Comparison of univariate and multivariate Granger causality in international asset pricing. Evidence from Finnish and Japanese financial economies

The study compares multivariate and univariate tests of causality in a rolling framework in testing whether the Japanese stock market 'causes' the Finnish cash and derivatives markets in the Granger sense. The multivariate algorithm generates the time pattern of causality of the underlying vector process. Significant causality is observed at distinct time intervals within the sample period, possibly during periods of regime switches, trend changes or major global disturbances. The multivariate causality is then decomposed into four univariate causality tests. The effect of each univariate causality pattern on the multivariate causality is tested by a recursive least squares regression (RLS) method.