Panel Regression Estimators - Section between xtgls and xtpcse

[Suren Perera]

Hi everyone,

I am currently conducting a research study investigating the impact of leverage (debt-to-asset ratio) on the financial performance (return on equity) of listed corporations. To achieve this objective, I have collected panel data from the annual reports of 20 companies over the past five years.

After conducting a panel regression analysis, I decided to use the fixed-effects model based on the results of the Hausman test. However, further diagnostic tests revealed several issues with my data:

• Wooldridge test: Detected autocorrelation.
• Wald test: Identified heteroskedasticity.
• Pesaran test: Indicated cross-sectional dependence.

Given these challenges and the dimensions of my panel (N = 20, T = 5), I am unsure whether to proceed with the xtgls (Generalized Least Squares) or xtpcse (Panel-Corrected Standard Errors) estimator to address these issues.

1. Which estimator (xtgls or xtpcse) is more suitable for my analysis to ensure accurate estimation of the relationship between leverage and financial performance?
2. Are there alternative approaches that might be more appropriate for my dataset?

Your advice and expertise would mean a lot as I navigate this step in my analysis! Thank you in advance for your help.

[Carlo Lazzaro]

Suren:
as you're dealing with a short (N>T) panel dataset, -xtgls- seems out of debate in favour of -xtreg,fe- with cluster robust standard errors.
I would consider -xtpsce- as a second choice only, as it is conceribed for T>N panel datasets.

[Suren Perera]

Dear Carlo,
Thank you so much for your invaluable guidance and prompt attention to my concerns. Now I have a clear path to continue my analysis. Your support has been truly valuable.

[Suren Perera]

Dear Carlo,

I recently read an article by Daniel Hoechle (2007) on "Robust Standard Errors for Panel Regressions with Cross-Sectional Dependence" (DOI: 10.1177/1536867X0700700301). On page 4, the article states that xtgls requires N < T for feasibility.

However, in your previous suggestion, xtgls was proposed as a suitable estimator when N > T. I am a bit confused about this—could the article be referring to a different aspect of feasibility, or might I have misunderstood its explanation?

I would greatly appreciate your clarification on this matter.

[Carlo Lazzaro]

Suren:
1) Daniel's article (page 285) about -xtgls- states:

N < T required for feasibility; tends to produce optimistic SE estimates

.
2) In addition, I previously replied:

...as you're dealing with a short (N>T) panel dataset, -xtgls- seems out of debate in favour of -xtreg,fe- with cluster robust standard errors.

.
HTH

[Suren Perera]

Dear Carlo,

I sincerely apologize for my misunderstanding of your comment. Upon rereading your explanation, I now fully grasp your point. I truly appreciate your patience and the time you took to clarify things for me.

Thank you once again for your thoughtful response. Your insights have been extremely helpful.

Best regards,
Suren