However, testing for Granger non causality this way might result in inconsistent estimates; primarily because the
correlation between lagged values of dependent variable and the disturbance terms are not properly taken into
account (see Kar et al., 2011: 688-689). In the presence of correlated disturbances with the lagged values of
dependent variable, estimation by within estimator would be inconsistent when T and N are small and it is consistent
when T and N tend to infinity. A first alternative is to use Generalized Method of Moments (GMM) estimators, as
long as T is big enough.