ABSTRACT

In a 1983 paper, Geweke et al. compared eight alternative tests of the absence of causal ordering, in the context of stationary variables. All were versions of the tests proposed by Granger (1969) and Sims (1972). Their unambiguous finding was that Wald variants of the tests are preferred. This chapter is meant to extend the results of Geweke et al. (1983) in two directions. First, cases where the variables are not I(0) [in the sense of Engle and Granger (1987)], but also integrated of higher order and/or cointegrated, are considered. Second, in addition to classical in-sample tests of Granger noncausality, various model selection procedures for checking causal ordering are outlined and examined.