The STATA manual states that predict res is the predicted innovation epsilon(t).
It is also my understanding that it is computed as the residual of the
prediction of the model:
res(t) = y(t) - E_(t-1) y(t).
With finite sample, this is not the same as the forecast at time t of
res(t); i.e.,
E_t res(t) need to equal to y(t) - E_(t-1) y(t). [By E_t, I mean the
expectation conditional on the values y(t), y(t-1), ..., y(0).]
For instance, is y(t) is MA(1) with no constant, E_t res(t) = E_t
y(t+1)/theta which in finite samples is different than y(t) - E_(t-1)
y_t.
Has anyone programmed E_t res(t)?
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/