OLS and fixed effects

[Larissa MM]

Hi,

Using OLS, I have panel data of 480 firms (10years).

reg lev prof size tang growth liq i.lifecycle i.industry i.year

lifecycle is a dummy variable created after categorization each of the firm year of the sample into different life cycle stages. A firm can have different life cycle stages in different year.

--
As it is a panel data, I will use fixed effects.

xtreg lev prof size tang growth liq, fe

Is this right? The lifecycle dummy is important as I need to interprete it. But how can I include the life cycle dummy in my fixed effect model?

[Carlo Lazzaro]

Larissa:
if -lifecycle- is time-varying, the -xtreg, fe-machinery will not wipe it out (unlike -industry-, I guess).
In general, you should test via -hausman- which specification (-fe- or -re-) fits your data better.
As an aside, -regress- is (usually) not the first choice when you deal with panel data with a continuous regressand.

[Larissa MM]

Larissa:
if -lifecycle- is time-varying, the -xtreg, fe-machinery will not wipe it out (unlike -industry-, I guess).
In general, you should test via -hausman- which specification (-fe- or -re-) fits your data better.
As an aside, -regress- is (usually) not the first choice when you deal with panel data with a continuous regressand.

Hi Carlo,

I have run Hausman, and FE fits better.

When I run, only industry is wiped out. (dc is lifecycle) I would like to know if the below looks right for what I am trying to run without FE?

xtreg mtd prof size tang growth liq dc1 dc2 dc3 dc4 i.industry i.year, fe
note: 9991.industry omitted because of collinearity
note: 9992.industry omitted because of collinearity
note: 9993.industry omitted because of collinearity
note: 9994.industry omitted because of collinearity
note: 9995.industry omitted because of collinearity
note: 9996.industry omitted because of collinearity
note: 9997.industry omitted because of collinearity
note: 9998.industry omitted because of collinearity
note: 9999.industry omitted because of collinearity

Fixed-effects (within) regression               Number of obs     =      4,820
Group variable: id                              Number of groups  =        482

R-sq:                                           Obs per group:
     within  = 0.2403                                         min =         10
     between = 0.1357                                         avg =       10.0
     overall = 0.1509                                         max =         10

                                                F(18,4320)        =      75.92
corr(u_i, Xb)  = -0.3107                        Prob > F          =     0.0000

------------------------------------------------------------------------------
         mtd |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
        prof |  -.2444728    .016683   -14.65   0.000      -.27718   -.2117656
        size |   .0943677   .0051354    18.38   0.000     .0842997    .1044357
        tang |   .1387465   .0162789     8.52   0.000     .1068316    .1706614
      growth |   -.004308   .0007211    -5.97   0.000    -.0057217   -.0028942
         liq |   -.003611    .000317   -11.39   0.000    -.0042325   -.0029895
         dc1 |   .0029403   .0062745     0.47   0.639     -.009361    .0152416
         dc2 |   .0040731    .004835     0.84   0.400    -.0054059    .0135521
         dc3 |   .0103336   .0051136     2.02   0.043     .0003084    .0203588
         dc4 |   .0184405   .0072045     2.56   0.011      .004316     .032565
             |
    industry |
       9991  |          0  (omitted)
       9992  |          0  (omitted)
       9993  |          0  (omitted)
       9994  |          0  (omitted)
       9995  |          0  (omitted)
       9996  |          0  (omitted)
       9997  |          0  (omitted)
       9998  |          0  (omitted)
       9999  |          0  (omitted)
             |
        year |
       2009  |  -.0644243   .0070173    -9.18   0.000    -.0781817   -.0506669
       2010  |  -.0886784   .0070299   -12.61   0.000    -.1024607   -.0748961
       2011  |  -.0840421   .0070341   -11.95   0.000    -.0978326   -.0702516
       2012  |  -.0879431   .0070839   -12.41   0.000    -.1018312    -.074055
       2013  |  -.1351823   .0071367   -18.94   0.000    -.1491738   -.1211908
       2014  |  -.1443554   .0072259   -19.98   0.000    -.1585219    -.130189
       2015  |  -.1545702   .0073154   -21.13   0.000    -.1689121   -.1402283
       2016  |  -.1549156   .0074161   -20.89   0.000    -.1694549   -.1403762
       2017  |  -.1658981   .0075199   -22.06   0.000     -.180641   -.1511553
             |
       _cons |  -.7250703   .0642256   -11.29   0.000    -.8509854   -.5991553
-------------+----------------------------------------------------------------
     sigma_u |  .19702641
     sigma_e |  .10866797
         rho |  .76675561   (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(481, 4320) = 22.89                  Prob > F = 0.0000

[Carlo Lazzaro]

Larissa:
your code looks correct to me.
As expected, -fe- machinery wipes out the -industry- due to its time-invariance.
The statistical significant F-test appearing at the foot of the -fe- outcome table tells you that -xtreg,fe- outperforms -regress-.

[Larissa MM]

Larissa:
your code looks correct to me.
As expected, -fe- machinery wipes out the -industry- due to its time-invariance.
The statistical significant F-test appearing at the foot of the -fe- outcome table tells you that -xtreg,fe- outperforms -regress-.

Thanks Carlo, I have one additional question but kinda not related to stata if you could help.

n this case, how do I present them in my thesis

xtreg mtd prof size tang growth liq dc1 dc2 dc3 dc4 i.industry i.year, fe

LEVit = B₀ + B1PROFit + B2SIZEit + B3TANGit + B4GROWTHit + B5LIQit + B6DC1it + B7DC2it + B8DC3it + B9DC4it + ɛjt


Should I just include the independent variables and the dummies or just the independent variables without lifecycle,industry and year

[Carlo Lazzaro]

Larissa:
you should include exactly what you typed in Stata, as yiur results refer to that code (hence, lifecycle, industry and year shoud be included).
Obviously, check what above with your supervisor.