Partially heterogeneous tests for Granger non-causality in panel data

The power of Granger non-causality tests in panel data depends on the type of the alternative hypothesis: feedback from other variables might be homogeneous, homogeneous within groups or heterogeneous across different panel units. Existing tests have power against only one of these alternatives and may fail to reject the null hypothesis if the specified type of alternative is incorrect. This paper proposes a new Union-Intersections (UI) test which has correct size and good power against any type of alternative. The UI test is based on an existing test which is powerful against heterogeneous alternatives and a new Wald-type test which is powerful against homogeneous alternatives. The Wald test is designed to have good size and power properties for moderate to large time series dimensions and is based on a bias-corrected split panel jackknife-type estimator. Evidence from simulations confirm the new UI tests provide power against any direction of the alternative.

More: https://www.lb.lt/en/publications/no-59-arturas-juodis-yiannis-karavias-partially-heterogeneous-tests-for-granger-non-causality-in-panel-data