Linear Models With Many Levels of Fixed Effects( REGHDFE)

lal mohan kumar

Join Date: May 2019

Posts: 265
#1

Linear Models With Many Levels of Fixed Effects( REGHDFE)

03 Oct 2020, 01:41

Dear Stata Members

I ran the following panel regression with firm fixed effects(in my case, xtset id year; where id denotes individual firms) and year dummies, and standard errors clustered at firm level

Code:

xtreg cash_ta2_w epu /// tra_cre2_w inven_ta_w /// size_w fixed_ass_w lever_w ltbankborr2_w stbankborr2_w rd_ta_w div_w /// sgrowth_w oppro_ta_w mb_w /// fcfvol icfvol fsalvol isalvol /// realinterestrate gdpgrowthannual /// i.year, fe vce(robust)

Now I want standard errors to be clustered by both firm and year to control for the possibility that standard errors are correlated across firms and over time.

I ran the following code

Code:

reghdfe cash_ta2_w epu /// tra_cre2_w inven_ta_w /// size_w fixed_ass_w lever_w ltbankborr2_w stbankborr2_w rd_ta_w div_w /// sgrowth_w oppro_ta_w mb_w /// fcfvol icfvol fsalvol isalvol /// realinterestrate gdpgrowthannual /// i.year, absorb(id) vce(cluster id year)

Did I run the right code?

References

HTML Code:

http://scorreia.com/software/reghdfe/
Tags: None
Carlo Lazzaro

Join Date: Apr 2014

Posts: 17699
#2

03 Oct 2020, 03:43

Ial:
your code seems correct.
Obviously, it would have been interesting to see, in addition:
-how you -xtset- your data before -xtreg-;
- the output of both codes that Stata gave you back.

Kind regards,
Carlo
(Stata 19.0)
Comment

lal mohan kumar

Join Date: May 2019
Posts: 265

03 Oct 2020, 04:27

Dear Carlo
Thanks for your prompt reply and guidance

how you -xtset- your data before -xtreg-;

Code:

encode companyname,gen(id)           //companyname denotes name of individual firms
xtset id year

Code:

. xtreg cash_ta2_w epu ///
> tra_cre2_w inven_ta_w ///
> size_w fixed_ass_w lever_w ltbankborr2_w stbankborr2_w rd_ta_w div_w ///
> sgrowth_w oppro_ta_w mb_w  ///
> fcfvol icfvol fsalvol isalvol ///
> realinterestrate gdpgrowthannual ///
> i.year, fe vce(robust)
note: 2008.year omitted because of collinearity
note: 2010.year omitted because of collinearity
note: 2012.year omitted because of collinearity

Fixed-effects (within) regression               Number of obs      =     23488
Group variable: id                              Number of groups   =      2773

R-sq:  within  = 0.1466                         Obs per group: min =         1
       between = 0.1991                                        avg =       8.5
       overall = 0.1990                                        max =        15

                                                F(30,2772)         =     20.62
corr(u_i, Xb)  = -0.0981                        Prob > F           =    0.0000

                                      (Std. Err. adjusted for 2773 clusters in id)
----------------------------------------------------------------------------------
                 |               Robust
      cash_ta2_w |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-----------------+----------------------------------------------------------------
             epu |   .0149305   .0024643     6.06   0.000     .0100985    .0197625
      tra_cre2_w |  -.1257609   .0098413   -12.78   0.000    -.1450578   -.1064639
      inven_ta_w |  -.2111026   .0137582   -15.34   0.000    -.2380801   -.1841252
          size_w |   -.001439   .0019053    -0.76   0.450    -.0051748    .0022969
     fixed_ass_w |  -.1758122   .0103571   -16.97   0.000    -.1961207   -.1555037
         lever_w |   .0056474   .0076046     0.74   0.458    -.0092639    .0205587
   ltbankborr2_w |  -.0241265   .0086593    -2.79   0.005    -.0411058   -.0071471
   stbankborr2_w |  -.0186561   .0092892    -2.01   0.045    -.0368705   -.0004416
         rd_ta_w |  -.0865674   .1515961    -0.57   0.568    -.3838202    .2106854
           div_w |   .4987488   .0830426     6.01   0.000     .3359171    .6615805
       sgrowth_w |  -.0014762   .0017478    -0.84   0.398    -.0049034     .001951
      oppro_ta_w |   .0958302   .0123161     7.78   0.000     .0716804    .1199799
            mb_w |   .0054963   .0011377     4.83   0.000     .0032656     .007727
          fcfvol |   .0104895   .0116456     0.90   0.368    -.0123455    .0333244
          icfvol |   .0352893   .0405422     0.87   0.384    -.0442067    .1147852
         fsalvol |   .0085543   .0042196     2.03   0.043     .0002804    .0168281
         isalvol |  -.0106887    .009929    -1.08   0.282    -.0301577    .0087803
realinterestrate |   .0006745   .0003318     2.03   0.042     .0000239    .0013252
 gdpgrowthannual |  -.0001855   .0003541    -0.52   0.600    -.0008798    .0005089
                 |
            year |
           2006  |   .0105316   .0018659     5.64   0.000     .0068728    .0141903
           2007  |   .0084752   .0021651     3.91   0.000     .0042297    .0127207
           2008  |          0  (omitted)
           2009  |  -.0088026   .0021055    -4.18   0.000    -.0129312    -.004674
           2010  |          0  (omitted)
           2011  |    .007997   .0014722     5.43   0.000     .0051102    .0108838
           2012  |          0  (omitted)
           2013  |  -.0029671   .0015151    -1.96   0.050     -.005938    3.84e-06
           2014  |   .0005053   .0027044     0.19   0.852    -.0047974    .0058081
           2015  |   .0022666   .0033117     0.68   0.494    -.0042272    .0087603
           2016  |   .0056691   .0035223     1.61   0.108    -.0012375    .0125757
           2017  |   .0045639   .0032355     1.41   0.158    -.0017804    .0109081
           2018  |   .0047014   .0033819     1.39   0.165    -.0019298    .0113327
           2019  |   .0037849   .0039063     0.97   0.333    -.0038747    .0114446
                 |
           _cons |   .0866509   .0167529     5.17   0.000     .0538015    .1195004
-----------------+----------------------------------------------------------------
         sigma_u |  .06773354
         sigma_e |  .05018485
             rho |  .64559608   (fraction of variance due to u_i)
----------------------------------------------------------------------------------

Code:

. reghdfe cash_ta2_w epu ///
> tra_cre2_w inven_ta_w ///
> size_w fixed_ass_w lever_w ltbankborr2_w stbankborr2_w rd_ta_w div_w ///
> sgrowth_w oppro_ta_w mb_w  ///
> fcfvol icfvol fsalvol isalvol ///
> realinterestrate gdpgrowthannual ///
> ,absorb(id) vce(cluster id year)
(dropped 278 singleton observations)
(MWFE estimator converged in 1 iterations)

HDFE Linear regression                            Number of obs   =     23,210
Absorbing 1 HDFE group                            F(  19,     14) =      27.63
Statistics robust to heteroskedasticity           Prob > F        =     0.0000
                                                  R-squared       =     0.6337
                                                  Adj R-squared   =     0.5892
Number of clusters (id)      =      2,495         Within R-sq.    =     0.1427
Number of clusters (year)    =         15         Root MSE        =     0.0503

                                   (Std. Err. adjusted for 15 clusters in id year)
----------------------------------------------------------------------------------
                 |               Robust
      cash_ta2_w |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-----------------+----------------------------------------------------------------
             epu |   .0063721   .0024511     2.60   0.021      .001115    .0116292
      tra_cre2_w |  -.1244352   .0096545   -12.89   0.000     -.145142   -.1037284
      inven_ta_w |  -.2085149   .0152315   -13.69   0.000    -.2411832   -.1758466
          size_w |  -.0002334   .0016184    -0.14   0.887    -.0037045    .0032377
     fixed_ass_w |  -.1780021   .0120293   -14.80   0.000    -.2038024   -.1522017
         lever_w |    .005923   .0082606     0.72   0.485    -.0117943    .0236403
   ltbankborr2_w |  -.0255102   .0087729    -2.91   0.011    -.0443261   -.0066942
   stbankborr2_w |  -.0189028   .0095001    -1.99   0.067    -.0392785    .0014728
         rd_ta_w |   -.077844   .1464953    -0.53   0.603    -.3920452    .2363573
           div_w |   .5102609   .0971987     5.25   0.000     .3017903    .7187315
       sgrowth_w |  -.0009679   .0022725    -0.43   0.677     -.005842    .0039062
      oppro_ta_w |   .0922582    .011566     7.98   0.000     .0674516    .1170648
            mb_w |   .0057251   .0012024     4.76   0.000     .0031462     .008304
          fcfvol |    .010845   .0127056     0.85   0.408    -.0164059    .0380958
          icfvol |   .0357164   .0421566     0.85   0.411    -.0547006    .1261333
         fsalvol |   .0089683   .0062749     1.43   0.175    -.0044901    .0224266
         isalvol |   -.004646   .0091316    -0.51   0.619    -.0242314    .0149393
realinterestrate |   .0002678   .0004508     0.59   0.562    -.0006991    .0012347
gdpgrowthannual |   -.000237    .000705    -0.34   0.742     -.001749     .001275
           _cons |   .1161762   .0191768     6.06   0.000      .075046    .1573064
----------------------------------------------------------------------------------

Absorbed degrees of freedom:
-----------------------------------------------------+
 Absorbed FE | Categories  - Redundant  = Num. Coefs |
-------------+---------------------------------------|
          id |      2495        2495           0    *|
-----------------------------------------------------+
* = FE nested within cluster; treated as redundant for DoF computation

However, if I put absorb(year id), I will get the results below

Code:

. reghdfe cash_ta2_w epu ///
> tra_cre2_w inven_ta_w ///
> size_w fixed_ass_w lever_w ltbankborr2_w stbankborr2_w rd_ta_w div_w ///
> sgrowth_w oppro_ta_w mb_w  ///
> fcfvol icfvol fsalvol isalvol ///
> realinterestrate gdpgrowthannual ///
> ,absorb(id year) vce(cluster id year)
(dropped 278 singleton observations)
note: epu is probably collinear with the fixed effects (all partialled-out values are close to zero; tol = 1.0e-09)
note: realinterestrate is probably collinear with the fixed effects (all partialled-out values are close to zero; tol = 1.0
> e-09)
note: gdpgrowthannual is probably collinear with the fixed effects (all partialled-out values are close to zero; tol = 1.0e
> -09)
(MWFE estimator converged in 7 iterations)
Warning: VCV matrix was non-positive semi-definite; adjustment from Cameron, Gelbach & Miller applied.
note: epu omitted because of collinearity
note: realinterestrate omitted because of collinearity
note: gdpgrowthannual omitted because of collinearity

HDFE Linear regression                            Number of obs   =     23,210
Absorbing 2 HDFE groups                           F(  16,     14) =      28.78
Statistics robust to heteroskedasticity           Prob > F        =     0.0000
                                                  R-squared       =     0.6353
                                                  Adj R-squared   =     0.5908
Number of clusters (id)      =      2,495         Within R-sq.    =     0.1395
Number of clusters (year)    =         15         Root MSE        =     0.0502

                                   (Std. Err. adjusted for 15 clusters in id year)
----------------------------------------------------------------------------------
                 |               Robust
      cash_ta2_w |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-----------------+----------------------------------------------------------------
             epu |          0  (omitted)
      tra_cre2_w |  -.1257609   .0094913   -13.25   0.000    -.1461176   -.1054041
      inven_ta_w |  -.2111026   .0152594   -13.83   0.000    -.2438309   -.1783744
          size_w |   -.001439   .0022175    -0.65   0.527     -.006195    .0033171
     fixed_ass_w |  -.1758122   .0123278   -14.26   0.000    -.2022526   -.1493717
         lever_w |   .0056474   .0082831     0.68   0.506    -.0121181    .0234128
   ltbankborr2_w |  -.0241265   .0090278    -2.67   0.018    -.0434891   -.0047638
   stbankborr2_w |  -.0186561   .0096456    -1.93   0.074    -.0393438    .0020317
         rd_ta_w |  -.0865674   .1450336    -0.60   0.560    -.3976335    .2244986
           div_w |   .4987488   .1003184     4.97   0.000     .2835872    .7139104
       sgrowth_w |  -.0014762   .0024878    -0.59   0.562     -.006812    .0038596
      oppro_ta_w |   .0958302   .0111398     8.60   0.000     .0719376    .1197228
            mb_w |   .0054963   .0011677     4.71   0.000     .0029919    .0080007
          fcfvol |   .0104895   .0122686     0.85   0.407     -.015824     .036803
          icfvol |   .0352893   .0451034     0.78   0.447    -.0614479    .1320265
         fsalvol |   .0085543   .0061449     1.39   0.186    -.0046253    .0217339
         isalvol |  -.0106887   .0091993    -1.16   0.265    -.0304192    .0090419
realinterestrate |          0  (omitted)
 gdpgrowthannual |          0  (omitted)
           _cons |   .1578272   .0178302     8.85   0.000     .1195852    .1960693
----------------------------------------------------------------------------------

Absorbed degrees of freedom:
-----------------------------------------------------+
 Absorbed FE | Categories  - Redundant  = Num. Coefs |
-------------+---------------------------------------|
          id |      2495        2495           0    *|
        year |        15          15           0    *|
-----------------------------------------------------+
* = FE nested within cluster; treated as redundant for DoF computation

.

Variables realinterestrate gdpgrowthannual & epu are same for all firms for a particular year as they are macro in nature (cross-sectionally invariant). I suspect that could be the reason why epu, realinterestrate & gdpgrowthannual got omitted when I put year inside the absorb.

Last edited by lal mohan kumar; 03 Oct 2020, 04:31.

Comment

Carlo Lazzaro

Join Date: Apr 2014

Posts: 17699
#4

03 Oct 2020, 04:52

Ial:
thanks for clarifying.
I would stop at -abs(id)- with standard errors clustered on both -id- and -year-.

Kind regards,
Carlo
(Stata 19.0)
Comment
lal mohan kumar

Join Date: May 2019

Posts: 265
#5

03 Oct 2020, 04:56

Thanks Carlo
Comment

Announcement