Re: st: Small Stata for teaching

 From "Brian P. Poi" To statalist@hsphsun2.harvard.edu Subject Re: st: Small Stata for teaching Date Wed, 15 Aug 2007 15:52:03 -0500 (CDT)

On Tue, 14 Aug 2007, Paulo Guimaraes wrote:

The course I am teaching is on economic forecasting. I am using the book
"Elements of Forecasting" by F. Diebold and contrary to the examples on the book (which are on Eviews) I decided to use Stata. Unfortunately, I am finding out that the students can not replicate some of the examples on the book because of the restrictions on Small Stata. For example, they are unable to estimate an ARMAX with 12 seasonal variables using 400 observations.

On Tue, 14 Aug 2007, Brian P. Poi wrote:
We fully intend for Small Stata to be able to run most examples that are encountered in textbooks; an ARMAX model with 12 variables should be within the grasp of Small Stata.
I will also report back to statalist once we arrive at a solution.

The solution to Paulo's problem of fitting ARMAX models with Small Stata is to specify "technique(bfgs)" or "technique(nr)" with the -arima- command.

By default -arima- and -arch- use the Berndt-Hall-Hall-Hausman (BHHH) optimization technique, and because of the way the likelihood evaluators for -arima- and -arch- are implemented, Stata needs to generate a double-precision temporary variable for each parameter in the model. In Paulo's model there were 18 parameters, and together with the original dataset and a couple of other temporary variables being created, Paulo was bumping up against Small Stata's maximum width of 200 bytes for each observation.

The BFGS and Newton-Raphson methods do not require the creation of score
variables for each parameter, so the observation width limit does not come
into play.

-- Brian Poi
-- bpoi@stata.com
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/