VAR Residual Portmanteau Tests for Autocor E6 vs E7

[skifozoa]

I get different results switching between Eviews 7 and Eviews 6 for the portmanteau test for autocorrelation in VAR model.
Note that the difference is situated in the degrees of freedom and following the probability levels. All the other values are identical.
Can anyone explain the difference ? More importantly which version is correct cause I'm preparing an exam.
(My university offers both through VPN. My teammate used eviews6 I used eviews7 and this is how I noticed the difference.)



Eviews 6:

Code: Select all

VAR Residual Portmanteau Tests for Autocorrelations Null Hypothesis: no residual autocorrelations up to lag h Date: 08/25/10 Time: 23:56 Sample: 1959Q1 1992Q1 Included observations: 130 Lags Q-Stat Prob. Adj Q-Stat Prob. df 1 0.137890 NA* 0.138959 NA* NA* 2 1.219885 NA* 1.237860 NA* NA* 3 8.480223 0.4866 8.669702 0.4683 9 4 23.83237 0.1606 24.50922 0.1390 18 5 41.92226 0.0335 43.32271 0.0243 27 6 52.70800 0.0357 54.63033 0.0240 36 7 60.38041 0.0624 62.73939 0.0412 45 8 76.65813 0.0230 80.08450 0.0121 54 9 83.77851 0.0411 87.73450 0.0214 63 10 89.99748 0.0742 94.47171 0.0391 72 11 93.35951 0.1642 98.14452 0.0945 81 12 95.71260 0.3204 100.7369 0.2062 90 *The test is valid only for lags larger than the VAR lag order. df is degrees of freedom for (approximate) chi-square distribution

Eviews7

Code: Select all

VAR Residual Portmanteau Tests for Autocorrelations Null Hypothesis: no residual autocorrelations up to lag h Date: 08/25/10 Time: 23:54 Sample: 1959Q1 1992Q1 Included observations: 130 Lags Q-Stat Prob. Adj Q-Stat Prob. df 1 0.137890 NA* 0.138959 NA* NA* 2 1.219885 NA* 1.237860 NA* NA* 3 8.480223 0.9333 8.669702 0.9264 16 4 23.83237 0.5291 24.50922 0.4901 25 5 41.92226 0.1649 43.32271 0.1313 34 6 52.70800 0.1474 54.63033 0.1100 43 7 60.38041 0.1988 62.73939 0.1463 52 8 76.65813 0.0852 80.08450 0.0512 61 9 83.77851 0.1248 87.73450 0.0744 70 10 89.99748 0.1868 94.47171 0.1130 79 11 93.35951 0.3278 98.14452 0.2157 88 12 95.71260 0.5179 100.7369 0.3773 97 *The test is valid only for lags larger than the VAR lag order. df is degrees of freedom for (approximate) chi-square distribution

[EViews Gareth]

There was a bug in the DF calculation in early versions of EViews 6. Thus it isn't a difference between EViews 7 and EViews 6 per se, rather it is just the case that your copy of EViews 6 is not up-to-date.

The EViews 7 results should be correct.

[skifozoa]

Thanks for your swift response.

[skifozoa]

Actually, I have an additional question.

The residual serial correlation LM test for VAR tests about the same right? That there is no autocorrelation left in your residuals.

How come, if Eviews7 calculates the correct amount of degrees of freedom that it is actually the Eviews 6 output that is more in line with the LM results ?
These are the LM test results for the same VAR(0) regression. One would conclude that there might be some autocorrelation left. This is in line with the Eviews6 probabilities of the Q statistic and not the Eviews7 results.
Can you explain this? I don't know the exact differences in these two tests (Q versus LM) since the course is about being able to produce and interpret output rather than the technical statistical details at the base of these results... so maybe I'm confused over nothing.
Any insights are sincerely appreciated.

Code: Select all

VAR Residual Serial Correlation LM Tests Null Hypothesis: no serial correlation at lag order h Date: 08/26/10 Time: 01:07 Sample: 1959Q1 1992Q1 Included observations: 132 Lags LM-Stat Prob 1 21.60942 0.0102 2 5.031123 0.8316 3 4.417986 0.8818 4 18.08444 0.0342 5 15.04170 0.0898 6 12.77500 0.1731 7 8.602617 0.4747 8 15.78411 0.0715 9 7.472913 0.5880 10 6.544042 0.6845 11 6.016431 0.7383 12 4.569485 0.8701 Probs from chi-square with 9 df.