F - Test with Dummy Variables

[Jarno Stappers]

Dear reader,

I have some problems with using the appropriate command for conducting an F-test for dummy variables. It is important to mention that I have Panel Data. The model consists of multiple fixed effects (year, firm, and executive). The executive fixed effects consist of three dummy variables namely: CEO, CFO, and Other. I have seven different models in which I want to test different fixed effects. In model 4, 5, 6, and 7 I would like to test for the executive effects so:

Model 4: Baseline regression (including only control variables) + CEO effects
Model 5: Baseline regression (including only control variables) + CFO effects
Model 6: Baseline regression (including only control variables) + Other effects
Model 7: Baseline regression (including only control variables) + CEO, CFO, and Other effects.

In STATA I tested different commands but I doubt that I used them correctly. I am aware of the different commands -regress- , -xtreg,fe-, and -reghdfe-.
Since, I only add one effect for models 4, 5, and 6 I guess that I have to work with the -xtreg,fe- command. However, I am also not sure which group variable (xtset) I have to use.

In model 7, I want to use the -reghdfe- command but I am not sure if I have to cluster or not.

The command that I used in model 4:

xtset CEO

xtreg GAAPETR RD ADVERTISING SGA CAPITALex LEVERAGE CASHHOlDINGS FOREIGN_OPERATIONS Net_Operating_Loss SIZE ESTIMATED_OPTION_EXPENSE INTANGIBLETOTOTALASSETS GROSSPPETOTOTALASSETS,fe

My questions are regarding which commands to use in these specific cases.
It would be nice if someone can explain this to me.

[Carlo Lazzaro]

Jarno:
if you're really interested in retrieving 3 fixed effects, the community-contributed programme -reghdfe- is probably your best bet.
That said, read carefully the related help file, that, among other topics, covers clustered standard errors.

[Jesse Wursten]

I would recommend using the reghdfe command, even if only for clarity. You should definitely cluster you standard errors, though it is not obvious at which level, probably firm or higher. You should be able to find guidance for this in related literature.

[Jarno Stappers]

Thanks Carlo and Jesse for your replies!

I will take a look at the literature for model 7.
However, do I also have to use the -reghdfe- command if I only include one fixed-effect (e.g. model 4)?

[Carlo Lazzaro]

Jarno:
no, for model 4 and 5 the excellent community-contributed command -reghdfe- is probably an overkill.
Going -xtreg,fe- should be enough.
I would add -i.time- among the predictors.
Be careful with clustered standard errors, consider the community-contributed programm -xtoverid- in lieu of -hausman- (if you use non-default standard errors) and, last but not least, run a comprehensive post estimataion routine (especially focused on possible model misspecification).

[Jesse Wursten]

Jarno:
no, for model 4 and 5 the excellent community-contributed command -reghdfe- is probably an overkill.
Going -xtreg,fe- should be enough.
I would add -i.time- among the predictors.
Be careful with clustered standard errors, consider the community-contributed programm -xtoverid- in lieu of -hausman- (if you use non-default standard errors) and, last but not least, run a comprehensive post estimataion routine (especially focused on possible model misspecification).

Personally I would use reghdfe for code consistency, but I suppose that's a matter of preference

[Carlo Lazzaro]

Jesse:
I get your point.
I think that the main issue with using community-contributed programme is a sound background knowledge of the underlying theory earned with practicing with Stata built-in commands (such -xtreg-, in this case).
As the original poster (OP) seems to be taking his first steps the demanding realm of panel data regression, it is probably advisable to keep things (commands) as simple as possible whenever feasible (that's why I recommended -xtreg,fe. for models 4 and 5 described by the OP).
My guess is that -reghdfe- requires a deep knowledge of (and practice with) -xtreg -fe- estimator (since it is a sort of evolution of -xtreg-).
That said, I find interesting to compare the results obtained with -xtreg,fe- and -reghdfe- when you are interested in one fixed effect only.