How to normalize this Fixed effect model.

[Olalere Oluwaseyi]

Hello,

Please am confuse about the outcome of my regression results. Using return on equity as a dependent variable, after the Hausman test was in favour of Fixed Effect and it seems some of the coefficients of the Fixed Effect Model is more than 1 i.e CR2, MR, Size, GDP.
Now, i wouldn't know what to do as regards this issue. I really need an honest advice please.

 xtreg ROE CR1 CR2 LR OR MR SIZE GDP, fe

Fixed-effects (within) regression               Number of obs      =       112
Group variable: Bank                            Number of groups   =        16

R-sq:  within  = 0.1174                         Obs per group: min =         7
       between = 0.0368                                        avg =       7.0
       overall = 0.0063                                        max =         7

                                                F(7,89)            =      1.69
corr(u_i, Xb)  = -0.7678                        Prob > F           =    0.1213

------------------------------------------------------------------------------
         ROE |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         CR1 |  -.0865407   .0656812    -1.32   0.191    -.2170478    .0439664
         CR2 |   1.049958   .4802478     2.19   0.031     .0957156      2.0042
          LR |   .1999621   .1125024     1.78   0.079    -.0235778    .4235019
          OR |  -.2387946    .162367    -1.47   0.145    -.5614144    .0838253
          MR |  -2.491781   1.385326    -1.80   0.075    -5.244394    .2608312
        SIZE |   8.620314   7.450672     1.16   0.250    -6.184012    23.42464
         GDP |   2.642453   3.375826     0.78   0.436    -4.065241    9.350146
       _cons |  -1.245036   1.261884    -0.99   0.326    -3.752373    1.262302
-------------+----------------------------------------------------------------
     sigma_u |  .39923402
     sigma_e |  .48045632
         rho |  .40844989   (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0:     F(15, 89) =     1.16              Prob > F = 0.3172

[Phil Bromiley]

There is nothing wrong with parameters greater than one. Would be a problem with lagged dependent variables having parameters greater than one since they suggest the dv should increase without limit.

You might want to make sure you don't have any crazy ROE's. Since equity can be near zero, ROE can take on some extreme values for some firms.

You're estimating a lot of parameters with very little data. The F won't reject the hypothesis that all the structural parameters are zero and the F at the bottom won't reject the hypothesis that the fixed effects have parameters of zero even though you have a reasonable r-square within. This is likely to be due to a small sample size. If you only look at the F on the u_i, you could just go with OLS. This may help reduce the standard errors on your betas. However, I'd really work to get a larger sample if feasible.

[Olalere Oluwaseyi]

Thanks Phil. I will look into that.