st: Testing a restriction with a Garch model turns problematic

[Mehmet Dicle <mfdicle@gmail.com>]

Im having a problem that may be more of a statistics issue than a Stata issue.
I am using Stata 12.1 for Windows.

I would like to test a restriction on a model similar to Granger type
non-causality tests.

I have the following code:

webuse stocks, clear

arch honda L(1/2).honda, arch(1) garch(1)
predict double resid_restricted, resid
predict double var_restricted, variance
gen double std_resid_restricted=resid_restricted/sqrt(var_restricted)

arch honda L(1/2).honda L(1/2).nissan, arch(1) garch(1)
predict double resid_unrestricted, resid
predict double var_unrestricted, variance
gen double std_resid_unrestricted=resid_unrestricted/sqrt(var_unrestricted)

My problem is with the variance of residuals. With the inclusion of
additional variables, the variance of residuals should decrease.
However, it is increasing.

tabstat std*, stat(variance)

   stats |  std_re..  s~unre~d
---------+--------------------
variance |  .9993687  .9993756
------------------------------

I would very much appreciate any explanation and/or directions why this is so.

Thanks,
Mehmet F. Dicle, Ph.D.
College of Business (Finance)
Loyola University New Orleans

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/