Page 1 of 1

OLS panel estimation with clustered standard errors

Posted: Thu Sep 12, 2013 10:38 am
by truetears
Hi everyone, please find my Eviews workfile attached ("6.wf1"). Basically eq01 is the OLS panel regression output (without clustered standard errors), how can I use clustered standard errors? I have created a variable "key" which is the clustering identifier, but I am unsure of what to click to use clustered standard errors. Thanks.

Re: OLS panel estimation with clustered standard errors

Posted: Thu Sep 12, 2013 4:42 pm
by EViews Glenn
Your data are not structured as a panel. What is the panel identifier?

Re: OLS panel estimation with clustered standard errors

Posted: Fri Sep 13, 2013 12:27 am
by truetears
Ah yes, I managed to import the excel datafile into Eviews 7 as panel after searching through some old forum posts.

I have uploaded my excel datafile here: https://www.dropbox.com/s/gy12j5y913tox ... 0file.xlsx (too big to upload as an attachment)

The yellow highlighted columns are my dependent variables, the blue columns are my independent variables and the green column is the identifier for each bank. Eg, "1" corresponds to the some bank, "2" corresponds to a different bank, and finally "23" corresponds to the last bank. I have a dated balanced panel from 1/1/2008-28/2/2013 (1348 observations) across 23 banks.

I have uploaded my eviews workfile here: https://www.dropbox.com/s/3ftgw94mon76k ... rkfile.wf1

eq01 is the OLS regression I want to run, but I want to do a few extra things which I am unsure of:

1. I want to use clustered standard errors (clustered by bank), how do I do this?

2. I also want to use bank fixed effects. To do this, do I just select "fixed" under effects specification -> cross-section?

Many thanks for your help.

Re: OLS panel estimation with clustered standard errors

Posted: Fri Sep 13, 2013 9:54 am
by EViews Glenn
I can't open your workfile, but if you are structured with bank as the cross-section identifier, then cross-section effects will give you fixed effects, and "White period" coefficient covariances will give the bank clustered covariances.

Re: OLS panel estimation with clustered standard errors

Posted: Fri Sep 13, 2013 11:19 am
by truetears
Thanks Glenn. Also I just want to clarify, I see there are two types of White coeficient covariances, "White - Period" and "White - Cross Section", so for clustering by bank I use White-Period, but what is "White - Cross Section"? Thanks in advance.

Re: OLS panel estimation with clustered standard errors

Posted: Fri Sep 13, 2013 4:58 pm
by EViews Glenn
Contemporaneous correlation across cross-sections (clustering by period, if you will).

Re: OLS panel estimation with clustered standard errors

Posted: Sat Sep 14, 2013 8:42 am
by truetears
Ah I see, also just one last thing, what is the difference between fixed effects for "cross-section" and "period" under Panel options -> effects specification? So cross-section fixed effects (assuming I have bank as the cross-section identifier) gives me bank fixed effects, what if I had "period" fixed effects, what does that do?

Re: OLS panel estimation with clustered standard errors

Posted: Mon Sep 16, 2013 9:30 am
by EViews Glenn
If it's a dated panel, time effects.