Causality tests and conditional heteroskedasticity:: Monte Carlo evidence

This paper investigates the reliability of causality tests based on least squares when conditional heteroskedasticity exists. Monte Carlo evidence documents considerable size distortion if the conditional variances are correlated. Inference based on a heteroskedasticity and autocorrelation consistent covariance matrix offers little improvement. This size distortion traces to an inability to discriminate between causality in mean and causality in variance. As a result, this paper endorses conducting causality tests based on an empirical specification that explicitly models the conditional means and conditional variances. The relationship between money and prices serves as an illustrative example.