ivreg2 with interaction terms

[David Crawford]

Hello,

I am trying to understand how to make a 2SLS with ivreg2 when there are interaction terms. In particular, I am now estimating the following:

*****************
* 2SLS regression
*****************

***Fist Stage
reg pres lond eng eur con age realgdpgr inflation lagdebt unemp emu interest sov, robust
predict pres_hat
reg fed lond eng eur con age realgdpgr inflation lagdebt unemp emu interest sov, robust
predict fed_hat
reg corp lond eng eurfrac con age realgdpgr inflation lagdebt unemp emu interest sov, robust
predict corp_hat
reg prop lond eng eurfrac con age realgdpgr inflation lagdebt unemp emu interest sov, robust
predict prop_hat

***Second Stage
reg deficit realgdpgr inflation unemp emu interest sov lagdebt pres_hat corp_hat prop_hat fed_hat, robust
reg deficit realgdpgr inflation unemp emu interest sov lagdebt c.pres_hat##c.corp_hat c.prop_hat##c.corp_hat c.fed_hat##c.corp_hat, robust

Where the variable of interet is deficit, the controls are realgdpgr inflation unemp emu interest sov lagdebt, while the intruments are lond eng eur con age and the intrumented pres, prop, corp and fed.

Now, if I use the above method everything seems to work fine. However, I would now like to use ivreg2:

ivreg2 deficit realgdpgr inflation unemp emu interest sov lagdebt (c.pres##c.corp c.prop##c.corp c.fed##c.corp= lond eng eur con age), robust

However, Stata says that there are no enough instruments. Could you help me understand where the mistake is and how to fix it?

Thank you!

[Andrew Musau]

c.pres##c.corp c.prop##c.corp c.fed##c.corp

This implies 9 endogenous variables after the factor variable expansion, i.e., c.pres##c.corp= c.pres, c.corp, c.pres#c.corp, and so on. You have 5 instruments. For identification, you need at least as many instruments as there are endogenous regressors.

[David Crawford]

Thank you Andrew for your reply. Just a couple of clarifications. If I intereract two of the five instruments, would the resulting interaction be used as the 6th instrument?

Thank you

[Andrew Musau]

Yes, as it is a variable in its own right.

[David Crawford]

Thank you very much!