AUTOCORRELATION CANNOT BE REMOVED

[sirooskhadem]

Hi,
I have been trying to remove the Autocorrelation from my model but it seems impossible. The best results come from an ARMA-EGARCH model with Student’s t error distribution. I have also tried with no pre-sampling but still nothing. The data is 15-MINUTES Australian Dollar exchange rate returns against USD with 5 working days dummy variables. When I check the correlogram standardized of residuals, AC and PAC are very close to zero but highly significant. There is also a note there saying;”*Probabilities may not be valid for this equation specification.” Which I don’t know what does it mean.

Any suggestion?
Thanks
Siroos

[eobumneke]

Hello, if you want to remove autocorrelation in a model, there are one out of the 3 basic things you could do: Run a general-to-specific model, or change the dynamic specification of the model or add sufficient lags to both the dependent and independent variables. Moreso, bear in mind that Durbin watson test for autocorrelation is used ONLY for first order autocorrelation! once any of such correction methods i earlier stated are utilized, you have to use LM test for autocorrelation. Regards.

[sirooskhadem]

Thank you very much for your reply.