RE: st: Forecasting with montly data

[Cameron McIntosh <cnm100@hotmail.com>]

Perhaps, given that you're not going too far into the future, but there are some subtleties to consider. I would consider a six-month forecast to be "medium" range, whereas exponentials are best for short range (e.g., one month), but you should be OK with exponential... maybe don't just use single (simple) exponential. Depending on what the data look like, you should maybe consider double (trend) or triple exponential (trend and seasonality). Some of the smoothing methods available in Stata are presented in the Yaffee paper I cite below. Also see:

Box, G., Jenkins, G.M., & Reinsel, G. (2008). Time Series Analysis: Forecasting & Control (4th ed.) Hoboken, NJ: Wiley.

Kantz, H., & Schreiber, T. (2003). Nonlinear Time Series Analysis (2nd ed.). Cambridge, UK: Cambridge University Press.

and these links:

http://www.stata.com/features/time-series/
http://www.stata.com/bookstore/ts.html
http://www.stata.com/bookstore/imtsa.html
Cam
----------------------------------------
> Date: Sat, 7 Jan 2012 16:31:40 -0800
> From: lisa_owen100@yahoo.ca
> Subject: Re: st: Forecasting with montly data
> To: statalist@hsphsun2.harvard.edu
>
> I dont have individual level data. I cant use ARMA/ARIMA because of the small time periods. How about exponential smoothing?
>
> Regards
> Lisa Owen
>
>
> ----- Original Message -----
> From: Cameron McIntosh 
> To: STATA LIST 
> Cc:
> Sent: Saturday, January 7, 2012 4:41:14 PM
> Subject: RE: st: Forecasting with montly data
>
> Hi Lisa,
> Well, as the simplest method, you could just extrapolate from the existing trend, say by regressing visits on time to get average monthly change and then away you go with your six-month prediction of the state of the world. But that's pretty crude. I'm assuming, though, that you also have some (time-stable and time-varying) covariates that you could use to model the trend -- that would help in building a more complex prediction model. Do you also have some other aggregate outcomes that you could model as co-evolving with doctor visits? Too bad you don't have individual-level data, as a microsimulation would be a useful approach (following a statistical model to get parameter estimates).
> Anyway, take a look at:
>
> Yaffee, R.A. (2007). Stata 10 (Time Series and Forecasting). Journal of Statistical Software, 23, Software Review 1.http://www.jstatsoft.org/v23/s01/paper
>
> Yaffee's book on the subject is also soon to be available:
>
> http://www.amazon.ca/Introduction-Forecasting-Time-Using-Stata/dp/1597180157
>
> Cam
>
> > Date: Sat, 7 Jan 2012 06:52:43 -0800
> > From: lisa_owen100@yahoo.ca
> > Subject: st: Forecasting with montly data
> > To: statalist@hsphsun2.harvard.edu
> >
> > Dear Statalist,
> > Happy new year. I have montly aggregate data for total number of doctor visits in a district for 12 months. I want to use this data to forecast what the number of visits will be in the next 6 months, if possibe make a monthly prediction for each month with update.
> > Please can somebody suggest what is the appropriate method of doing this in stata.
> > Thanks
> > Lisa Owen
> > *
> > *  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/
>
> *
> *  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/
>
> *
> * 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/
 		 	   		  
*
*   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/