ShortPanel, too few individuals, singular

The Grunfeld example has fewer firms than time periods

estimating an unrestricted covariance matrix results in a reduced rank matrix and causes a failure of the Cholesky decomposition.

Stata produces numbers, but prints a warning "not converged"

>>> mod = ShortPanelGLS2(res.model.endog, res.model.exog, firm_id)
>>> res_p = mod.fit()
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "e:\josef\eclipsegworkspace\statsmodels-git\statsmodels-josef_new\statsmodels\sandbox\panel\panel_short.py", line 120, in fit
    self.cholsigmainv_i = np.linalg.cholesky(np.linalg.pinv(sigma_i)).T
  File "C:\Python26\lib\site-packages\numpy\linalg\linalg.py", line 531, in cholesky
    Cholesky decomposition cannot be computed'
numpy.linalg.linalg.LinAlgError: Matrix is not positive definite -         Cholesky decomposition cannot be computed

>>> sigma_i = mod.get_within_cov(mod.res_pooled.resid)
>>> np.linalg.eigvals(sigma_i)
array([ 118599.13113534,   28749.15411872,   10884.5047154 ,
          1567.26735086,     455.28185451,     283.0793974 ,
           118.27313134,      47.77001344,      44.56802788,
            34.72886823,       5.18697788,       0.        ,
            -0.        ,       0.        ,       0.        ,
            -0.        ,      -0.        ,      -0.        ,
            -0.        ,       0.        ])

Start with statsmodels/sandbox/panel/panel_short.py at the fit path for ShortPanelGLS2 and reproduce the Grunfeld example with fewer firms than time periods. Investigate the reduced-rank covariance and Cholesky failure; done should mean this case no longer fails unexpectedly and its behavior is covered by an appropriate regression test.