Autoregressive (AR) modelling is based on assumption of stationarity which implies homoscedasticity and constant mean. Heteroscedasticity makes AR modelling unreliable for forecasting. To study the sensitivity of AR modelling to heteroscedasticity can best be done by simulation.