Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: Re: AIC and BIC values after ARIMA estimation

From   Alejandro Mosiño <[email protected]>
To   [email protected]
Subject   st: Re: AIC and BIC values after ARIMA estimation
Date   Thu, 27 Jan 2011 18:38:48 +0100

Thank you very much for your answer,

Actually yes, i cheked the signs. They are all negatives, so... believe me, that's is not the problem.

I understand that AICs and BICs are not comparable across software, but i think the decision of what model to choose should be (if not equal) very similar. In my particular case, i'm using data for the yen / dollar exchange rate. The results using Eviews are published in a book by Francis X. Diebold (Elements Forecasting, chapter 13). I replicated the analysis using the same data (that you can download from the author's web page or the book's companion site) on Gretl and Stata. As i told you, Gretl is giving me similar results, but Stata does not.

I'm a big fan of Stata, so, i don't know. Maybe there is another way of computing AIC and BIC. Any suggestions?



El 27/01/2011 04:18 p.m., Maarten buis escribió:
--- On Thu, 27/1/11, Alejandro Mosiño wrote:
I have this problem. I'm selecting ARIMA models by using
AIC and BIC criteria. However, i got very different results
from different softwares (Eviews, Gretl and Stata). In one
case for instance, Gretl and Eviews tell me that the right
model is an ARMA(2,0), while Stata tells me that it is an
ARMA(2,2). Another example, Gretl and Eviews tell me that
some other model should be an ARIMA(1,1,0), while Stata
tells me that the right model is an ARIMA(0,1,1).

Then, while Gretl and Eviews agree, Stata gives me VERY
different results. Do you know why?
The absolute numbers for (log) likelihood values and thus
BICs and AICs are often not comparable across software.
There is often a term in the likelihood function that does
not depend on the parameters, and programmers can choose
to leave that term in or out without changing the maximum.

Did you check the sign and adjusted what you call small
and large accordingly (it may sound like a silly mistake
to make, but you do not want to know (and I do not want to
tell) how often I made a mistake like that...)

Hope this helps,

Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen

*   For searches and help try:

*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index