Pool & Panel

[mfb]

When doing a regression using a Pool object, we can check residuals with View/Residuals (to have a look at Table, Graphs, Covariance Matrix, Correlation Matrix of residuals grouped by periods and cross-sections).

How can one check residuals (Table, Graphs, Covariance Matrix, Correlation Matrix) when doing regression with panel data using the menu Quick/Estimate equation ? Is it possible at all?

Thanks,

[mfb]

This is not a question about econometrics as such. This is a question about Eviews. Could the Eviews team please provide an answer?

Is there any simple procedure to examine residuals, when estimating with Quick/Estimate equation/Panel data, as we have when estimating with a Pool object (by using the View/Residuals menu option)?

Thanks

[EViews Gareth]

Proc->Make Residuals ?


(note, we tend to not even read posts in the Econometric Discussions forum :) )

[mfb]

I believe that is just for making a copy of the residuals.

What I am looking for is something that will allow testing residuals from a panel data regression for heteroskedasticity and serial correlation.

I need this in order to justify using any of the GLS Weights or the Coef Covariance Methods (heterosk-robust standard errors) options available in Eviews panel equation configuration box.

If these options are available, we should have the means of testing the data in order to justify their use (like doing heterosk tests in single equation models to see if GLS or robust errors or some other procedure is needed).

Are there any such tests for panel data in Eviews?

Thanks.

[EViews Gareth]

The views you mentioned for the pool can all be done from a series in a panel.

There are no built in heteroskedasticity tests for either pools or panels.

[mfb]

Ok. Then what I have tried to do is looking at the Eviews equation output menu options to see what could be used for this purpose.

Using an example panel data set of 18 cross-sections and 6 periods, I did this:

1. I have run the Likelihood Ratio and Hausman tests to select Fixed Effects (as against Random Effects).

2. Estimate the equation with Fixed Effects with no GLS Weights and no Robust Errors (Coef Covariance Method=none).

3. Open the RESID series on a spreadsheet and run the tests below.

4. Cross-section Heteroskedasticity
Use the Bartlett and Levene tests for equality of variances:
View / Descriptive statistics & tests / Equality tests by classification
(using the identity series cn to classify cross-sections)

5. Period Heteroskedasticity
Use the Bartlett and Levene tests for equality of variances:
View / Descriptive statistics & tests / Equality tests by classification
(using the identity series yr to classify periods)

5. Contemporaneous covariance
View / Panel covariance / Contemporaneous covariance

6. Between periods covariance
View / Panel covariance / Between periods covariance

The conclusions I arrived at, in this particular case, were this:
1. Cross-section heteroskedasticity
2. Period heteroskedasticity
3. Cross-section contemporaneous correlation
4. Period correlation (serial correlation)

I will then have to select GLS Weights and/or Robust Standard Errors (Coef Covariance Method) to account for all 4 these features in my panel data.

Please see attached Excel spreadsheet which records results of these tests.

Is there any other thing I could possibly do in Eviews to improve on this?

Thanks,

Panel data_FE_RESID_Eviews tests.xls

(29 KiB) Downloaded 980 times

[mfb]

Could the Eviews team provide some feedback on this, please?

I've read the User's Guide but found no further guidance on these issues.

So I need confirmation this is all there is in terms of Eviews testing facilites for panel data GLS/Robust Standard Errors.

Thanks,

[EViews Gareth]

We don't offer econometric advice.

[mfb]

My query is not really about econometrics. It is about usage of EVIEWS.

I checked the User's Guide. But I might be missing something.

So, what I am asking is:

What facilities exist in EVIEWS to analyse the RESID series for the purpose of GLS/Coef Covariance Methods selection?

Just those I illustrated in my earlier post or are there any others?

[EViews Glenn]

In an Equation object look at the entries under View/Residual Diagnostics. In addition, there are add-ins for testing residuals for cross-sectional dependence and random effects.

[Vap01]

Ok. Then what I have tried to do is looking at the Eviews equation output menu options to see what could be used for this purpose.

Using an example panel data set of 18 cross-sections and 6 periods, I did this:

1. I have run the Likelihood Ratio and Hausman tests to select Fixed Effects (as against Random Effects).

2. Estimate the equation with Fixed Effects with no GLS Weights and no Robust Errors (Coef Covariance Method=none).

3. Open the RESID series on a spreadsheet and run the tests below.

4. Cross-section Heteroskedasticity
Use the Bartlett and Levene tests for equality of variances:
View / Descriptive statistics & tests / Equality tests by classification
(using the identity series cn to classify cross-sections)

5. Period Heteroskedasticity
Use the Bartlett and Levene tests for equality of variances:
View / Descriptive statistics & tests / Equality tests by classification
(using the identity series yr to classify periods)

5. Contemporaneous covariance
View / Panel covariance / Contemporaneous covariance

6. Between periods covariance
View / Panel covariance / Between periods covariance

The conclusions I arrived at, in this particular case, were this:
1. Cross-section heteroskedasticity
2. Period heteroskedasticity
3. Cross-section contemporaneous correlation
4. Period correlation (serial correlation)

I will then have to select GLS Weights and/or Robust Standard Errors (Coef Covariance Method) to account for all 4 these features in my panel data.

Please see attached Excel spreadsheet which records results of these tests.

Is there any other thing I could possibly do in Eviews to improve on this?

Thanks,

Panel data_FE_RESID_Eviews tests.xls

I am having the same problem. I'm trying to test heteroskedascity for a panel data and this abscence of a built in test is kiling me. Could you please give me more details on these procedure you have adopted? Thank you!