Dear Stata-sticians,

I am a little puzzled by Stata's calculation of the Mean Square Error when using a predict command following ARIMA estimation to dynamically predict the first 8 periods after the initial dataset ended

I am using the code:

arima D.lnwpi, ar(1 2 4)

tsappend, add(8)

predict dynamic, dynamic(tq(1991q1)) y

predict dynamic_mse, dynamic(tq(1991q1)) mse


What I don't really understand is how Stata is calculating MSE values for these predictions when there are no observed values against which to compare them... :-/ The Stata manual is really unhelpful in terms of how it calculates the MSE (which incidentally also seems to exhibit variation throughout the data), so either I fundamentally misunderstand what the MSE is, or Stata has a particular way of calculating it.

Either way, any help would be greatly appreciated!

Thanks,

Tim