Regression output

[Marco Errico]

Hi all,

I'm running a series of regression with industry and year fixed effect
My dependent variable is a proxy for manipulation of earnings. My key variable of interest is a dummy (quint1) that doesn't change over time and after I have a series of controls.
Since I have several proxies of manipulation of earnings I'm running multiple regression.
I'm struggling to understand why, for some of the regressions, I have this output.

reg absmoddd quintB1 aver_shareturn analysts_log logat salevolatility cashvolatility workcapit proploss delsalegrowth earn futearn ppe deltaWRC changesale cfo accr bigaud roa i.fyear i.ffind, cluster(gvkey)
note: 2016.fyear omitted because of collinearity
note: 2017.fyear omitted because of collinearity
note: 2018.fyear omitted because of collinearity

Linear regression                               Number of obs     =     38,454
                                                F(54, 5145)       =          .
                                                Prob > F          =          .
                                                R-squared         =     0.5437
                                                Root MSE          =     .08192

                               (Std. Err. adjusted for 5,146 clusters in gvkey)
-------------------------------------------------------------------------------
              |               Robust
     absmoddd |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
--------------+----------------------------------------------------------------
      quintB1 |   .0021172   .0063768     0.33   0.740     -.010384    .0146185
aver_sharet~n |   4.83e-08   4.07e-09    11.86   0.000     4.03e-08    5.63e-08
 analysts_log |  -.0030811   .0005967    -5.16   0.000    -.0042508   -.0019114
        logat |  -.0055064   .0003689   -14.93   0.000    -.0062296   -.0047832
salevolatil~y |   1.110494          .        .       .            .           .
cashvolatil~y |   2.592445          .        .       .            .           .
    workcapit |   .6596937          .        .       .            .           .
     proploss |  -3.900873          .        .       .            .           .
delsalegrowth |   .0014231   .0008765     1.62   0.105    -.0002953    .0031414
         earn |  -.2315857   .0174807   -13.25   0.000    -.2658552   -.1973161
      futearn |  -.0102936   .0073057    -1.41   0.159    -.0246158    .0040287
          ppe |  -.0255437   .0037311    -6.85   0.000    -.0328583   -.0182291
     deltaWRC |   .1210259   .1170141     1.03   0.301    -.1083714    .3504233
   changesale |   .0291029   .0031282     9.30   0.000     .0229703    .0352355
          cfo |   .1973238   .0143676    13.73   0.000     .1691573    .2254903
         accr |   .8036271   .2924855     2.75   0.006     .2302311    1.377023
       bigaud |   -.182379          .        .       .            .           .
          roa |  -.0468239   .0108017    -4.33   0.000    -.0679998   -.0256479
              |
        fyear |
        2002  |   .0684815          .        .       .            .           .
        2003  |   .0275448          .        .       .            .           .
        2004  |  -.0781455          .        .       .            .           .
        2005  |  -.1743677          .        .       .            .           .
        2006  |  -.3005914          .        .       .            .           .
        2007  |  -.3505113          .        .       .            .           .
        2008  |  -.2788789          .        .       .            .           .
        2009  |  -.1475632          .        .       .            .           .
        2010  |  -.0795899          .        .       .            .           .
        2011  |  -.0488357          .        .       .            .           .
        2012  |  -.0239605          .        .       .            .           .
        2013  |  -.0673017          .        .       .            .           .
        2014  |   -.136407          .        .       .            .           .
        2015  |  -.0694582          .        .       .            .           .
        2016  |          0  (omitted)
        2017  |          0  (omitted)
        2018  |          0  (omitted)
              |
        ffind |
           2  |   .0165133   .0038744     4.26   0.000     .0089178    .0241088
           3  |   .0061173   .0060999     1.00   0.316     -.005841    .0180757
           4  |   .0137559   .0068625     2.00   0.045     .0003026    .0272092
           6  |   .0261899   .0050341     5.20   0.000      .016321    .0360588
           7  |   .0271067   .0040535     6.69   0.000     .0191601    .0350534
           8  |   .0307849   .0091985     3.35   0.001      .012752    .0488178
           9  |   .0121585   .0038337     3.17   0.002     .0046428    .0196742
          10  |   .0129211   .0041018     3.15   0.002     .0048799    .0209624
          11  |   .0387873   .0049461     7.84   0.000     .0290907    .0484838
          12  |   .0333732   .0042868     7.79   0.000     .0249692    .0417772
          13  |   .0778667   .0053883    14.45   0.000     .0673034    .0884301
          14  |   .0248361    .007387     3.36   0.001     .0103544    .0393177
          15  |   .0124818   .0077165     1.62   0.106    -.0026458    .0276094
          16  |   .0252366   .0113663     2.22   0.026     .0029538    .0475195
          17  |   .0180586   .0073686     2.45   0.014     .0036129    .0325043
          18  |   .0174014   .0080347     2.17   0.030       .00165    .0331528
          19  |   .0161763   .0074397     2.17   0.030     .0015914    .0307612
          20  |  -.0102544   .0102338    -1.00   0.316     -.030317    .0098081
          21  |   .0195901   .0073257     2.67   0.008     .0052287    .0339515
          22  |   .0173262   .0078094     2.22   0.027     .0020165    .0326359
          23  |    .021891   .0074394     2.94   0.003     .0073066    .0364754
          24  |   .0132493     .00759     1.75   0.081    -.0016303    .0281289
          25  |   .0011462   .0076273     0.15   0.881    -.0138066     .016099
          26  |   .0048563   .0175949     0.28   0.783    -.0296372    .0393498
          27  |   .0379749   .0078397     4.84   0.000     .0226057     .053344
          28  |   .0451368   .0079863     5.65   0.000     .0294803    .0607933
          29  |   .0075163   .0079214     0.95   0.343    -.0080129    .0230455
          30  |   .0469983   .0075624     6.21   0.000     .0321727    .0618239
          33  |   .0141865   .0074641     1.90   0.057    -.0004464    .0288194
          34  |   .0422195   .0072676     5.81   0.000     .0279718    .0564671
          35  |   .0408326   .0075694     5.39   0.000     .0259933    .0556719
          36  |   .0478678   .0073668     6.50   0.000     .0334258    .0623099
          37  |   .0174294    .007406     2.35   0.019     .0029105    .0319483
          38  |   .0110031   .0074203     1.48   0.138    -.0035438    .0255501
          39  |   .0037448   .0075109     0.50   0.618    -.0109796    .0184693
          40  |   .0172039   .0082013     2.10   0.036     .0011259     .033282
          41  |   .0172645     .00734     2.35   0.019      .002875     .031654
          42  |   .0130322   .0072594     1.80   0.073    -.0011993    .0272637
          43  |   .0102852   .0077753     1.32   0.186    -.0049577    .0255281
          46  |   .0520101   .0101473     5.13   0.000     .0321171     .071903
          47  |   .0435883   .0092769     4.70   0.000     .0254015     .061775
-------------------------------------------------------------------------------

It is clear that some of the year dummies are omitted thanks to multicollinearity.
But why do I have a series of dots between year dummies 2002 and 2015 for the columns of std error, p statistcs etc.?
The same problem happens also for three controls (cashvolatility, salesvolatility, proprloss)
Do you have any suggestions?
Thanks

[Carlo Lazzaro]

Marco:
you may want to take a look at this old Stata thread: https://www.stata.com/statalist/arch.../msg00860.html.
In addition: does the same problem creep up when you use default standard errors?
Eventually: why not switching to -xtreg,fe-?

[Marco Errico]

Dear Carlo Lazzaro ,
Many thanks for your reply.
I'm nout using firm fixed effect (xtreg, fe) since my key variable of interest is a dummy that doesn't change over time (the dummy is if the firm appears in a first poistion when listed in alphabetical order).
I read that thread and if I will not use cluster, I have no problems (selecting no options for the errors)
I also tried with the option robust, but the problem appears sometimes.
Do you have any suggestion regarding the source of the problem?
As far as I know, if I cluster standard error I'm controlling for heteroskedasticity and autocorrelations of the residuals. With robust only for heteroskedasticity.
Am I wrong?
Thanks in advance

[Carlo Lazzaro]

Marco:
If I got you right, you use -regress- in order to avoid the wiping-out if the time-invariant predictors you're intrested in.
However, there's might be a consistency issue, as you do not know which specification (-fe- or -re-) fits your data better: that said, I woud give -xtreg- a try; if -fe- is the way to go, you can consider an hybrid model (search the community-contribute command -xthybrid-); if -re- is the way to go, you wlll have all your coefficients estimate (under the tight constraints that panel-wise effect is not correlated with the vector of regressors).
Eventually, you're right about your last statement about non-default standard error options and their job.

[Marco Errico]

Dear Carlo,
Thank you, Actually I'm not an expert of -re-. I immediately avoi the option -xtreg, fe- since my dumy doesn't change over time.
I will search the command that you suggest me.
Anyway, it seems that the dots are displayed when some variables are highmulticollinear and stata doesn't drop them automatically. Indeed, after thre gression i write -vif- and I noticed that the variables with the dots (instead of displaying SE-t statistics and confidence interval) have a vif of more than 2500

[Carlo Lazzaro]

Marco:
hence you have to re-consider your model getting rid of variables that create quasi-extreme multicollinearity.