If you model a time series Yt using a stationary ARMA process
with a nonzero constant (µ unequal to 0) and use it to forecast
future values of Yt, then as you forecast further and further into
the future, the confidence interval widths for your forecasts
will
(a) continue to increase and eventually reach arbitrarily large
values.
(b) gradually decay to zero.
(c) cutoff to zero after some lag.
(d) converge to a non-zero limiting value.
The correct answer is option (d) i.e.
If you model a time series Yt using a stationary ARMA process with a nonzero constant (µ unequal to 0) and use it to forecast future values of Yt, then as you forecast further and further into the future, the confidence interval widths for your forecasts will converge to a non-zero limiting value.
If you model a time series Yt using a stationary ARMA process with a nonzero constant...