I am using STATA 13 for windows. I have run an arima model: arima lcurr, arima(3,1,1) noconstant where lcurr is in natural logs of currency in circulation. My data are monthly and I generate forecasts (using forecast estimates solve) 48 periods out into the future.
arima lcurr, arima(3,1,1) <---actual data available from January 1994 to April 2016
estimates store currency
tsappend, last(2020m12) tsfmt(tm) <--so adding rows for data from May 2016 through December 2020
forecast create fcurrency, replace
forecast estimates currency, names(flcurr)
forecast solve, prefix(f_) simulate(errors, statistic(stddev,prefix(sd_)) reps(50))
gen upper=f_flcurr+(1.96*sd_lcurr)
gen lower=f_flcurr-(1.96*sd_lcurr)
I then use the above information to create levels (since all the results thus far have been in logged differences rather than log levels). I'm expecting the prediction bands to widen the further out into my forecast horizon, but it remains quite narrow.
Any advice?
This is my first time posting so I may not be providing enough info.
arima lcurr, arima(3,1,1) <---actual data available from January 1994 to April 2016
estimates store currency
tsappend, last(2020m12) tsfmt(tm) <--so adding rows for data from May 2016 through December 2020
forecast create fcurrency, replace
forecast estimates currency, names(flcurr)
forecast solve, prefix(f_) simulate(errors, statistic(stddev,prefix(sd_)) reps(50))
gen upper=f_flcurr+(1.96*sd_lcurr)
gen lower=f_flcurr-(1.96*sd_lcurr)
I then use the above information to create levels (since all the results thus far have been in logged differences rather than log levels). I'm expecting the prediction bands to widen the further out into my forecast horizon, but it remains quite narrow.
Any advice?
This is my first time posting so I may not be providing enough info.
Comment