Fixed effect xtreg - Panel data - P-value

[Jack Smith]

Hey everyone,

I'm trying to find the regression of my panel dataset. I'm trying to calculate the relationship between Acquisitions (DV) and CEO stock options (IV). I also have some control variables such as CEO age, CEO tenure, firm performance ROA, etc.
From what I saw on many articles, they use xtreg fe to find the regression. My problem is that I find a really high p-value when performing this. I'm not sure whether I'm doing everything correctly.
Could you please check out my code and whether I could do something to reduce this p-value. Also, are the control variables supposed to be added after the independent variable?

Thank you!

. xtreg Acquisitions Valueofstockoptionawards CEOage Marketvalueasofeveryfiscal FirmperformanceROA Stockownership CEOtenure, fe

Fixed-effects (within) regression               Number of obs     =        614
Group variable: Company1                        Number of groups  =        155

R-sq:                                           Obs per group:
     within  = 0.0258                                         min =          1
     between = 0.0711                                         avg =        4.0
     overall = 0.0572                                         max =          5

                                                F(6,453)          =       2.00
corr(u_i, Xb)  = -0.0956                        Prob > F          =     0.0646

--------------------------------------------------------------------------------------------
              Acquisitions |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
---------------------------+----------------------------------------------------------------
  Valueofstockoptionawards |   8.77e-09   3.54e-08     0.25   0.804    -6.08e-08    7.83e-08
                    CEOage |  -.0125978   .0185358    -0.68   0.497    -.0490246    .0238291
Marketvalueasofeveryfiscal |   1.36e-11   5.36e-12     2.54   0.012     3.06e-12    2.41e-11
        FirmperformanceROA |   .0033968   .0101177     0.34   0.737    -.0164867    .0232803
            Stockownership |   2.52e-09   1.14e-09     2.22   0.027     2.92e-10    4.75e-09
                 CEOtenure |  -.0036318   .0165066    -0.22   0.826    -.0360708    .0288072
                     _cons |   1.764198    .994452     1.77   0.077    -.1901134     3.71851
---------------------------+----------------------------------------------------------------
                   sigma_u |  1.5235763
                   sigma_e |  1.1794857
                       rho |  .62526671   (fraction of variance due to u_i)
--------------------------------------------------------------------------------------------
F test that all u_i=0: F(154, 453) = 6.58                    Prob > F = 0.0000

[Carlo Lazzaro]

Jack:
welcome to this forum.
I would check whether:
1) your model is correctly specified given the data generating process (ie, are all the relevant predictors included?);
2) your model suffers from quasi-extreme multicollinearity (as per the width of your 95% CI);
3) if -fe- is actually the way to go or -re- is better for your data (see -help hausman-).

As an aside, please note that panel data are regressions are pretty tricky and the -click and see- approach is rarely rewarding (and should be discouraged altogether): hence, studying any decent panel data econometrics textbook is highly recommended.

[Phil Bromiley]

To add to Carlo's comments, almost certainly, your dv is not continuous - it is probably a count or a dummy. If so, you should consider estimators built for count (e.g., xtpoisson, xtnbreg) or dummy (xtlogit, xtprobit) dependent variables.

Whether the right variable is awards or total value of exercisable options or some other option-related metric is not clear. I'm not sure what discipline you come from, but in strategic management there has been a lot of work on how these differences matter. You also should be careful that the awards are not in the year of the acquisitions since acquisitions in t certainly can influence awards in t.

[Jack Smith]

Ok thank you both!

Would it make sense to standardise (with standard deviation and mean of each variable) all the variables before doing the regression?

[Carlo Lazzaro]

Jack:
as we do not know which type of variables are you dealing with, I cannot advise on that.
Things would have been probably easier had you shared an excerpt/example of your data via -dataex-. Thanks.

[Jack Smith]

Ah ok my bad, here is an example of the variables:

* Example generated by -dataex-. To install: ssc install dataex
clear
Acquisitions double(Valueofstockoptionawards Marketvalueasofeveryfiscal FirmperformanceROA Stockownership) byte CEOage double CEOtenure
1                  0          814340000  14.34             589395 66  18
2                  0          891110000   6.81             565847 67  19
1                  0         1280720000  11.69             576681 68  20
1                  0         1264880000  11.87             572781 69  21
1                  0         1189170000   11.5             584041 70  22
2             357638        53995560000  16.19             386328 59   2
3            1563162        84013310000   7.24 482146.99999999994 60   3
2            2369580        1.04275e+11  13.92             382471 61   4
5 3212968.9999999995        96842380000  11.37             355012 62   5
1             786774        1.01764e+11   8.81             314905 63   6
1 503533.99999999994          522550000   9.27              60005 47  10
3             612505         1063340000   3.92             159111 48  11
1             861232         1563480000  45.61             218853 49  12
1             841798         3831240000  12.24             399252 50  13
3            2417748         4888170000  11.83             354760 51  14
4            2206260        1.03534e+11   4.97            1456713 58  15
3 3651733.9999999995  59265230000.00001   5.91            1514972 59  16
3            4589739  67790630000.00001  11.72            1507693 60  17
2            5987405        66993060000   3.75            3193255 61  18
2            2317123  56551380000.00001   1.71 3210544.9999999995 62  19
2            2581809  998320000.0000001   3.71             478529 57  19
2            1724090         1205220000      3             495400 58  20
1            1835586 1714409999.9999998   1.94 400549.99999999994 59  21
2 1808955.9999999998 1846190000.0000002  -1.95             534588 60  22
1            3981293          866950000 -16.66             564285 61  23
3                  0        15490690000   4.69             620953 56   4
3            3727440        24843290000   4.41             729194 57   5
1            6030413        31072730000   4.94             844089 58   6
1            8745183        37701440000   4.34  925317.0000000001 59   7
3            2897933        43515080000   3.53             795237 60   8
4             155356          189050000   2.77             509086 61  22
2             247239          307130000   4.16             485551 62  23
6             267367          277170000   5.26             508751 63  24
3             302531          375540000   3.81 523196.99999999994 64  25
2            5161551 457460000.00000006   3.66             501199 65  26
1             572523          224090000  -1.54             168417 59   3
1             422983          616330000    .06             266297 60   4
4                  0        10987080000  -3.24             399995 52   7
4                  0 29240040000.000004  -3.63             402865 53   8
7            2133833        68229250000   4.98              65759 45   2
9                  0        1.23164e+11  11.97              99912 46   3
4             210648        7.87705e+10  -2.77             180931 47   4
2  53191.99999999999         1429210000   6.37              87828 51  11
1                  0         2274470000   8.34              85051 52  12
2             255004         1726220000   8.42              79303 53  13
2             446253 1647959999.9999998   7.14 116477.99999999999 54  14
3            2040907         1275020000  14.65                  0 46   4
3            1135482         2377530000   5.69            2161235 47   5
1             797536 3903480000.0000005   9.37            2149971 48   6
1             797536         4269490000  10.93            2073274 49   7
1            1876428          2.734e+09    .19            2105172 50   8
1  850198.9999999999         2258420000   8.29             216526 57  12
1             311082         4589410000  16.42             216968 58  13
1                  0          4.489e+09  13.93 13270.000000000002 58   1
1                  0         5234430000  15.92              30206 59   2
2             649627  7660539999.999999  15.14  51710.99999999999 60   3
3             109432          9.088e+08   5.51              38567 55   7
1             253572         1100860000    5.8  63100.99999999999 56   8
1             470101         1046910000   5.45              59138 57   9
2             529047 1024979999.9999999   1.95              56321 58  10
1                  0         1033210000 -12.85              11742 70   2
1            6175917        18212310000   8.64             721176 55  22
2            6505436        26055140000  17.61  853794.0000000001 56  23
2            7140022        36689020000   2.03 114209.99999999999 49   1
1            4906068        42980550000    3.5                  0 50   2
1            2220201        2.73979e+10   3.78              33337 54   1
1            4970660         1.1905e+09  -7.15             175015 57   6
2                  0          351490000 -14.65             371521 55  32
1                  0          4.749e+08   3.07             154347 57   1

[Carlo Lazzaro]

Jack:
as Phil skillfully surmised, you should go -xtpoisson- (or -xtbnreg- if -xtpoisson- shows overdispersion), as your regressand takes on integer values.