The consequences of a potential pitfall in approximate GLS estimation of the regression model with ARMA(p, q) disturbances