This paper points out that F statistic for testing Granger causality often
leads to spurious causality between two independent and irrelative processe
s {x(t)} and {y(t)} where one of or both of them is or are non-stationary.
To show this we derive non-standard asymptotic distributions of F statistic
s and simulate the distributions of them by Monte Carlo experiment under th
ree simple data generating processes. (C) 2001 Elsevier Science B.V. All ri
ghts reserved.