Heteroskedastic Residuals

What should we do if an ARIMA/SARIMA model has heteroskedastic residuals?

Answer

If we fit an ARIMA or SARIMA model and the residuals are not white noise and exhibit heteroskedastic (time‐varying) variance, this indicates that the model has not fully captured the dynamics of the series.

In such cases, we need to model the variance of the residuals separately. This can be done by using the ARCH (Autoregressive Conditional Heteroskedasticity) or GARCH (Generalized ARCH) models (see below).

Back to collection