Dear All,
I have a time series model, which was modeled as a multple linear regression, the seasonal effect was asjusted by the monthly dummy variables. After all the potential predictor variables in the model, there was no autocorrelation in the residuals as suggested from SAS output, and four normality tests were passed. However the residuals looked nonconstant and a bit fanning out could be observed. My first thoughts were, this could be 1. outliers, 2. heteroscedasticity.

I firstly adjusted one or two outliers in the model, then all assumptions of regression passed, no arch effect , no autocorrelatiion in the residuals, and normal test passed as well.

My questions then are:

1. Should I study heteroscedasticity first or outliers ??

2. The test for arch effect without outliers adjusted showed the potential heteroscedasticity, however after modelling the residual of regression using GARCH, the final residual didn't improve much. How can we tell the outliers are the main cause or the non-constant residual is.....

Thanks