Clustered errors

[MalaMi]

Thanks!

[Sielenna]

Hi,

I have a cross-sectional dataset of 130 observations. These observations are selected over 20 different countries, for different periods in time. I do not have an equal number of observations for each country, nor do I have equal time periods for each observations.
Eg: 3 observations for country 1, [1980-1983], [ 2000, 2004] and [2006-2009] , 2 observations for country 2: [2005], [2007-2009], and so on.

I have included "fixed effect" dummies for each country, but I was wondering if I should also consider " country-clustered standard errors". I want to allow my errors to be correlated with the other observations of the same country, but not with observations from a different country. White-period standard errors look appropriate but unfortunately you can only apply them in a panel dataset. How can I adjust these standard errors for my case?

Kind greetings,

[zeca]

Hi,
My sample have the 100 largest publicly traded banks in Europe from 1996 to 2011 (a total of 1133 observations). Banks are from 25 diferent countries.
I'm using eviews to estimate a regression in which Leverage is the dependent variable and then I have 5 explanatory variables (please see attached the equation).

Following the available literature I have to use time and country fixed effects (Ct and Cc). Regardind time fixed effects I found it easy (panel options / Effects specification / Period:Fixed)
Q1: Could you give me some guidelines to deal with country fixed effects in eviews? (One of my available variables is country)

Two aditional questions:
Q2: what should I do to adjust the standard errors for clustering at the bank level?
Q3: variable Div is a dummy (1 if bank pays dividends in a given period, 0 otherwise). Once is a dummy it has the same processing and interpretation of remaining variables?

Many thanks!

[EViews Glenn]

To compute clustering in the cross-section workfile, simple structure as a non-dated panel using the variable you want to cluster over as the cross-section identifier.

[EViews Glenn]

I must admit I'm finding it difficult to follow the multiple conversations in this single thread. In general, one should create a new topic/thread, rather than hijacking an existing one, unless the comments pertain to the early discussion.

On banking. How do you have the workfile structured?

[zeca]

I must admit I'm finding it difficult to follow the multiple conversations in this single thread. In general, one should create a new topic/thread, rather than hijacking an existing one, unless the comments pertain to the early discussion.

On banking. How do you have the workfile structured?

Thanks for the advice. I'll move it to a new topic

[mfb]

Could I interpret all you said in the following way, assuming panel data structured with N cross-sections (firms, countries, etc) over T periods (years, etc):

White Cross-section:
N equations giving a NxN error matrix. White Cross-section robust errors assumes and corrects for diagonal heteroskedastic variances and non-zero off-diagonal covariances.
It’s like when some countries are somewhat constant over time while others may vary much over time. Also, some countries may move together up or down while others may be moving independently from the rest.

White Period:
T equations giving a TxT error matrix. White Period robust errors assumes and corrects for diagonal heteroskedastic variances and non-zero off-diagonal covariances.
It’s like there are calm years when each country behaves like all other countries and wild years when you have countries moving haphazardly apart from each other. Also, there may be some persistence over time: wild years tend to follow wild years and calm years tend to follow calm years.

White Diagonal:
NTxNT error matrix. White Diagonal assumes and corrects for individual observations heteroskedasticity but makes no assumptions whatsoever concerning covariances of any type.


Any of these options can be used before or after controlling for Fixed or Random Effects, cross-wise or period-wise (or cross-wise and period-wise).

Is this ok?

[mfb]

Could the Eviews team please provide some feedback on whether my interpretation as stated in my post of Wed July 10 is ok in terms of Eviews implementation?
Thanks,

[EViews Gareth]

The EViews team does not offer econometric advice (no matter how many times, or how differently you ask).

[mfb]

I conclude that my interpretation as in my post of Wed July 10 corresponds exactly to what Eviews does.
Thanks.

[paris223]

I have the same issue. I am trying to use clustered standard errors for an unstructured file. Could anyone help with some guiding instructions for how to do this in Eviews 7? i am pretty nw with Eviews but I need to use it urgently for my thesis.

Thanks a lot in advance!

[adc]

I get the difference now between white cross-section and white-period, but could you explain what is the difference between white-period and period SUR?

many thanks.

[EViews Glenn]

The former are adjustments to the computation of the standard errors that make them robust to cross-section intercluster covariances where we do not assume common structure on the covariances. The latter robust method assumes that the intercluster covariances are common across cross-sections, estimates those covariances, and then adjusts the standard error calculations accordingly. I believe that this is discussed in the manual.