How Large are the Effects of Simultaneity on Testing Granger Causality?

Abstract

Interpreting Granger causality as economic causality implies that the underlying VAR model is a structural economic model. This is wrong in the case of simultaneity, and causal conclusions can be misleading. Nevertheless, the empirical relevance of this problem still needs to be determined. Therefore, the magnitude and stability of possible errors are analysed in a simulation study. It is shown that economic misinterpretations of tests of Granger causality can occur with probability one for realistic parameter values. Thus, policy conclusions from Granger causality tests can be wrong with probability one if simultaneity occurs. Furthermore, the power of the test can be rather low even with a sample size of T = 50.