Forecasting under non-gaussianity

[nazif]

Dear Tom,
I urgently need your valuable suggestions about my inflation forecasting paper under reviewing process. in that paper we are comparing forecasting accuracy of alternative models ANN, ARIMA and unobserved components models.
Referee reports are quite positive but they want additional estimates. One of the referees suggest us to use a forecasting model that would allow asymmetry and heavy tales, however he did not explicitly refer to any methodology. The only clue that we have just this sentence. Do you suggest any method to deal with asymmetry in the data? is it possible to use GARCH type models in this case? If so do you have any written RATS code for dynamic and and static forecasting with GARCH models?
Any help would be greatly appreciated.
Thanks
ps I'm using RATS 7.1

[TomDoan]

Dear Tom,
I urgently need your valuable suggestions about my inflation forecasting paper under reviewing process. in that paper we are comparing forecasting accuracy of alternative models ANN, ARIMA and unobserved components models.
Referee reports are quite positive but they want additional estimates. One of the referees suggest us to use a forecasting model that would allow asymmetry and heavy tales, however he did not explicitly refer to any methodology. The only clue that we have just this sentence. Do you suggest any method to deal with asymmetry in the data? is it possible to use GARCH type models in this case? If so do you have any written RATS code for dynamic and and static forecasting with GARCH models?
Any help would be greatly appreciated.
Thanks
ps I'm using RATS 7.1

The forecasts for an equaiton don't change if you have a GARCH error process. Perhaps what they mean is that you should estimate the model allowing for a GARCH process. That will give you a different set of estimates, which presumably would be less sensitive to the behavior of the data where there are clusters of big residuals. If you estimate an equation or model (for multivariate) using GARCH, you can then use FORECAST on it exactly as if you estimated it with LINREG or ESTIMATE.