# Why is the value of SE for the transformed regression model higher than the initial model with autocorrelation?

Is it right that with positive autocorrelation in the errors, the model underestimates the SE? Hence, using generalized differencing (such as Cochrane-Orcutt), the transformed model has a higher value of SE? And if so, what happens to a model with negative autocorrelation? I can't quite explain how these happen.

• Could you provide more details? E.g. autocorrelation in what: dependent variable, independent variables, errors? Also, what do you mean by generalized differencing? May 16 at 16:43
• Autocorrelation in errors and generalized differencing such as Cochrane-Orcutt May 17 at 1:10