Dear statalist expert,
I am using the pvar command to estimate a panel VAR model. Before estimation, I ran the pvarsoc command to decide optimal lag length and the output returns a Hansen's J-stat which is quite different from that one I obtain after running pvar with its overid option.
The pvarsoc output and code:
pvarsoc NFFGDP lnRPL , maxlag(5) pinstlag(2/10) exog( lnRGDPCH)
Running panel VAR lag order selection on estimation sample
.....
Selection order criteria
Sample: 16 - 45 No. of obs
> = 882
No. of panels
> = 35
Ave. no. of T
> = 25.200
+---------------------------------------------------------------+
| lag | CD J J pvalue MBIC MAIC |
|-------+-------------------------------------------------------|
| 1 | .9858371 44.09921 .0754545 -172.9309 -19.90079 |
| 2 | .9832904 23.90654 .686459 -165.9948 -32.09346 |
| 3 | .9786814 22.31654 .5603802 -140.4561 -25.68346 |
| 4 | .9759816 16.7584 .6686012 -118.8854 -23.2416 |
| 5 | .9301767 13.28738 .6516404 -95.24582 -18.71262 |
+---------------------------------------------------------------+ +-------------------+
| lag | MQIC |
|-------+-----------|
| 1 | -78.41601 |
| 2 | -83.29428 |
| 3 | -69.56988 |
| 4 | -59.81361 |
| 5 | -47.97557 |
The pvar code and its output
pvar NFFGDP lnRPL , lags(1) gmmstyle instlags(2/10) overid exog(lnRGDPCH)
....
Test of overidentifying restriction:
Hansen's J chi2(32) = 89.982543 (p = 0.000)
In sum, I run the model with one lag based on the MBIC and the associated J-stat has p-value of 0.075. However when I run the model, the option "overid" reports a J-stat with zero p-value.
Any idea about what could be the discrepancy between these two incompatible reports? Did I commit any mistake in running the model? Which J-stat should be relliable one?
Best regards,
Sait.
I am using the pvar command to estimate a panel VAR model. Before estimation, I ran the pvarsoc command to decide optimal lag length and the output returns a Hansen's J-stat which is quite different from that one I obtain after running pvar with its overid option.
The pvarsoc output and code:
pvarsoc NFFGDP lnRPL , maxlag(5) pinstlag(2/10) exog( lnRGDPCH)
Running panel VAR lag order selection on estimation sample
.....
Selection order criteria
Sample: 16 - 45 No. of obs
> = 882
No. of panels
> = 35
Ave. no. of T
> = 25.200
+---------------------------------------------------------------+
| lag | CD J J pvalue MBIC MAIC |
|-------+-------------------------------------------------------|
| 1 | .9858371 44.09921 .0754545 -172.9309 -19.90079 |
| 2 | .9832904 23.90654 .686459 -165.9948 -32.09346 |
| 3 | .9786814 22.31654 .5603802 -140.4561 -25.68346 |
| 4 | .9759816 16.7584 .6686012 -118.8854 -23.2416 |
| 5 | .9301767 13.28738 .6516404 -95.24582 -18.71262 |
+---------------------------------------------------------------+ +-------------------+
| lag | MQIC |
|-------+-----------|
| 1 | -78.41601 |
| 2 | -83.29428 |
| 3 | -69.56988 |
| 4 | -59.81361 |
| 5 | -47.97557 |
The pvar code and its output
pvar NFFGDP lnRPL , lags(1) gmmstyle instlags(2/10) overid exog(lnRGDPCH)
....
Test of overidentifying restriction:
Hansen's J chi2(32) = 89.982543 (p = 0.000)
In sum, I run the model with one lag based on the MBIC and the associated J-stat has p-value of 0.075. However when I run the model, the option "overid" reports a J-stat with zero p-value.
Any idea about what could be the discrepancy between these two incompatible reports? Did I commit any mistake in running the model? Which J-stat should be relliable one?
Best regards,
Sait.
