Hallo,
I'm currently working on my panel data models.
I want to find evidence if there's a problem of heterskedasticity or serial correlation in my data.
Of course, I know the xtserial and xtgls commands see http://www.stata.com/support/faqs/st...tocorrelation/
However I see myself confronted with another starting condition due to my panale data structure.I analyze executive compensation of index listed german companies between 2006 and 2012 (year).
Because I include not only one executive but several executives per firm per year into my examination, the panel variable is not "firm" (ID) but an artifically created mixture out of executive's name and the company (personIDfirmID). This is due to the fact, that STATA does not allow for multiple observations per year.
So, I defined my panel data set like this:
. xtset personIDfirmID year
panel variable: personIDfirmID (unbalanced)
time variable: year, 2006 to 2012, but with gaps
delta: 1 unit
However in my regression I control (fixed or random effects) on the firm level. That's why my regression looks like this:
xtreg indepvar devars i.year, i.(ID) re robust
xtreg indepvar devars i.year, i.(ID) fe robust
(depending on Hausman test, the "robust" was told me to be "state of the art", don't ask me why I do this)
Now back to the topic of heteroskedasticity and serial correlation:
The link above explains procedures, but the commans refer to the xtset command and thus treat "personIDfirmID" as the panel variable. However I control on firm level (ID).
Is this even a problem? Are there other tests for Heteroskedasticity and Serial Correlation in Panel Data, where I can adjust the panel variable?
Thanks,
Simon
I'm currently working on my panel data models.
I want to find evidence if there's a problem of heterskedasticity or serial correlation in my data.
Of course, I know the xtserial and xtgls commands see http://www.stata.com/support/faqs/st...tocorrelation/
However I see myself confronted with another starting condition due to my panale data structure.I analyze executive compensation of index listed german companies between 2006 and 2012 (year).
Because I include not only one executive but several executives per firm per year into my examination, the panel variable is not "firm" (ID) but an artifically created mixture out of executive's name and the company (personIDfirmID). This is due to the fact, that STATA does not allow for multiple observations per year.
So, I defined my panel data set like this:
. xtset personIDfirmID year
panel variable: personIDfirmID (unbalanced)
time variable: year, 2006 to 2012, but with gaps
delta: 1 unit
However in my regression I control (fixed or random effects) on the firm level. That's why my regression looks like this:
xtreg indepvar devars i.year, i.(ID) re robust
xtreg indepvar devars i.year, i.(ID) fe robust
(depending on Hausman test, the "robust" was told me to be "state of the art", don't ask me why I do this)
Now back to the topic of heteroskedasticity and serial correlation:
The link above explains procedures, but the commans refer to the xtset command and thus treat "personIDfirmID" as the panel variable. However I control on firm level (ID).
Is this even a problem? Are there other tests for Heteroskedasticity and Serial Correlation in Panel Data, where I can adjust the panel variable?
Thanks,
Simon
