Interpreting STATA FE regression output

[Deepika Deshpande]

Hello,

I am running a fe regression on panel data and had some basic questions on how to interpret the results:
1. R-sq within group is just 0.0306 seems to be extremely small. How should I be interpreting this number?
2. There are 2 F-values i,e, F(12,1099) with value 2.89 and F(122, 1099) with value 19.64 . My understanding is that the second one is a reflection of the quality of the overall model. Would that be a correct assumption? If so, what does the first F statistic represent?

(In case my questions are too basic my apologies. Please feel free to point me to any book material where I could find the answers. Thank you.

. xtreg ROA_win05 Ln_Revenue Ln_LTDTA CoAge TPSD wGDPpc wCPI wDCF wExpgr wGDPgr wCons c.l1.G
> SD##c.l1.GSD if FOREIGNSALESTOTALSALES >10 & Year_ < YearInactive, fe 

Fixed-effects (within) regression               Number of obs     =      1,234
Group variable: n_CUSIP                         Number of groups  =        123

R-sq:                                           Obs per group:
     within  = 0.0306                                         min =          1
     between = 0.4929                                         avg =       10.0
     overall = 0.3306                                         max =         18

                                                F(12,1099)        =       2.89
corr(u_i, Xb)  = 0.3905                         Prob > F          =     0.0006

-------------------------------------------------------------------------------
    ROA_win05 |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
--------------+----------------------------------------------------------------
   Ln_Revenue |   2.073987   .6697842     3.10   0.002     .7597872    3.388188
     Ln_LTDTA |  -.7476576   .1945061    -3.84   0.000    -1.129303   -.3660123
        CoAge |  -.0010398   .0901846    -0.01   0.991    -.1779934    .1759137
         TPSD |  -.9328759   .9147456    -1.02   0.308    -2.727721    .8619692
       wGDPpc |  -.0000301   .0001556    -0.19   0.847    -.0003354    .0002753
         wCPI |  -.2623995   .2668114    -0.98   0.326    -.7859167    .2611178
         wDCF |  -2.71e-14   1.20e-12    -0.02   0.982    -2.39e-12    2.33e-12
       wExpgr |   .1684812   .0990232     1.70   0.089    -.0258147    .3627772
       wGDPgr |  -.3395405   .2691567    -1.26   0.207    -.8676594    .1885785
        wCons |  -5.39e-14   4.31e-13    -0.13   0.900    -8.99e-13    7.91e-13
              |
          GSD |
          L1. |   7.670794   3.519331     2.18   0.029     .7654273    14.57616
              |
cL.GSD#cL.GSD |  -5.572459   2.683354    -2.08   0.038    -10.83754   -.3073827
              |
        _cons |  -39.89984   13.55392    -2.94   0.003    -66.49434   -13.30535
--------------+----------------------------------------------------------------
      sigma_u |  16.843882
      sigma_e |  7.0694025
          rho |  .85023217   (fraction of variance due to u_i)
-------------------------------------------------------------------------------
F test that all u_i=0: F(122, 1099) = 19.64                  Prob > F = 0.0000

[Felix Bittmann]

1. The within variation tells you how much variation within a given firm is explained by the change in the explanatory vars. As always, you should judge the size of this coefficient in relation to other models and studies. If they have similar vars and firms and much different results, this might be concerning. However, maybe this is a typical result? This must be judged from a theoretical perspective as well.
2. To judge the overall model fit, the first value is needed F(12, 1099). This is the omnibus test of the model. The second one refers to the "diagnostic" test (F test that all u_i=0.). For more annotated output see https://www.princeton.edu/~otorres/Panel101.pdf

[Deepika Deshpande]

Thank you Felix.

[Carlo Lazzaro]

Deepika:
as an aside to Felix's helpful advice, in my opinion 123 panels call for clustered standard errors.