Am 26.02.2019 um 00:42 schrieb Allin Cottrell:
Use of "fcast" with panel data is now enabled (to varying
degrees) for
pooled OLS, fixed effects, random effects and dpanel.
Thanks for the summary, Allin.
Caveat: It's not clear what we should do if the model
specification
includes time dummies. At present we impute an out-of-sample time effect
equal to the mean of the in-sample time effects. This is obviously
debatable; feel free to debate it. Perhaps we should ban such prediction
altogether.
Spontaneously I'd say yes, ban them. A model with time fixed effects
wants to do it in a conditional way, but we cannot condition on the
unknown future. Also, you might say that the time dummy removes the mean
in a particular period across the units, but following that logic you
would have to demean the forecasts cross-sectionally and then label them
not as forecasts of the variables, but as forecasts of the deviations.
In any case, you wouldn't use the average of the past.
Now to out-of-sample in the cross-sectional dimension. What are we
going
to do with individual effects, for the fixed- and random-effects
estimators? For now, we set the predicted individual effect to the
global constant (for fixed effects) or to its expected value of zero
(for random effects). This is obviously another debatable point,
analogous to the out-of-sample time effect case.
Yes. Same answer from me for the fixed effects case. For random effects
it's different IMHO, because it is explicitly treated as a random
variable and as you say, its expectation in the cross-section is zero.
Of course, if you want to evaluate forecast uncertainty the variance of
the random intercept would have to be taken into account.
cheers
sven