I am currently doing a project on the link between transportation and economic growth. Particularly, am interested in the potential heterogeneoua nature of causality relations across regions. However, I have some uncertainties regarding the empirical strategy.
I have a strongly balanced panel of regions spanning 19 years with data on GDP and transport infrastructure.
According to panel unit root tests both variables are I(1). Moreover, panel cointegration test point towards cointegration. Would this mean that for every region a causal relationship exists in at least one direction? I have found papers that find non-causality in some regions with cointegrated variables after rejecting the hypothesis of homogeneous causality in a VAR setting. Is this common or a violation of what is expected from cointegrated data?
With cointegrated data it seems to me that the appropriate strategy would be to run a panel VECM and base Granger causality on that. Are there panel Granger procedures that allow for heterogeneous causality based on VEC models? (I know they exist for VAR settings.)