st: FW: VAR lag selection and lags autocorrelation

[Muhammad Akram <aasim548@hotmail.com>]

Dear Statlist et al,
 
I have few questions about lag order selection for var (vector autoregressive). I have preestimation varsoc with max lag 8 option. I got the following results
 
varsoc Dependant IND1 IND2 IND3, maxlag(8)
   Selection-order criteria
   Sample:  1989 - 2009                         Number of obs      =        21
  +---------------------------------------------------------------------------+
  |lag |    LL      LR      df    p      FPE       AIC      HQIC      SBIC    |
  |----+----------------------------------------------------------------------|
  |  0 | -198.518                      2796.63   19.2874   19.3306   19.4863  |
  |  1 | -144.204  108.63   16  0.000  75.5051   15.6384   15.8543   16.6332  |
  |  2 | -110.247  67.914   16  0.000  16.6656   13.9283   14.3169   15.7189  |
  |  3 | -93.6263  33.241   16  0.007  28.5872   13.8692   14.4305   16.4556  |
  |  4 | -28.3657  130.52   16  0.000  1.4264*  9.17768   9.91172   12.5599  |
  |  5 |  2275.82  4608.4   16  0.000        .  -208.745  -207.838  -204.567  |
  |  6 |  2451.92  352.19   16  0.000        .  -225.516  -224.609  -221.338  |
  |  7 |  2555.54  207.25   16  0.000        .  -235.385  -234.478  -231.207  |
  |  8 |  2571.33  31.577*  16  0.011        .  -236.889* -235.982* -232.711* |
  +---------------------------------------------------------------------------+

This result suggests 8 lags. but my data has only 29 observations for time means T<30. If I use more than 4lags it gives no t or p values in estimations. Moreover, with more than 3lags varstable test doesn't satisfies. 
 
Another question is about LM-test for autocorrelation.
 
varlmar, mlag(3)
   Lagrange-multiplier test
  +--------------------------------------+
  | lag  |      chi2    df   Prob > chi2 |
  |------+-------------------------------|
  |   1  |   26.8331    16     0.04338   |
  |   2  |   12.3209    16     0.72161   |
  |   3  |   14.3204    16     0.57486   |
  +--------------------------------------+
   H0: no autocorrelation at lag order
If there is autocorrelation at any of the lags what can be used as a solution in var model. Remember my data is non-stationary at level so I am using var at first difference for all variables. But for lag order selection I am using variables at level.
 
I would be greatly thankful for supportive replies.
 
Thanks
aasim


 

 

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