Panel Data Fixed Effects

[Ali Hasan]

Hi Everyone,

I needed help in dealing with endogeneity in panel data with fixed effects analysis. I have tried to setup my instrumental regression like this

xtivreg log_loan_amount refinancing_loan home_purchase_loan applicant_sex minority (Fintech log_applicant_income = minority applicant_sex home_purchase_loan loan_type tract_to_msamd_income), fe first
(Fintech and applicant income are endogenous variables)

But the R2 in this way is too low.

The other way which gives me better R2 is through solving this way:

xtreg (log_applicant_income Fintech) minority loan_type loan_type refinancing_loan home_purchase_loan applicant_sex,fe robust
predict v1hat, resid
xtreg log_loan_amount log_applicant_income Fintech refinancing_loan home_purchase_loan applicant_sex v1hat,fe robust
test v1hat

Can you please tell me if the second alternative method is right to solve for endogeneity?

Kind regards.

[Maxence Morlet]

The R-Squared in IV regression is quite a weird unusual concept.

I presume you're in economics. In labour economics at least, we care a lot more about causality than the percentage of variability in the DV that our model explains. I would stick with IV if you believe in the exclusion restriction for your instruments.

[Maxence Morlet]

Because your model is overidentified you can produce something like the Sargan J Statistic

[Ali Hasan]

Thank you very much, Maxence Morlet. Really appreciate your help.

Yeah, this is coming from an economic perspective. Can you please explain a little more about why would you like to stick with IV. For giving you a better idea, I am sharing with you the final regression of xtvireg.

[Ali Hasan]

I am also sharing with you the result of the second methodology with v1hat. Please correct me if I am missing something. I am really looking forward to your reply.

Thank you once again,

[Maxence Morlet]

Can I see the first stage as well? Type option “first” in xtivreg

[Maxence Morlet]

The xtivreg by the way also automatically adjusts standard errors for the loss of degrees of freedom due to the fitted values being estimated

[Ali Hasan]

Dear Maxence Morlet,

Yeah sure. Please see the first stage. Thank you very much for your help.