EViews.com

I have a question regarding exactly how Eviews applies the Newey-West HACV estimator in cases of panel data in a standard linear regression case. Does the estimator appropriately account for truncation of lags across different cross sectional units? How are lag orders automatically chosen in such cases? What if the panel is unbalanced? Finally, in my particular case, there are multiple observations on a single cross sectional unit within a given time period (a year). Does Eviews treat this properly? I am thinking that clustering std error corrections would be more appropriate. Do you agree?

How are you trying to estimate the HACV for panel data in EViews? It really isn't part of the standard suite of options...

Hi--

Actually, I am reviewing some work done by someone else and I do not know exactly how it was done (though I know it was done in Eviews). By your response, I take it that Eviews does not have an automatic option to allow one to apply Newey-West in a panel (in a way that appropriately truncates the lags so as to prevent adjustments for correlation across different cross-sectional units)--correct? I am sure that one could accomplish this by hard-coding, but that is not what was done.

I very much appreciate your response.

BG

Correct, EViews does not have an automatic way to implement Newey-West standard errors in a Panel at all (let alone appropriately truncating lags etc...). Whatever was done must have been done through a custom EViews program (or was simply done on non-Panel data).

Hi,

I am actually dealing with the same problem. bkgwatfiv, you do not happen to work on the extrapolation theory do you ? =)

Coming to the details of my problem. as i mentioned above i am dealing with the extrapolation theory. therefore, I use a set of independent variables in time t and the return of period t+1 to t+4 as dependent variables. since I am dealing with this in an unbalanced panel, I do have serious serial correlation in my regression (DW stats much smaller than 2), as the returns overlap when moving along in time. I found several coef. coveriance estimation methods. the "period sur (pcse) coef. covariance estimation method", - according to the help file - allows for general correlation of residuals across periods for a specific cross-section.

Should this not be the correct estimation method for my model? Do you have any other suggestion? In my paper, the author is also mentioning the newey-west procedure which I was not able to find in eviews - as i know now because it does not exist.

Thanks for your help

All the best

Kai

Hi-- I wish I could be of more assistance, but I only know that a straight Newey-West needs modification (in a way that Eviews apparently does not do). You may find a great paper by Mitch Peterson of Northwestern of interest.

“Estimating Standard Errors in Finance Panel Data Sets: Comparing Approaches,” Review of Financial Studies, January 2009, Volume 22, pages 435 - 480.

Hope that helps.

Best,

BG

Thanks for that paper !

Some more general questions. I cannot find the white test for heteroskedasticity in the panel equation window. all it tells me for the residuals is table, graph, correlations and covariances. how can i test for heteroskedasticity in an unbalanced panel ?

Similarily, how can I test for serial correlation in an unbalanced panel? Is there only the Durbin Watson Test ? How can I correct for the serial correlation in my regression model ? By coef. coveriance estimation methods? Moreover, coming back to my question above, how can I determine which coef. coveriance estimation methods is the most appropriate for my model ?

All the best and thanks

Kai

Hi
I am working on unbalanced data in e-views 10, I am trying to apply Peterson (2009) or as you refer in your website and guide II as Petersen (2009) cluster by firm and year, I found a useful information in the following help link: http://www.eviews.com/help/helpintro.ht ... rrors.html
but my questions are:
1- is this is available only in pooled data not panel data.
2- Could I use Period SUR (PCSE) method in stead as robust as I want to cluster by cross section only
3- If not, How could I convert my attached panel work file to pooled work file.
Thanks in advance