ivreghdfe and conformability error

[Johanna Krenz]

Hi,

when I want to estimate an equation with ivreghdfe and time FE plus when I try to calculate country-specific ("group" is a country identifier which is my cross-section) coefficients for the shock I get the error: conformability error r(503), i.e., if I run

ivreghdfe delta_GDP1 (c.`shock'#i.group=c.`instr'#i.group) L(`q'/`p').c.int_GDP1, absorb(mdate) vce(cluster mdate)


However, I don't get the error when I run

ivreghdfe delta_GDP1 (c.`shock'p=c.`instr') L(`q'/`p').c.int_GDP1, absorb(mdate) vce(cluster mdate)

or when I take out the time FE

ivreghdfe delta_GDP1 (c.`shock'#i.group=c.`instr'#i.group) L(`q'/`p').c.int_GDP1, absorb(mdate) vce(cluster mdate)


But I want the country-specific coefficients.


When I just use the shock directly, i.e. no instrumental variable regression, and use reghdfe it runs also with the group interaction, i.e.,

reghdfe delta_GDP1 c.`shock'#i.group L(`q'/`p').c.int_GDP1, absorb(mdate) vce(cluster mdate)

Which makes me think that there is something weird going on with ivreghdfe.

Would be very happy about any ideas what could be going wrong here, or what I could be doing wrong.

Thanks!

[Johanna Krenz]

Hi,

I would like to add that I ran some more test and I also get the "conformability error" when I try to "trick" ivreghdfe by using time dummies directly instead of in the absorb option, i.e.

qui tabulate mdate, gen(dmonth)
ivreghdfe delta_GDP1 (c.`shock'#group=c.`instr'#group) L(`q'/`p').c.int_GDP1 dmonth*, absorb(group) vce(cluster mdate)

and also when I use ivreg2, i.e.

qui tabulate mdate, gen(dmonth)
qui tabulate countrykey, gen(dcountry)
ivreg2 delta_GDP1 (c.`shock'#group=c.`instr'#group) L(`q'/`p').c.int_GDP1 dcountry* dmonth*, cluster(mdate)

So, does anyone has an idea why I get a conformabiltiy error when I try to find the country-specific effects of my instrumented shock when using time FE? But not when I use the instrument directly as a shock (reghdfe delta_GDP1 c.`instr'#i.group L(`q'/`p').c.int_GDP1, absorb(group mdate) vce(cluster mdate))?

Thanks for any ideas!