[Lydia Gad]

Hi Chris,
I thank you for your effort.

I have run the regression but still the model omits the interaction term:

. xtreg I3631_Coverage_ratio_A i.Period##i.treat A1100_Loans_and_advances_A Total_Liabilities
> _A ROE_A  ROA_A  Debt_Securities_A  Total_Assets_A, fe
note: 1.Period omitted because of collinearity
note: 1.treat omitted because of collinearity
note: 1.Period#1.treat omitted because of collinearity

Fixed-effects (within) regression               Number of obs     =        439
Group variable: new_refere~a                    Number of groups  =         27

R-sq:                                           Obs per group:
     within  = 0.2018                                         min =          1
     between = 0.0105                                         avg =       16.3
     overall = 0.0095                                         max =         19

                                                F(6,406)          =      17.11
corr(u_i, Xb)  = -0.4651                        Prob > F          =     0.0000

--------------------------------------------------------------------------------------------
    I3631_Coverage_ratio_A |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
---------------------------+----------------------------------------------------------------
                  1.Period |          0  (omitted)
                   1.treat |          0  (omitted)
                           |
              Period#treat |
                      1 1  |          0  (omitted)
                           |
A1100_Loans_and_advances_A |  -1.52e-09   3.55e-09    -0.43   0.669    -8.51e-09    5.47e-09
       Total_Liabilities_A |   1.49e-08   3.75e-09     3.98   0.000     7.54e-09    2.23e-08
                     ROE_A |  -.6494135   .2313493    -2.81   0.005    -1.104205   -.1946216
                     ROA_A |    8.34393   2.401678     3.47   0.001     3.622653    13.06521
         Debt_Securities_A |  -2.16e-07   7.96e-07    -0.27   0.786    -1.78e-06    1.35e-06
            Total_Assets_A |  -1.32e-08   2.88e-09    -4.57   0.000    -1.88e-08   -7.50e-09
                     _cons |   44.55126   2.798249    15.92   0.000     39.05039    50.05213
---------------------------+----------------------------------------------------------------
                   sigma_u |  11.396829
                   sigma_e |  4.7123562
                       rho |  .85399623   (fraction of variance due to u_i)
--------------------------------------------------------------------------------------------
F test that all u_i=0: F(26, 406) = 58.27                    Prob > F = 0.0000

.

.
Hi Carlo,
Thank you so much for your advice to implement the -xtoverid-.
I really want to make sure that I am implementing the right model.
Now this is the output, which I think still leads me to Fixed Effects model:

. xtreg I3631_Coverage_ratio_A  A1100_Loans_and_advances_A Total_Liabilities_A ROE_A  ROA_A  
> Debt_Securities_A  Total_Assets_A, re

Random-effects GLS regression                   Number of obs     =        463
Group variable: new_refere~a                    Number of groups  =         27

R-sq:                                           Obs per group:
     within  = 0.1966                                         min =          1
     between = 0.0086                                         avg =       17.1
     overall = 0.0121                                         max =         20

                                                Wald chi2(6)      =      99.71
corr(u_i, X)   = 0 (assumed)                    Prob > chi2       =     0.0000

--------------------------------------------------------------------------------------------
    I3631_Coverage_ratio_A |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
---------------------------+----------------------------------------------------------------
A1100_Loans_and_advances_A |  -1.07e-09   2.74e-09    -0.39   0.698    -6.44e-09    4.31e-09
       Total_Liabilities_A |   1.39e-08   3.04e-09     4.56   0.000     7.90e-09    1.98e-08
                     ROE_A |  -.5834656    .206057    -2.83   0.005    -.9873298   -.1796013
                     ROA_A |   8.096589   2.200684     3.68   0.000     3.783328    12.40985
         Debt_Securities_A |   6.55e-09   7.37e-07     0.01   0.993    -1.44e-06    1.45e-06
            Total_Assets_A |  -1.15e-08   1.74e-09    -6.59   0.000    -1.49e-08   -8.07e-09
                     _cons |    42.8582   2.088609    20.52   0.000     38.76461     46.9518
---------------------------+----------------------------------------------------------------
                   sigma_u |  9.1938164
                   sigma_e |  4.7635099
                       rho |   .7883641   (fraction of variance due to u_i)
--------------------------------------------------------------------------------------------

. xtoverid

Test of overidentifying restrictions: fixed vs random effects
Cross-section time-series model: xtreg re   
Sargan-Hansen statistic  14.712  Chi-sq(6)    P-value = 0.0226

[Chris Boudreaux]

It is typical for the treatment variable to drop out in a fixed effects estimation because the treatment often does not vary within group (i.e., the reference area is either always treated or never treated). I'm not sure, however, why the period dummy and interaction drops out. I would check the code because the period dummy is also collinear. Maybe Carlo or someone else can be more helpful here.

[Carlo Lazzaro]

Lydia:
yes, you should go -fe-.
Chris' diagnosis on the cause of omitted interaction under -fe- specification is correct since, as expected, time-invariant predictors are wiped out.
In addition, I do share Chris' wise recomendation about revision your code, esepcially as far as the omitted interaction is concerned.

[Lydia Gad]

Dear Chris and Carlo,

Thank you so much for your advice and guidance.
I will revise my code.

Best regards,
Lydia Gad

[Lydia Gad]

Hi Carlo Lazzaro,

I run the hausman test and -xtoverid- test using different dependent variables; and with some dependent variables the p value is higher than 0.05 which means I need to go for RE.
However, in my scenario where I have panel dataset with countries, a FE is very advised to address the fact that country specific effects are related to the independent variables and I need to control them.

What do you advise?

Best regards,
Lydia Gad

[Carlo Lazzaro]

Lydia:
you may want to consider the community-contributed programme -xthybrid-.

[Lydia Gad]

Hi Carlo,

Thank you for your assistance, I will look into it.

Best regards,
Lydia Gad