Simultaneous Equation Model with Fixed Effects

[Solomon Lin]

Hi,

Can someone please clarify the appropriate method for running a 3 equation simultaneous equation model with fixed effects estimation? In my model, each of the dependent variables depend on the other two i.e. Y1 depends on Y2 and Y3; Y2 depends on Y1 and Y3; and Y3 depends on Y1 and Y2. I will be demeaning over time and using two time periods. I will need robust standard errors.

I believe this is the basic command:

xtivreg2depvar [varlist1] (varlist2=varlist_iv) [weight] [ifexp], fe

1) How do I adjust it so that I can have both Y2 and Y3 in the above equation for Y1?

I know that if you use the reg3 command, you can specify the instruments for each independent variable separately (but you cannot use reg3 for fixed effects), e.g.:

reg3 (Y1 X1 Y2 Y3) (Y2 IV1 IV2) (Y3 IV3 IV4), 2sls

Is there a way I can specific instruments separately with xtivreg2? Or does it not matter whether I do that or not? Also, what if some of the dependent variables share the same instruments, e.g. if IV1 is used for Y2 and Y3?

2) How can I include robust standard errors? Is it just be typing in "r" after "fe"?

Thanks so much!

[Emad Shehata]

Simultaneous equations models as 3SLS for panel data not available yet
So you can use now a single equation xtivreg for panel 2SLS or Fixed and Random Effects in panel data using SEM

For more detail
Fixed and Random Effects in Panel Data Using Structural Equations Models

http://papers.ccpr.ucla.edu/papers/P...R-2008-003.pdf

[Solomon Lin]

Simultaneous equations models as 3SLS for panel data not available yet
So you can use now a single equation xtivreg for panel 2SLS or Fixed and Random Effects in panel data using SEM

For more detail
Fixed and Random Effects in Panel Data Using Structural Equations Models

http://papers.ccpr.ucla.edu/papers/P...R-2008-003.pdf

Hi Emad,

Thank you so much for the reply. I actually don't need 3SLS (I was just using reg3 as an example).

Could you just explain a bit more about how I would conduct the test as I don't fully understand your comment?

For the first option you provided, I believe you mean that I can just use the xtivreg command (I prefer to use xtivreg2 though because I read that xtivreg2 allows for robust standard errors). But how do I write it if I have 2 endogenous variables with separate lists of instruments? Can I write this?

xtivreg2 depvar [varlist1] (endogenous variable 1=varlist_iv1)(endogenous variable 2=varlist_iv2), fe

Also, can the instruments overlap? As in, can some of the instruments in varlist_iv1 be the same as varlist_iv2?




EDIT: I will read the paper for the second method but I am interested in using the xtivreg2 command. Thank you for suggesting the paper.

[Emad Shehata]

Like this Form only

xtivreg2 y1 x1 x2 (y2 y3 =x3 x4 x5), fe

This command not correct
xtivreg2 y1 x1 x2 (y2 =x3 x4) (y3 =x3 x4 x5), fe

[Solomon Lin]

Like this Form only

xtivreg2 y1 x1 x2 (y2 y3 =x3 x4 x5), fe

This command not correct
xtivreg2 y1 x1 x2 (y2 =x3 x4) (y3 =x3 x4 x5), fe

Ah, is there an alternative way to conduct this test? I am running a simultaneous equation model and not all of the instruments are meant for all of the endogenous variables.