reghdfe with nocons option

[Lucy Garcia]

Dear statalist,

I want to run a regression with firm and year fixed effects, with firm-clustered standard errors, and without the constant term, I believe -reghdfe- can be combined with -nocons-, but why when I run the code below, stata says "invalid options: nocons"?

reghdfe y x1 x2 x3 x4 x5 x6, absorb(firm year) nocons vce(cluster firm)

Thanks a lot for any suggestion!

[Carlo Lazzaro]

Lucy:
works for me in the following toy-example:

use "https://www.stata-press.com/data/r17/nlswork.dta"
. reghdfe ln_wage c.age##c.age , abs(idcode year) vce(cluster idcode) nocons
(dropped 551 singleton observations)
(MWFE estimator converged in 8 iterations)

HDFE Linear regression                            Number of obs   =     27,959
Absorbing 2 HDFE groups                           F(   2,   4158) =      44.91
Statistics robust to heteroskedasticity           Prob > F        =     0.0000
                                                  R-squared       =     0.6593
                                                  Adj R-squared   =     0.5995
                                                  Within R-sq.    =     0.0115
Number of clusters (idcode)  =      4,159         Root MSE        =     0.3013

                             (Std. err. adjusted for 4,159 clusters in idcode)
------------------------------------------------------------------------------
             |               Robust
     ln_wage | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
         age |   .0728746   .0136873     5.32   0.000     .0460402    .0997089
             |
 c.age#c.age |  -.0010113   .0001076    -9.39   0.000    -.0012224   -.0008003
------------------------------------------------------------------------------

Absorbed degrees of freedom:
-----------------------------------------------------+
 Absorbed FE | Categories  - Redundant  = Num. Coefs |
-------------+---------------------------------------|
      idcode |      4159        4159           0    *|
        year |        15           0          15     |
-----------------------------------------------------+
* = FE nested within cluster; treated as redundant for DoF computation

.

[Lucy Garcia]

Hi Carlo,

Thanks a lot for your reply. How strange! My code is similar to yours, except for I have 6 IVs and some control variables, and I don't have the interaction term. To be more specific, my 6 IVs are 6 time dummies, which represents 3 years before an event (enforcement of a law) happens, 2 years before, 1 year before, the event year, 1 year after, and 2 years after (pre3, pre2, pre1, event, post1, post2); the DV, error, is a continuous variable that measures analyst forecast error.

reghdfe error pre3 pre2 pre1 event post1 post2 controls, absorb(firm year) vce(cluster firm) nocons

And stata says

invalid options: nocons
r(198);

I could use -nocons- if I use reg instead.

reg error pre3 pre2 pre1 event post1 post2 controls i.firm i.year ,vce(cluster firm) nocons

------------------------------------------------------------------------------------
                   |               Robust
             error |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------------+----------------------------------------------------------------
              pre3 |   -1.43105   2.077697    -0.69   0.491    -5.510459    2.648359
              pre2 |  -2.310634   2.163936    -1.07   0.286    -6.559368    1.938099
              pre1 |  -1.541373   2.072042    -0.74   0.457    -5.609679    2.526933
             event |  -1.890391   2.030509    -0.93   0.352    -5.877149    2.096368
             post1 |  -1.700703   2.093798    -0.81   0.417    -5.811725     2.41032
             post2 |  -1.620117   1.934716    -0.84   0.403    -5.418793    2.178558

Can you spot anything wrong with the code using -reghdfe-? Why it doesn't work the same way as using reg?

Thanks!

[Andrew Musau]

ssc install reghdfe, replace

To be more specific, my 6 IVs are 6 time dummies, which represents 3 years before an event (enforcement of a law) happens, 2 years before, 1 year before, the event year, 1 year after, and 2 years after (pre3, pre2, pre1, event, post1, post2); the DV, error, is a continuous variable that measures analyst forecast error.

If the event occurs at the same year for all units (analysts) in your panel, then the event, pre and post dummies are collinear with the year dummies. In any case, if your goal is to estimate these coefficients, then they are not identified in the case that they are collinear with the year effects. The -nocons- option in reghdfe is essentially a "hide cons" option, and will not affect what is omitted from the estimation. You have to go back and respecify the model. If your intention is, e.g., to examine the impact of the enforcement of the law on the outcome variable, then define a treatment dummy (=1 if analyst \(i\) in year \(t\) was subject to the law, and 0 otherwise) and then estimate the regression:

xtset analyst_id year
xtreg outcome treatment controls i.year, fe robust

[Lucy Garcia]

Hi Andrew,

Thanks for your help! The event doesn't occur in the same year for all firms; the enforcement of law is a staggered implementation, firms in different industries have enforcement in different years (sorry that I should make this clear earlier).

I think my -reghdfe- if not up to date. When I ssc install reghdfe, replace, it works.