eivreg in rolling window setting (estimates not stored properly)

[Andres Salman]

Dear Statalist,

I might be shooting my shot with this post but I want to implement the eivreg command in a rolling-windows setting using an unbalanced panel (id x tid).
Currently, I am able to run the program but the estimates are not stored correctly. I would like to save the regression coefficients in a _b_* variables and the t-stats in _t_* variables.

The specifications for the rolling windows are the last 36 months (minimum 12 months of observations) by id. r_rf, mkt, and smb are assumed to be random variables.
The matrix r(table) stores the estimates for each window.

From similar threads (with other regressors), I ended up with this code.

g _b_mkt=.
g _b_smb=.
g _t_mkt=.
g _t_smb=.

capture program drop eiv_rol
program define eiv_rol

    qui: eivreg  r_rf mkt  smb ,  r(mkt 0.6 smb 0.6) 
    matrix mattest= r(table)
    
    replace _b_mkt= mattest[1, 1] if mi(_b_mkt)
    replace _b_hml= mattest[1, 2] if mi(_b_smb)
    replace _t_mkt= mattest[3, 1] if mi(_t_mkt)
    replace _t_hml= mattest[3, 2] if mi(_t_smb)
    
    exit
end

rangerun eiv_rol, by(id) interval(tid -36 -1)

If the estimates can be stored properly, I want to re-run the program changing the reliability (in this example case the reliability for both variables is 0.6). I am planning to increment by .1 in each step from 0 to 1. What other procedure would you recommend to speed up the process?

Another problem that I am facing is that I cannot break the code in between. I need to wait until the end to made changes, which might take several minutes. Do you know how to change this behavior?
Thank you

Ricardo

[Andres Salman]

I think I got it. The values should be generated inside the program instead of replacing them. Here is an example with data.

webuse grunfeld, clear

xtset company  year
capture program drop eiv_rol
program define eiv_rol
        xtset id tid
        qui: eivreg   invest mvalue kstock ,  r(mvalue 0.95 kstock 0.95)  
        matrix mattest= r(table)    
        foreach v in  mvalue kstock  {
        gen b_`v' = _b[`v']
        gen t_`v' = _b[`v']/_se[`v']
        }
exit
end

rangerun eiv_rol, by(company) interval (year -6 -1) verbose