How to process fixed effect model with lagged independent variable

[David Chin]

Hi guys,
i am trying to study relationship between wage and firm performance with lagged wage variable
i have 832 company in 5 year data, i used xtreg and hausman test to decide fixed or random model and the result is fixed .

however , when i run xtreg ,fe it will auto demean my lagged wage variable and it seems strange in economic explanation

so i use reg with manually demean variable ,which means i use demean wage and lagged of this variable

but next question is if i want to test heteroskedasticiy and cross sectional dependence by using xttest3 and xtcsd, pesaran abs

stata says "last estimates not found" it seems these commend can only run after xtreg

how can i test for these ?

and on the other hand, use reg to run within fixed effect model , it gives wrong standard error of varialbe and R square because of wrong degree of freedom (reg and xtreg doesn't creat dummy )
usually i use areg to correct this problem but i turn to same question mention above


is there any suggested procedure to study fixed effect model with lagged independent variable?
can i use hausman test as before aftre take this question account?
thx

[Phil Bromiley]

I think you're making problems where they don't exist. If the lagged variable is endogenous (e.g., a lagged dependent variable) then you do have problems. Otherwise, it is being estimated properly. xtreg, fe with demeaning gives you the same parameters and standard errors you would get if you put in dummy variables for all but one of the panels. That it happens to be more efficient to either difference over time or de-mean the variables to calculate these parameters is not relevant. Treat the lagged exogenous iv's like any other variable.

Phil

[David Chin]

that's what i think before
but in fact, the coefficient and standard error are different when i run these two commend as you can see from the pictures

gw_ variable are demean varible

look at picture
row1-6 are data of same firm and in xtreg it will demean column lagavewage1 using mean of row 2-6
and in column lagavewage2 it will demean with mean of row 3-6

[Phil Bromiley]

Try checking the FAQs. Statalist does not like photos - use the program that gives a data example. I can't (and won't) check that you have actually demeaned exactly the way xtreg does. If you put in the fixed effects as dummies, you should get identical results. You're just making problems for yourself.