytic approach that combines the p-values of the individual statistics. The Monte Carlo simulations indicate that the combination tests show good size and power properties and appear suitable for the panels where cross sections may have different unit root or co-integration properties. We finally re-investigate Granger causality between export and economic growth in OECD countries. The results shed light on the importance of accounting for cross-sectional dependence within a factor model framework in determining direction of Granger causality for country-specific analysis. The results further reveal that export and economic growth do not cause each other in the majority of the European Union countries.