Estimating Standard Errors for FEVD's of VECM parameters

[Andrew Slaven]

Good Morning,

I am currently running variance decompositions on some parameters in my estimated VAR and VEC models. I have found that when running a VAR model, I am able to generate Monte-Carlo standard errors along with my estimates. However, when running a VECM model, I can only generate estimates, as any attempt to produce standard errors receives a warning of 'Matrix size mismatch'.

My query is whether this is a problem can be remedied through a different specification or if Monte-Carlo simulations are simply not compatible with Error Correction Models.

many thanks,

Andrew

[EViews Mirza]

Hello Andrew!

Would you mind sharing your WF and the exact steps (code) which reproduce the unintended behaviour?

Kind regards,
Mirza

[Andrew Slaven]

Please may you let me know what WF stands for, apologies this is fairly new to me. In terms of the exact steps, I did not use code I simply used the drop down menus to specify my model, I will outline those steps now in case they are of use:

1) Select the variables I wish to model ( 7 variable system )
2) Open and VAR in levels to determine lag order + Cointegrating vectors
3) If cointegration is present run a VEC (p-1) in Levels with r cointegrating vectors
4) Assuming model is stable, homoscedastic and without autocorrelation at p-1 lags, select the variance decomposition tab and select add Monte-Carlo standard errors with 999 repetitions ( warning persists at lower and higher repetitions )

The warning only pops up when I use a VEC model, when running VAR in differences in the event that there are no cointegrating relationships then I am able to generate error bands without issue. The variables and process does not change for a VAR model, apart from the use of first differences and I am able to generate variance decompositions without errors in both cases.

I am using EViews 12

many thanks,

Andrew

[EViews Mirza]

WF stands for "workfile". In any case, I was able to replicate the behaviour you're experiencing. It appears that this is an issue in EViews 12. EViews 13 is correctly handling this scenario.

The issue is that FEVD standard errors are currently not supported with VEC models. The behaviour you're experiencing is due to the fact that we didn't invalidate the Monte Carlo SE dialog option in the variance decomposition dialog when the underlying estimation type is a VEC. We'll correct for this and issue a patch.

[Andrew Slaven]

Understood, thank you for responding so quickly. Are there currently any alternatives to determining the significance of FEVD's of VEC models currently supported by EViews version 13? I shall be sure to update my version in any case.

again, many thanks,

Andrew