Missing F-Stat for regression with robust SE - Not a problem with singleton dummy variables

Cecilia Roesen

Join Date: Nov 2018

Posts: 12
#1

Missing F-Stat for regression with robust SE - Not a problem with singleton dummy variables

14 Dec 2018, 06:13

Dear Statalist community,

Currently I am conducting an event study on ad hoc disclosures. The regression model I use contains some evidence for heteroscedasticity so I decided to use robust SE. Regrettably, Stata does not report an F-statistic in this case (when I use the “normal” –reg- command it does report it by the way).

I googled a lot and read other Statalist threads but they always report problems with singleton dummy variables. My model contains 17 dummy variables but none of them seems to be a singleton dummy variable. Also I don't get an error message or something like that. Does anybody have an idea what the problem might be?

The command I used to regress using robust SE is the following:

Code:

reg CAR_gesamt Regelpublizität Dividenden Aktienrückkauf Anleihe Kapitalerhöhung MA Kooperation Beteiligung Verkauf Personalveränderung Rechtssache Strategie Auftragseingang Insolvenz Sonstiges Zeit GeneralStandard Scale MarketCapln TotalAssetsln ReturnonAssets if n_disclosurecount_tag==1, vce(robust)

CAR_gesamt, MarketCapln, TotalAssetsln, ReturnonAssets and Zeit are numeric variables, the rest of them are dummy variables.

If anybody could help me I would be really grateful!

Thanks in advance,

Cecilia
Tags: None
William Lisowski

Join Date: Dec 2014

Posts: 10150
#2

14 Dec 2018, 07:33

Perhaps you are overlooking something in your regression output that will provide further information.

As section 12.1 of the Statalist FAQ tells us,

12.1 What to say about your commands and your problem

Say exactly what you typed and exactly what Stata typed (or did) in response. N.B. exactly!.

It would be particularly helpful to copy your command and output from your Stata Results window and paste that into your Statalist post using code delimiters [CODE] and [/CODE], as described in section 12 of the FAQ.

The more you help others understand your problem, the more likely others are to be able to help you solve your problem.
Comment

Cecilia Roesen

Join Date: Nov 2018
Posts: 12

14 Dec 2018, 08:01

I am sorry, you are right. I hoped it would be more like a general problem, so I did not post the output.

I actually have three outputs because I have three dependent variables: CAR_gesamt which means total cumulative abnormal returns, CAR_gesamt_positive which refers only to positive cumulative abnormal returns and CAR_gesamt_negativ which contains only negative cumulative abnormal returns.

Thus, I get the following three outputs using the aforementioned command, just changing the dependent variables:

Code:

reg CAR_gesamt Regelpublizität Dividenden Aktienrückkauf Anleihe Kapitalerhöhung MA Kooperation Beteiligung Verkauf Personalveränderung Rechtssache Strategie Auftragseingang Insolvenz Sonstiges Zeit MarketCapln TotalAssetsln GeneralStandard Scale ReturnonAssets if n_disclosurecount_tag==1, vce(robust)

Code:

Linear regression                               Number of obs     =      1,910
                                                F(20, 1888)       =          .
                                                Prob > F          =          .
                                                R-squared         =     0.0203
                                                Root MSE          =     .22202

-------------------------------------------------------------------------------------
                    |               Robust
         CAR_gesamt |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
--------------------+----------------------------------------------------------------
    Regelpublizität |   .0346433   .0098256     3.53   0.000     .0153732    .0539134
         Dividenden |    .036643   .0118994     3.08   0.002     .0133057    .0599803
     Aktienrückkauf |   .0418391   .0096198     4.35   0.000     .0229725    .0607057
            Anleihe |   .0343144   .0280485     1.22   0.221     -.020695    .0893237
    Kapitalerhöhung |    .017955   .0135757     1.32   0.186    -.0086699    .0445799
                 MA |   .0481007   .0121212     3.97   0.000     .0243284    .0718729
        Kooperation |   .1034536   .0270035     3.83   0.000     .0504938    .1564135
        Beteiligung |   .0744864   .0144003     5.17   0.000     .0462443    .1027285
            Verkauf |   .0479639   .0281547     1.70   0.089    -.0072537    .1031815
Personalveränderung |   .0155133   .0125134     1.24   0.215    -.0090282    .0400548
        Rechtssache |   .0090666   .0210047     0.43   0.666    -.0321283    .0502615
          Strategie |   .0370645   .0578847     0.64   0.522    -.0764601    .1505892
    Auftragseingang |   .0854783    .020662     4.14   0.000     .0449555    .1260011
          Insolvenz |   .0939476   .1720563     0.55   0.585     -.243493    .4313881
          Sonstiges |  -.0064238   .0209867    -0.31   0.760    -.0475833    .0347357
               Zeit |  -1.60e-14   9.60e-15    -1.67   0.096    -3.48e-14    2.82e-15
        MarketCapln |   .0152107   .0123968     1.23   0.220    -.0091021    .0395235
      TotalAssetsln |  -.0091281   .0087437    -1.04   0.297    -.0262765    .0080203
    GeneralStandard |   .0146988   .0150551     0.98   0.329    -.0148275    .0442252
              Scale |    .024613   .0160091     1.54   0.124    -.0067843    .0560103
     ReturnonAssets |  -.0006545   .0006804    -0.96   0.336    -.0019889    .0006799
              _cons |  -.1459188   .0657184    -2.22   0.027    -.2748072   -.0170304
-------------------------------------------------------------------------------------

Code:

 reg CAR_gesamt_positiv Regelpublizität Dividenden Aktienrückkauf Anleihe Kapitalerhöhung MA Kooperation Beteiligung Verkauf Personalveränderung Rechtssache Strategie Auftragseingang Insolvenz Sonstiges Zeit MarketCapln TotalAssetsln GeneralStandard Scale ReturnonAssets if n_disclosurecount_tag==1, vce(robust)

Code:

Linear regression                               Number of obs     =        909
                                                F(20, 887)        =          .
                                                Prob > F          =          .
                                                R-squared         =     0.1618
                                                Root MSE          =      .1708

-------------------------------------------------------------------------------------
                    |               Robust
 CAR_gesamt_positiv |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
--------------------+----------------------------------------------------------------
    Regelpublizität |   .0133361   .0108227     1.23   0.218     -.007905    .0345772
         Dividenden |   .0007233   .0160668     0.05   0.964      -.03081    .0322566
     Aktienrückkauf |   .0018905   .0114669     0.16   0.869     -.020615    .0243959
            Anleihe |   .0294088   .0314745     0.93   0.350    -.0323645    .0911821
    Kapitalerhöhung |   .0023494   .0157071     0.15   0.881    -.0284781    .0331769
                 MA |   .0325903   .0116005     2.81   0.005     .0098227     .055358
        Kooperation |   .0202378   .0286863     0.71   0.481    -.0360632    .0765388
        Beteiligung |   .0128323   .0119777     1.07   0.284    -.0106755    .0363402
            Verkauf |   .0530642   .0366234     1.45   0.148    -.0188144    .1249428
Personalveränderung |   .0004881   .0117124     0.04   0.967    -.0224992    .0234754
        Rechtssache |   .0228309   .0234322     0.97   0.330    -.0231581    .0688199
          Strategie |   .0690418   .0407355     1.69   0.090    -.0109073     .148991
    Auftragseingang |   .0330857   .0200519     1.65   0.099     -.006269    .0724404
          Insolvenz |   .3863657    .216513     1.78   0.075    -.0385718    .8113033
          Sonstiges |   .0358996    .027408     1.31   0.191    -.0178924    .0896917
               Zeit |  -3.79e-11   5.62e-11    -0.67   0.500    -1.48e-10    7.24e-11
        MarketCapln |  -.0064207   .0139103    -0.46   0.644    -.0337218    .0208803
      TotalAssetsln |  -.0051993   .0106461    -0.49   0.625    -.0260939    .0156952
    GeneralStandard |   .0351406   .0178779     1.97   0.050     .0000527    .0702285
              Scale |  -.0178708    .013933    -1.28   0.200    -.0452163    .0094747
     ReturnonAssets |  -.0010079   .0005695    -1.77   0.077    -.0021256    .0001098
              _cons |  -71.58165   106.4161    -0.67   0.501    -280.4384    137.2751
-------------------------------------------------------------------------------------

Code:

   reg CAR_gesamt_negativ Regelpublizität Dividenden Aktienrückkauf Anleihe Kapitalerhöhung MA Kooperation Beteiligung Verkauf Personalveränderung Rechtssache Strategie Auftragseingang Insolvenz Sonstiges Zeit MarketCapln TotalAssetsln GeneralStandard Scale ReturnonAssets if n_disclosurecount_tag==1, vce(robust)

Code:


Linear regression                               Number of obs     =      1,001
                                                F(20, 979)        =          .
                                                Prob > F          =          .
                                                R-squared         =     0.1381
                                                Root MSE          =     .20264

-------------------------------------------------------------------------------------
                    |               Robust
 CAR_gesamt_negativ |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
--------------------+----------------------------------------------------------------
    Regelpublizität |   .0407928   .0106433     3.83   0.000     .0199066    .0616791
         Dividenden |    .050257   .0135227     3.72   0.000     .0237201    .0767939
     Aktienrückkauf |   .0517575   .0125206     4.13   0.000     .0271873    .0763277
            Anleihe |   .0068625   .0353374     0.19   0.846    -.0624833    .0762083
    Kapitalerhöhung |   .0247777   .0145614     1.70   0.089    -.0037974    .0533529
                 MA |   .0102019   .0146016     0.70   0.485     -.018452    .0388559
        Kooperation |   .0775656   .0283728     2.73   0.006     .0218872     .133244
        Beteiligung |   .0438398   .0112344     3.90   0.000     .0217936     .065886
            Verkauf |  -.0090703   .0279294    -0.32   0.745    -.0638786     .045738
Personalveränderung |   .0177453   .0150552     1.18   0.239    -.0117989    .0472895
        Rechtssache |   .0174292   .0214035     0.81   0.416    -.0245728    .0594313
          Strategie |   -.084871   .1113765    -0.76   0.446    -.3034351    .1336932
    Auftragseingang |   .0419292   .0154199     2.72   0.007     .0116692    .0721891
          Insolvenz |  -.3709435   .1349949    -2.75   0.006     -.635856   -.1060309
          Sonstiges |  -.0411137   .0207829    -1.98   0.048    -.0818979   -.0003295
               Zeit |   2.61e-14   6.63e-15     3.94   0.000     1.31e-14    3.91e-14
        MarketCapln |   .0193138    .015125     1.28   0.202    -.0103673    .0489948
      TotalAssetsln |  -.0004801   .0091585    -0.05   0.958    -.0184526    .0174924
    GeneralStandard |    -.01767   .0146548    -1.21   0.228    -.0464284    .0110884
              Scale |   .0383542   .0234047     1.64   0.102    -.0075749    .0842834
     ReturnonAssets |  -.0004225   .0009551    -0.44   0.658    -.0022968    .0014518
              _cons |  -.2934315   .0946442    -3.10   0.002    -.4791603   -.1077027
-------------------------------------------------------------------------------------

Just in case it helps I also give you the command and output on "normal" Regression of CAR_gesamt:

Code:

 reg CAR_gesamt Regelpublizität Dividenden Aktienrückkauf Anleihe Kapitalerhöhung MA Kooperation Beteiligung Verkauf Personalveränderung Rechtssache Strategie Auftragseingang Insolvenz Sonstiges Zeit MarketCapln TotalAssetsln GeneralStandard Scale ReturnonAssets if n_disclosurecount_tag==1

Code:

      Source |       SS           df       MS      Number of obs   =     1,910
-------------+----------------------------------   F(21, 1888)     =      1.86
       Model |   1.9296357        21  .091887414   Prob > F        =    0.0100
    Residual |   93.062259     1,888  .049291451   R-squared       =    0.0203
-------------+----------------------------------   Adj R-squared   =    0.0094
       Total |  94.9918947     1,909  .049760029   Root MSE        =    .22202

-------------------------------------------------------------------------------------
         CAR_gesamt |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
--------------------+----------------------------------------------------------------
    Regelpublizität |   .0346433   .0190342     1.82   0.069    -.0026869    .0719736
         Dividenden |    .036643   .0327038     1.12   0.263    -.0274964    .1007824
     Aktienrückkauf |   .0418391   .0293057     1.43   0.154    -.0156358     .099314
            Anleihe |   .0343144    .028042     1.22   0.221    -.0206823     .089311
    Kapitalerhöhung |    .017955   .0213754     0.84   0.401    -.0239669    .0598769
                 MA |   .0481007   .0209343     2.30   0.022     .0070438    .0891575
        Kooperation |   .1034536   .0402453     2.57   0.010     .0245237    .1823836
        Beteiligung |   .0744864   .0364062     2.05   0.041     .0030858     .145887
            Verkauf |   .0479639   .0273432     1.75   0.080    -.0056621    .1015899
Personalveränderung |   .0155133   .0193277     0.80   0.422    -.0223926    .0534193
        Rechtssache |   .0090666   .0340207     0.27   0.790    -.0576554    .0757887
          Strategie |   .0370645   .0264856     1.40   0.162    -.0148795    .0890086
    Auftragseingang |   .0854783    .038209     2.24   0.025     .0105421    .1604145
          Insolvenz |   .0939476    .047098     1.99   0.046      .001578    .1863172
          Sonstiges |  -.0064238   .0231799    -0.28   0.782    -.0518848    .0390371
               Zeit |  -1.60e-14   4.25e-14    -0.38   0.706    -9.93e-14    6.73e-14
    GeneralStandard |   .0146988   .0146801     1.00   0.317     -.014092    .0434897
              Scale |    .024613   .0208717     1.18   0.238    -.0163211    .0655471
        MarketCapln |   .0152107   .0046583     3.27   0.001     .0060747    .0243467
      TotalAssetsln |  -.0091281   .0040891    -2.23   0.026    -.0171476   -.0011085
     ReturnonAssets |  -.0006545   .0002769    -2.36   0.018    -.0011975   -.0001115
              _cons |  -.1459188   .0882021    -1.65   0.098    -.3189026     .027065
-------------------------------------------------------------------------------------

As you can see, the F-statistic (and adjusted R²) are reported conducting "normal" linear Regression (the same holds for CAR_gesamt_positiv and CAR_gesamt_negativ), but not for robust SE regression.

Maybe you need information on the dummy variables as well:

Code:

tabstat Regelpublizität Dividenden Aktienrückkauf Anleihe Kapitalerhöhung MA Kooperation Beteiligung Verkauf Personalveränderung Rechtssache Strategie Auftragseingang Insolvenz Sonstiges GeneralStandard Scale if n_disclosurecount_tag==1, s(sum) col(stat)

Code:

    variable |       sum
-------------+----------
Regelpubli~t |       270
  Dividenden |        56
Aktienrück~f |        73
     Anleihe |        83
Kapitalerh~g |       183
          MA |       196
 Kooperation |        35
 Beteiligung |        43
     Verkauf |        88
Personalve~g |       254
 Rechtssache |        51
   Strategie |        95
Auftragsei~g |        39
   Insolvenz |        28
   Sonstiges |       139
GeneralSta~d |       432
       Scale |       141

Since none of them equals 1 I assume that there is no singleton dummy variable included in the model.

Please inform me if you need any further information

.

Thanks a lot for your help!!

Last edited by Cecilia Roesen; 14 Dec 2018, 08:11.

Comment

Cecilia Roesen

Join Date: Nov 2018

Posts: 12
#4

14 Dec 2018, 08:20

Maybe I should add the fact, that from a theoretical point of view the results for the p-values of the robust SE Regression are a lot more plausible than the results of the normal linear regression. So it would be really great if there was a way to report F-stat and R² adj. as well!!
Comment
William Lisowski

Join Date: Dec 2014

Posts: 10150
#5

14 Dec 2018, 08:34

Your Zeit variable is strange. Apparently (given the coefficient estimates) it takes very large values. Can you tell us more about it?

Code:

describe Zeit codebook Zeit summarize Zeit, detail

If you were to tell us more about Zeit, and post the results of running these commands, someone might be able to suggest a better way of treating Zeit,

In particular, I wonder if the vce(robust) code is finding Zeit difficult to deal with. You might try leaving Zeit out of one of your vce(robust) models and seeing if that results in the F-statistic returning to the model results.
Comment

Cecilia Roesen

Join Date: Nov 2018
Posts: 12

14 Dec 2018, 08:59

It's funny, I also thought that "Zeit" might be the problem. Zeit stands for "time" and basically represents the time when the ad hoc message was published. Therefore it is formatted as %tcHH:MM:SS.

Code:


. describe Zeit

              storage   display    value
variable name   type    format     label      variable label
---------------------------------------------------------------------------------------------------------
Zeit            double  %tc..                 Zeit



codebook Zeit

-----------------------------------------------------------------------------------------------------------------------------
Zeit                                                                                                                     Zeit
-----------------------------------------------------------------------------------------------------------------------------

                  type:  numeric (double)

                 range:  [-1.893e+12,1.842e+12]       units:  10000
         unique values:  845                      missing .:  0/21,010

                  mean:  -1.9e+12
              std. dev:   1.2e+11

           percentiles:        10%       25%       50%       75%       90%
                          -1.9e+12  -1.9e+12  -1.9e+12  -1.9e+12  -1.9e+12



 summarize Zeit, detail

                            Zeit
-------------------------------------------------------------
      Percentiles      Smallest
 1%    -1.89e+12      -1.89e+12
 5%    -1.89e+12      -1.89e+12
10%    -1.89e+12      -1.89e+12       Obs              21,010
25%    -1.89e+12      -1.89e+12       Sum of Wgt.      21,010

50%    -1.89e+12                      Mean          -1.89e+12
                        Largest       Std. Dev.      1.21e+11
75%    -1.89e+12       1.84e+12
90%    -1.89e+12       1.84e+12       Variance       1.46e+22
95%    -1.89e+12       1.84e+12       Skewness       30.85452
99%    -1.89e+12       1.84e+12       Kurtosis       953.0015

Please note that the number of observations is misrepresented.
Nonetheless the data is arranged the way that mean/sd etc. are not affected.

It becomes clear if I type in the "correct" command for summarize, so please don't worry about the number of observations:

Code:

. summarize Zeit if n_disclosurecount_tag==1, detail

                            Zeit
-------------------------------------------------------------
      Percentiles      Smallest
 1%    -1.89e+12      -1.89e+12
 5%    -1.89e+12      -1.89e+12
10%    -1.89e+12      -1.89e+12       Obs               1,910
25%    -1.89e+12      -1.89e+12       Sum of Wgt.       1,910

50%    -1.89e+12                      Mean          -1.89e+12
                        Largest       Std. Dev.      1.21e+11
75%    -1.89e+12      -1.89e+12
90%    -1.89e+12      -1.89e+12       Variance       1.46e+22
95%    -1.89e+12       1.84e+12       Skewness       30.85452
99%    -1.89e+12       1.84e+12       Kurtosis       953.0015

I simply wonder if it is even "allowed" to use time variables in regression models. I thought of including dummy variables for certain time periods but then I got multicollinearity problems so this is not really an option, either. Is it may be possible to adjust the format? I don't really know how Stata works with respect to time variables, but if you, for example, define a time variable in Microsoft Excel and change the format to "default" you get numerical values that are not too high. Stata seems to covert time variables into rather high values… so may be it could work if the format is adjusted?

Thanks for everything!

Comment

Cecilia Roesen

Join Date: Nov 2018

Posts: 12
#7

14 Dec 2018, 09:03

Sorry I forgot to say that I tried exactly what you proposed (leave out time and run the regression again) and then I got results for F and R² adj. That's why I originally thought that Time might be the problem But I cannot just leave the variable out because it is crucial to my analysis...
Comment

Cecilia Roesen

Join Date: Nov 2018
Posts: 12

14 Dec 2018, 09:08

In case it is needed: Here's the output for the robust regression without the time variable:

Code:

Linear regression                               Number of obs     =      1,910
                                                F(20, 1889)       =       4.11
                                                Prob > F          =     0.0000
                                                R-squared         =     0.0202
                                                Root MSE          =     .22197

-------------------------------------------------------------------------------------
                    |               Robust
         CAR_gesamt |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
--------------------+----------------------------------------------------------------
    Regelpublizität |   .0346615    .009821     3.53   0.000     .0154003    .0539226
         Dividenden |   .0355877   .0118186     3.01   0.003     .0124089    .0587665
     Aktienrückkauf |    .041782   .0096126     4.35   0.000     .0229295    .0606345
            Anleihe |   .0343009   .0280414     1.22   0.221    -.0206945    .0892963
    Kapitalerhöhung |    .018033   .0135665     1.33   0.184    -.0085739    .0446399
                 MA |   .0480606   .0121178     3.97   0.000     .0242949    .0718263
        Kooperation |   .1035219   .0270019     3.83   0.000     .0505652    .1564786
        Beteiligung |   .0744821   .0143985     5.17   0.000     .0462436    .1027207
            Verkauf |   .0479677   .0281469     1.70   0.089    -.0072345      .10317
Personalveränderung |   .0155206   .0125071     1.24   0.215    -.0090085    .0400497
        Rechtssache |   .0090357   .0209919     0.43   0.667     -.032134    .0502054
          Strategie |   .0370767   .0578729     0.64   0.522    -.0764248    .1505781
    Auftragseingang |   .0855951   .0206544     4.14   0.000     .0450872    .1261029
          Insolvenz |   .0939014   .1720241     0.55   0.585    -.2434759    .4312786
          Sonstiges |   -.006833   .0209348    -0.33   0.744    -.0478908    .0342249
        MarketCapln |   .0151355   .0123817     1.22   0.222    -.0091477    .0394187
      TotalAssetsln |   -.009038    .008723    -1.04   0.300    -.0261459    .0080698
    GeneralStandard |    .014412   .0149802     0.96   0.336    -.0149674    .0437915
              Scale |   .0246499    .016006     1.54   0.124    -.0067414    .0560412
     ReturnonAssets |  -.0006537     .00068    -0.96   0.337    -.0019874      .00068
              _cons |  -.1157676   .0619208    -1.87   0.062     -.237208    .0056728
-------------------------------------------------------------------------------------

.

Comment

Rich Goldstein

Join Date: Mar 2014

Posts: 4466
#9

14 Dec 2018, 09:11

your time variable is very strange; first, note that there is very little variation; second, note the large jump between the 3rd largest and the 2nd largest values; finally, recall that this measures millisecpnds since 1/1/1960 - I think that there is something seriously wrong with this variable - maybe you could explain more about what it should be and why there is so little variation (and, of course, how you got it into this format - i.e., what commands did you use?)
Comment

Cecilia Roesen

Join Date: Nov 2018
Posts: 12

#10

14 Dec 2018, 09:44

Yes, you are right, indeed. I have to admit that I did the data recollection in Exel, so I basically just typed in 15:02:00 for example. Later on I imported it into Stata and I did not change the format because "HH:MM:SS" looked quite plausible to me.

What's really strange is that if I type in -br- I see the time in the format I expect it to be, but if I generate a -dataex- example only really high values appear.

Below you can see, for example three ad hoc messages of the company 1&1 Drillisch that were published at different times:

Code:

* Example generated by -dataex-. To install: ssc install dataex
clear
input int(DisclosureID Day) double Zeit str36 CompanyName
 911 101      -1893378780000 "1&1 DRILLISCH"
 911 102      -1893378780000 "1&1 DRILLISCH"
 911 103      -1893378780000 "1&1 DRILLISCH"
 911 104      -1893378780000 "1&1 DRILLISCH"
 911 105      -1893378780000 "1&1 DRILLISCH"
 911 106      -1893378780000 "1&1 DRILLISCH"
 911 107      -1893378780000 "1&1 DRILLISCH"
 911 108      -1893378780000 "1&1 DRILLISCH"
 911 109      -1893378780000 "1&1 DRILLISCH"
 911 110      -1893378780000 "1&1 DRILLISCH"
 911 111      -1893378780000 "1&1 DRILLISCH"
2345 101      -1893410520000 "1&1 DRILLISCH"
2345 102      -1893410520000 "1&1 DRILLISCH"
2345 103      -1893410520000 "1&1 DRILLISCH"
2345 104      -1893410520000 "1&1 DRILLISCH"
2345 105      -1893410520000 "1&1 DRILLISCH"
2345 106      -1893410520000 "1&1 DRILLISCH"
2345 107      -1893410520000 "1&1 DRILLISCH"
2345 108      -1893410520000 "1&1 DRILLISCH"
2345 109      -1893410520000 "1&1 DRILLISCH"
2345 110      -1893410520000 "1&1 DRILLISCH"
2345 111      -1893410520000 "1&1 DRILLISCH"
2798 101 -1893438300000.0002 "1&1 DRILLISCH"
2798 102 -1893438300000.0002 "1&1 DRILLISCH"
2798 103 -1893438300000.0002 "1&1 DRILLISCH"
2798 104 -1893438300000.0002 "1&1 DRILLISCH"
2798 105 -1893438300000.0002 "1&1 DRILLISCH"
2798 106 -1893438300000.0002 "1&1 DRILLISCH"
2798 107 -1893438300000.0002 "1&1 DRILLISCH"
2798 108 -1893438300000.0002 "1&1 DRILLISCH"
2798 109 -1893438300000.0002 "1&1 DRILLISCH"
2798 110 -1893438300000.0002 "1&1 DRILLISCH"
2798 111 -1893438300000.0002 "1&1 DRILLISCH"
end
format %tcHH:MM:SS Zeit

And here you can see a screenshot of my data editor:

Click image for larger version

Name: scrren.PNG
Views: 1
Size: 39.5 KB
ID: 1474918

So I don't really understand why the descriptive variables show such low standard deviations…
How do you see that it measures miliseconds since 1960? The thing is that it should not have anything to do with a date but only with the time the ad hoc message was published...

Do you have an idea what the appropiate time variable format should be so that it is usable in a regression?

Comment

William Lisowski

Join Date: Dec 2014

Posts: 10150
#11

14 Dec 2018, 09:48

First of all, some substantive advice, as opposed to advice on using Statalist.

Stata's "date and time" variables are complicated and there is a lot to learn. If you have not already read the very detailed Chapter 24 (Working with dates and times) of the Stata User's Guide PDF, do so now. If you have, it's time for a refresher. After that, the help datetime documentation will usually be enough to point the way. You can't remember everything; even the most experienced users end up referring to the help datetime documentation or back to the manual for details. But at least you will get a good understanding of the basics and the underlying principles. An investment of time that will be amply repaid.

All Stata manuals are included as PDFs in the Stata installation (since version 11) and are accessible from within Stata - for example, through the PDF Documentation section of Stata's Help menu.

Added in edit: this was being written while post #10 was posted. You definitely need to do the reading.

You seem to suggest that your time data was imported from Excel. Suppose you issue the command

Code:

format Zeit %tc browse Zeit

what sorts of dates and times do you see? I may be wrong, but I don't think they're what you think they should be. Values on the order of -1.89e+12 correspond to a lot of milliseconds before January 1 1960.

Code:

. display %tc -1.89e12 09feb1900 00:00:00

I think that however you got your data in from Excel (did you use Stata's import excel command?) your date was not imported correctly. If that's the case, we'll have to figure out what to correct once we know what you did that failed. There's a FAQ somewhere on this, and other advice, but I can't quickly locate it.

But more to the point, do you really believe that time makes a linear contribution to CAR? I don't think so, because you say that you tried to use dummy variables for certain time periods but ran into trouble. You were on the right track there, you are (almost certainly) on the wrong track trying to use time - either as it is, or re-expressed as a date or year - as a linear term in your model. Multicollinearity is not a problem with dummy variables if you do them correctly.

So your work is cut out for you.
Read about dates and times in Stata

Decide if your times are correct in Stata

If not, they need to be corrected

Figure out how to create and incorporate indicator variables for time periods in your regressions

With regard to #4, your approach should be to create a categorical variable that indicates different time periods, e.g 1 for 2000 and before, 2 for 2000-2008, 3 for 2009 onward, and then use Stata's factor variable notation to include indicator variables in your regressions. For that, you will want to review the output of help factor variables for instruction in that technique.

Last edited by William Lisowski; 14 Dec 2018, 09:50.
Comment
Cecilia Roesen

Join Date: Nov 2018

Posts: 12
#12

16 Dec 2018, 13:43

Thanks a lot for your advice, I really appreciate it! I'll definitely do the reading! Having read the information on -help datetime- just makes me wonder if it implies that time is always linked to a date? That is not really what I want. I only want to refer to certain time periods like for example 9:00-17:00pm, 17:01-22:00pm, 22:01-7:00am and 7:01-8:59am, independent of the date the ad hoc message was published. Previous studies, for example, have come to the conclusion, that if ad hoc messages are not published during trading hours the overall reaction is lower on the actual event day in comparison to messages published during trading hours (something I could confirm via univariate analysis as well). So I actually need to somehow include time variables in my regression model, testing for interactions with other relevant variables... Well, if I still can make out how it works I guess creating factor variables is a good idea- I have done that before, this shouldn't be a problem.
Comment

Announcement