The relationship between the F-test and the Schwarz criterion: Implications for Granger-causality tests

In applied research, the Schwarz Bayesian Information Criterion (BIC) and the F-test might yield different inferences about the causal relationships being investigated. This paper examines the relationship between the BIC and the F-tests in the context of Granger-causality tests. We calculate the F-test significance levels as a function of the model dimensionality and the sample size that would lead to the same conclusion as the BIC. We illustrate that the BIC would reject the null hypothesis of no-causality less often compared to an F-test conducted at five percent significance level for sample sizes above 50 especially when the chosen model dimensionality is small. Putting the philosophical issues aside, we suggest that the decision to choose between the F-test and the BIC should be made in view of the sample size.