T-statistics in AR process

[nubbey554]

Hey I was wondering if someone could help me. I have two questions:

I have estimated an AR(1) process, and eviews outputs t-statistics for the estimated coefficients.
-In an AR process can these t-statistics be interpreted like a non-stochastic model, ie do they test if the coefficient on the dependent lagged variable is equal to 0?
-My other question is if they are interpreted like a normal non-dynamic regression, why do we do a dickey fuller test instead of just a t-test to see if the coefficient is equal to 1 ??

Thanks so much for any help provided! Loving the software.

[EViews Glenn]

I'm not certain what you mean by "like a non-stochastic model". The t-statistics test for an AR coefficient whether that specific dependence in the error structure is present in the error when estimated under the alternative hypothesis (in contrast to the serial correlation LM test which evaluates under the null). It is a test of the coefficient on the lagged dependent variable only in pure AR models with no regressors apart from the intercept.

You can't use this statistic for testing unit roots since under the null that the first-order AR coefficient is equal to 1, the distribution of the statistic does not follow the conventional t-distribution.

[nubbey554]

Thanks a lot glenn, really appreciate the fast reply. So we cannot do a T test as the dependent variable does not follow a T distribution as in the DF test our null is the variable is non-stationary yes ?