Newey-West standard error correction in VAR

[fabian]

Hi All,

I am currently trying to estimate a VAR with 3 variables (and suggested is roughly 20 lags), but have found out that I need to correct for heteroscedasticity and preferrably also for autocorrelation between my error terms. The dataset(s) contain over 20000 obs.

I am using Eviews 6. (if an easy solution in Eviews 7 exists I am able to use that as well)

Is there any way I can use Newey-West corrected standard-errors in my VAR and later plot the Impulse response and Variance-decomposition?
Or is there a way to plot the Impulse Responses and correct t-stats for Newey-West later?

Perhaps if I use OLS on all 3 equations first? Or if there is some piece of code that can help me?

Any help is much appreciated, If more information is needed let me know.

Many Thanks!

Fabian

[EViews Gareth]

There is nothing built in that will do it for you.

[fabian]

Thanks for the quick reply,

Is there any solution to this problem with a workaround in that case?
I have seen some papers where is stated that they use the Newey-West corrected standard errors when dealing with heteroskedastic and/or autocorrelated VAR residuals..

I am kind of stuck on this, so any suggestions are welcome, even if it's not that easy to implement

Thanks

[EViews Gareth]

I can't think of a workaround, especially if you want IRFs.

[fabian]

Thanks again,

Now say that I can remove the autocorrelation if I add more lags (which should not be to much of an issue given my rather large # of observations I think).
Is it then possible to compensate the standard errors only for heteroskedasticity (i.e. use White heteroskedasticity-robust standard errors)??
Direct/using a workaround/program something myself?

I just don't fully get it, if the VAR is basically built up from 3 lagged OLS regressions (where you can correct for HAC) why this should be in no way possible when estimating as a VAR..

Many thanks for your help!

[EViews Gareth]

Because the code that estimates a VAR is completely separate from the code that estimates OLS.

[huugh]

I have a similar problem. I intended to estimate it as a system (thus with OLS), but the HAC options are greyed out and not available. Why?

[EViews Gareth]

Because EViews doesn't offer HAC standard errors on systems estimated by OLS.

[EViews Glenn]

But you can specify a GMM model that is equivalent to the OLS specification (use all of the regressors as instruments and use TSLS weighting), and obtain the robust standard errors that way.

[KElizabeth925]

If using EViews 8--is there any way to run a VAR utilising the Newey-West errror correction for autocorrelation and heteroskedasticity? Or must we run each equation via OLS , adjusting for HAC-Newey West?

[EViews Gareth]

Nothing has changed in EViews 8.

[EViews Glenn]

And you can still specify as a system, though the HAC options are less numerous than those for the individual equations.

[resbaby]

So what is the possible solution in case of VAR having autocorrelation and heteroscedasticity problem ???

[EViews Glenn]

To expand on our earlier comments, you can still create a system object from of the VAR, estimate using GMM with the appropriate set of instruments, and then tell EViews to do a full system HAC. No individual equations required.