reghdfe makes Mata forget Mata functions

[Peter Raupach]

Hi, I am a beginner with Mata and also have not written Stata programs (.ado files).

Inside a larger loop I run reghdfe after which I do some evaluations in Mata. I learned this can be only a one-liner (otherwise the loop breaks at Mata's "end") so I need to define a Mata function called by the one-liner. So far so good. My problem is that once reghdfe has run, the previously defined Mata function is forgot. Here is a MWE:

cls
clear all
adopath + // path to reghdfe.ado goes here

mata
    real matrix mtest()     { 
        return(runiform(3,4))
    }
end

mata: st_matrix("rt1",mtest()) // works
mat li rt1

sysuse auto
qui reghdfe price foreign mpg, absorb(i.rep78) // reghdfe makes mata forget the function mtest()

mata: st_matrix("rt2",mtest()) // error <istmt>:  3499  mtest() not found
mat li rt2

Somehow (I guess) I have to declare this function mtest() permanently by other means (not just in the .do file) but as I have never written any of these .ado programs or whatever kind of wrapper is needed here, I got stuck. Any idea?

Peter

[David Roodman]

I think reghdfe may do something clever where on the fly it compiles a version of the Mata code that is optimized for your estimation problem. This might include a "mata clear" command.
You could also post on the issues page at Github: https://github.com/sergiocorreia/reghdfe/issues.
Sergio Correia

[Peter Raupach]

Thank you David. I think I have to cope with reghdfe as it is. Finally I found a solution. While calling the mata function from a program...end wrapper (still defined inside the .do file) did not help, outsourcing the mata stuff and the wrapper to an ado file does help. This file seems to be unaffected by reghdfe's deletions (or it is re-defined each time I call it; luckily that's not critical to my application)