Robustness Check in Stata

[Anuradha Saikia]

Hello everyone
I tried to check robustness of my model by using alternative measures of dependent variable. In my main model its ROA and in the robustness its ROE . but the variables which are coming significant for ROA are not coming the same for ROE. what do I do.

[Anuradha Saikia]

reg ROE CL FirmAge FirmSize	Lindex Leverageratio Disputedtaxes
Source SS	df MS Number of obs =	1,236
F(6, 1229) =	17.76
Model 253717.694	6 42286.2824 Prob > F =	0.0000
Residual 2926071.68	1,229 2380.85572 R-squared =	0.0798
Adj R-squared =	0.0753
Total 3179789.37	1,235 2574.72824 Root MSE =	48.794
ROE Coef.	Std. Err. t P>t [95% Conf.	Interval]
CL -.0021681	.0045191 -0.48 0.631 -.0110341	.0066978
FirmAge .4118603	.0635544 6.48 0.000 .2871731	.5365475
FirmSize 3.665014	1.013118 3.62 0.000 1.677381	5.652646
Lindex .0033932	.0871871 0.04 0.969 -.1676589	.1744452
Leverageratio -1.495568	.2708771 -5.52 0.000 -2.027001	-.9641354
Disputedtaxes 6.92e-06	.0052459 0.00 0.999 -.0102849	.0102988
_cons -27.80335	6.668618 -4.17 0.000 -40.88649	-14.72021

[Anuradha Saikia]

this is the output with ROE

[Anuradha Saikia]

THIS IS THE output for ROE as dependent variable

[Carlo Lazzaro]

Anuradha:
your second -regress- code highlights the role of residuals vs model.
In addition:
1) return on assessments and return on equity are two different beasts (please consider that my experience with corporate finance and related indexes dates back to https://en.wikipedia.org/wiki/DuPont_analysis ); no wonder that the results of the two regression codes are not consistent;
2) the "ROE regression" shows two coefficients with a highly suspect P-value (approaching one) that deserve a more careful check.
That said, I would stick with the "ROA regression" and replace default standard errors with their bootstrapped counterparts (200 replications would be enough; see https://www.routledge.com/An-Introdu...IaAkG0EALw_wcB page 47; it is also worth noting that in his helpful and well-written textbook on bootstrap (and bootstrapping) (https://www.degruyter.com/document/d...110693348/html), Felix Bittmann recommends 500 replications (see page 98)).

[Anuradha Saikia]

Thanks Carlo Lazzaro . I always see papers with taking ROE to check robustness for ROA . I would look into the sources you mentioned though I have no experience of doing boot strapping.

[Anuradha Saikia]

Also I have noticed this. If I remove a variable of mine , Lindex which is a control variable, the results are different,

[Carlo Lazzaro]

Anuradha:
the adjusted R-sq of your last -regress- code is still dramatically low.
If I were a reviewer, I would perceive this result as an evidence of weakness of your original regression,

[Anuradha Saikia]

Thank you Carlo. will look into it again