Predicting residuals from instrumental regression

[Raja Hasan]

Model : ivreghdfe itotf (qt = lagqt lagcfo) cfo, absorb(gvkey fyear) endog(qt) cluster(gvkey) resid

----- Resid is not allowed

Another way I can find residuals is by predicting yhat and measuring the difference between DV and yhat. Though I can predict yhat from the above regression,
the yhat values are wrong.

Can any one help me either predict yhat correctly or residuals?

[Joro Kolev]

- ivreghdfe- is a user written command, and you should read the help file how you generate residuals with it.

If it follows standard syntax, it would be something like

 ivreghdfe itotf (qt = lagqt lagcfo) cfo, absorb(gvkey fyear) endog(qt) cluster(gvkey) 

predict myresiduals, resid

[Raja Hasan]

Thanks, Joro

For reghdfe this is not possible. That's the problem

[Joro Kolev]

Write to the authors, and tell them that what they have done does not really make sense :-).

Indeed, you are right, -ivreghdfe- takes us in circles, and it does not seem possible to generate residuals after it

. sysuse auto, clear
(1978 Automobile Data)

. qui ivreghdfe price (mpg = weight length), absorb(rep)

. predict myres, resid
you must add the resid option to reghdfe before running this prediction
r(9);

. ivreghdfe price (mpg = weight length), absorb(rep) resid
option resid not allowed
r(198);

Thanks, Joro

For reghdfe this is not possible. That's the problem

[Joro Kolev]

This might be what you need

. sysuse auto, clear
(1978 Automobile Data)

. qui ivreghdfe price (mpg = weight length), absorb(rep)

. predict pricehat
(option xb assumed; fitted values)

. gen resid = price - pricehat

[Raja Hasan]

Hi Joro,

Thank you very much.
Could you please tell me how I can do the #1 model using ivreg2?

[Joro Kolev]

Try the following, Raja:

ivreghdfe itotf (qt = lagqt lagcfo) cfo i.gvkey i.fyear, partial(i.gvkey i.fyear) cluster(gvkey)

It did work on the Mickey Mouse dataset. I do not know whether it is going to work on your large dataset.

. sysuse auto, clear
(1978 Automobile Data)

. ivreg2 price (mpg = weight length) i.rep, partial(i.rep)

IV (2SLS) estimation
--------------------

Estimates efficient for homoskedasticity only
Statistics consistent for homoskedasticity only

                                                      Number of obs =       69
                                                      F(  1,    63) =    22.20
                                                      Prob > F      =   0.0000
Total (centered) SS     =  568436416.2                Centered R2   =   0.2135
Total (uncentered) SS   =  568436416.2                Uncentered R2 =   0.2135
Residual SS             =  447080951.8                Root MSE      =     2545

------------------------------------------------------------------------------
       price |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         mpg |   -384.088   77.88926    -4.93   0.000    -536.7482   -231.4279
------------------------------------------------------------------------------
Underidentification test (Anderson canon. corr. LM statistic):          41.152
                                                   Chi-sq(2) P-val =    0.0000
------------------------------------------------------------------------------
Weak identification test (Cragg-Donald Wald F statistic):               45.809
Stock-Yogo weak ID test critical values: 10% maximal IV size             19.93
                                         15% maximal IV size             11.59
                                         20% maximal IV size              8.75
                                         25% maximal IV size              7.25
Source: Stock-Yogo (2005).  Reproduced by permission.
------------------------------------------------------------------------------
Sargan statistic (overidentification test of all instruments):          13.035
                                                   Chi-sq(1) P-val =    0.0003
------------------------------------------------------------------------------
Instrumented:         mpg
Excluded instruments: weight length
Partialled-out:       2.rep78 3.rep78 4.rep78 5.rep78 _cons
                      nb: total SS, model F and R2s are after partialling-out;
                          any small-sample adjustments include partialled-out
                          variables in regressor count K
------------------------------------------------------------------------------

.

Hi Joro,

Thank you very much.
Could you please tell me how I can do the #1 model using ivreg2?

[Raja Hasan]

Thanks Joro, For large dataset, this is a problem

[Jeff Wooldridge]

Try this:

xtset gvkey fyear
xtivreg itotf (qt = lagqt lagcfo) cfo i.fyear, fe
predict resid, e

[Raja Hasan]

Thank you very much Jeff. It has been extremely helpful.

[Silvia Peracchi]

Hi, I was returned the same error message when trying to run the ivreghdfe command with a resid(res) option to save residuals.

What worked for me was to re-install the packages via the installation codes available in the github code source (https://github.com/sergiocorreia/ivreghdfe#installation).

I had previously used the ssc install syntax to install all these packages, and the net install alternative suggested in github solved the issue.
My impression is that some updates available in github are not available in the ssc repo.