Query in Breusch-Pagan test for panel data

[Sudharshan Srikkanth]

Hello everyone, I am new to this forum.
I would like to test my panel data model for Heterskedasticity and so would like to run a Breusch-Pagan test using the xttest2 command. However, whenever I try to run the command, I get the Error: "too few common observations across panel. no observations". I would like to know how I can fix this issue so I can run the xttest2 or I would like to know how I can maually run a test for heteroskedasticity in panel data as this is for my thesis and I have to show the tests. Any input would be greatly appreciated.

[Carlo Lazzaro]

Sudharshan:
the first issue to investigate is the evidence of a panel-wise effect in your dataset.
If you're using default standard errors, what above is proved by the statistical significance reached by the F-statistic just below your -xtreg,fe- outcome table.

[Sudharshan Srikkanth]

Sudharshan:
the first issue to investigate is the evidence of a panel-wise effect in your dataset.
If you're using default standard errors, what above is proved by the statistical significance reached by the F-statistic just below your -xtreg,fe- outcome table.

Hello Carlo,
Thank you for your response. In the outcome table, the F statistic is significant at the 1% level with a statistic of 125.77. But I don't understand how this helps me solve my issue. Any repsonse is greatly appreciated.
Thanks

[Carlo Lazzaro]

Sudharshan:
how many observations per panel are you referring to?

[Sudharshan Srikkanth]

Hello Carlo,
The average observation per group is 25.6, I am checking for a relationship between industry concentration and firm profitability. Thus I have industry data from 1970 to 2021 and I have data relating to 45 industries. I just making a guess here: Is this error occurring because the number of years is greater than the number of groups or something of that nature? Also, if that is the case, am I better off using xtgls instead xtreg fe? Is there a varaiton of the Hausman Test to verify using the xtgls as opposed to other specifications? Aplogies for the barrage of questions and thank you for your response.

Edit: I would als like to add a request: Please suggest a method that I can test for heteroskedasticity as xttest2 is not working.

[Carlo Lazzaro]

Sudharshan:
the error probably occurs because the number of common observations across panels is too low.
That said, with such a long T dimension, I would cluster the standard errors on panelid, taking at the same time autocorrelation and heteroskedasticity into account.
If analytic tests do not work, visual inspection can be an easy work-around-

[Sudharshan Srikkanth]

Sudharshan:
the error probably occurs because the number of common observations across panels is too low.
That said, with such a long T dimension, I would cluster the standard errors on panelid, taking at the same time autocorrelation and heteroskedasticity into account.
If analytic tests do not work, visual inspection can be an easy work-around-

Thank you for your response. Will running a fixed effects regression with robust standard errors cluster the standard errors on panelid, like you mentioned, while taking autocorrelation and heteroskedasticity into account?

[Carlo Lazzaro]

Sudharshan:
correct.

[Sudharshan Srikkanth]

Hello Carlo, thank you for your repsonse.
What method would you recommend for visual representation of heteroskedasticity and what command is useful in implementing it?

[Carlo Lazzaro]

Sudarsham:
the following toy-example could be helpful:

. use "https://www.stata-press.com/data/r17/nlswork.dta"
(National Longitudinal Survey of Young Women, 14-24 years old in 1968)

. quietly xtreg ln_wage c.age##c.age, fe


. predict fitted, xb


. predict epsilon, e


. scatter e fitted

.

Heteroskedasticity is apparent in the graph: just go -vce(cluster idcode)- and do not plot epsilon vs. fitted again, as you will obtain the very same graph.