Re: st: Models ARIMA in STATA and Eviews

["Brian P. Poi" <bpoi@stata.com>]

On Fri, 21 Sep 2007, Joao Ricardo F. Lima wrote:

> Hi,
>
> I have one doubt with models ARMA-ARIMA-SARIMA. When i'm using eviews, i
> have one result. When i use Stata, the same model (ex arima y,
> arima(2,1,2)), the results are different. Sometimes, very different!
>
> Can help me and explais these differences, please?
>
> Thanks,
>
> Joao Lima
Getting different packages to report identical results when fitting ARIMA 
models is a challenge for several reasons.  There are several different 
estimators available for ARIMA models, including unconditional maximum 
likelihood, conditional maximum likelihood, and least squares.  Moreover, 
the choice of priming values or priors for the Kalman filter recursions 
will affect results, as will the choice of initial values used in the 
maximization algorithm.  Stata's full maximum likelihood results match 
those you would obtain using SAS.

In short, there are many variations in how ARIMA models are fit, and 
without knowing the implementation details of a package's ARIMA commands, 
it is hard to say whether appropriate options can be used with -arima- to 
match the other package's results.

A paper by Newbold, Agiakloglou, and Miller (1994, "Adventures with ARIMA 
software," Int'l. Journal of Forecasting, vol. 10) addresses some of the 
issues that arise when fitting ARIMA models.

   -- 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/