[Carlo Lazzaro]

Adam:
please repeat -testparm- just after -xtreg,fe-.

[Adam Klaas]

Adam:
please repeat -testparm- just after -xtreg,fe-.

. testparm i.Year

 ( 1)  2012.Year = 0
 ( 2)  2013.Year = 0
 ( 3)  2014.Year = 0
 ( 4)  2015.Year = 0
 ( 5)  2016.Year = 0
 ( 6)  2017.Year = 0

       F(  6,   349) = 1046.45
            Prob > F =    0.0000

[Carlo Lazzaro]

Adam:
that's good.
The -testparm- outcome tells you that, due to the joint statistical sigificance of -i.year- uou should keep this predictor in the right-hand side of your regression equation.
Therefore, stick with your last -xtreg,fe- equation.

[Adam Klaas]

[QUOTE=Adam Klaas;n1646852]


Now I am getting this when I use the Year as timevariable:

. xtreg log_realHP log_PopulationDensity log_Population logunemployment real_consCost log
> Real_income real_interest i.high_dev i.Year,fe vce(robust)
note: 2018.Year omitted because of collinearity
note: 2019.Year omitted because of collinearity

Fixed-effects (within) regression Number of obs = 3,020
Group variable: GM_code Number of groups = 350

R-sq: Obs per group:
within = 0.8950 min = 1
between = 0.0493 avg = 8.6
overall = 0.0021 max = 9

F(13,349) = 1091.08
corr(u_i, Xb) = -0.5984 Prob > F = 0.0000

(Std. Err. adjusted for 350 clusters in GM_code)
---------------------------------------------------------------------------------------
| Robust
log_realHP | Coef. Std. Err. t P>|t| [95% Conf. Interval]
----------------------+----------------------------------------------------------------
log_PopulationDensity | .0408971 .0300572 1.36 0.175 -.0182189 .100013
log_Population | .0885554 .0559159 1.58 0.114 -.0214192 .19853
logunemployment | .0626742 .0115337 5.43 0.000 .0399898 .0853585
real_consCost | .7189247 .0103547 69.43 0.000 .6985592 .7392901
logReal_income | .2008085 .0887648 2.26 0.024 .0262272 .3753898
real_interest | .5124789 .0076199 67.26 0.000 .4974922 .5274656
1.high_dev | -.0151478 .0041305 -3.67 0.000 -.0232717 -.007024
|
Year |
2012 | 1.015005 .0154797 65.57 0.000 .98456 1.04545
2013 | 2.625527 .0389241 67.45 0.000 2.548972 2.702082
2014 | 3.206205 .0472807 67.81 0.000 3.113214 3.299196
2015 | 2.712921 .0409138 66.31 0.000 2.632452 2.79339
2016 | 1.726945 .0260159 66.38 0.000 1.675777 1.778112
2017 | .692979 .0106379 65.14 0.000 .6720565 .7139014
2018 | 0 (omitted)
2019 | 0 (omitted)
|
_cons | -68.49108 1.869881 -36.63 0.000 -72.16874 -64.81343
----------------------+----------------------------------------------------------------
sigma_u | .29871983
sigma_e | .02811635
rho | .99121869 (fraction of variance due to u_i)
-----------------------------------------------------------------------------------

Some more question sir, I appreciate it so much that you make time and help us I again thank you for that sir.
So I dont really have to look at the significance of the year dummies, but I should keep it in all my regressions? What about the multicollinearitt of my low_devA and Year 2018 and 2019? The coefficient of the interest is positive, I would have expected that it would be negative. Because if the interest increases by 1 percentage point from the result it indicates that house prices increases by 0.5? Or does it have to do that people are scared that the interest keeps increasing and buy now rather than later?

[Carlo Lazzaro]

Adam:
1) yes, you have to look at the joint statistical significance of the -i.year- and keep it in the right-hand side of your regression equation. At the same time, you should not take the statistical significance of each single year into account;
2) -i.low_dwa- is collinear with the fixed effect (time-invariant predictor). No worries about its omission, bexause it is expected as per -fe- machinery;
3) the same comment holds for one of the omitted year, whereas the other one is omitted to protect your regression from the effects of yhe so called dummy trap (see Wikipedia for further details).

[Adam Klaas]

Adam:
1) yes, you have to look at the joint statistical significance of the -i.year- and keep it in the right-hand side of your regression equation. At the same time, you should not take the statistical significance of each single year into account;
2) -i.low_dwa- is collinear with the fixed effect (time-invariant predictor). No worries about its omission, bexause it is expected as per -fe- machinery;
3) the same comment holds for one of the omitted year, whereas the other one is omitted to protect your regression from the effects of yhe so called dummy trap (see Wikipedia for further details).

Thank you very much it makes sense!