Improving panel data regression model

[Simona Galletta]

Dear all, I've just signed up to Statalist. I am a new user of Stata and thanks to this forum I learned the basics commands. I implemented them in my dataset which is a panel of banks from 2011 to 2017. I run a regression with the command -xtreg- adding fixed effects and robust options plus I lagged my dependent variable. Now my questions are :
1)How can I improve my model ?
2)Do I have to correct for endogeneity?
3)What am I going to do as "robustness check" after this regression ?

I hope to make myself clear, since I am new.
Any help would be great.
Regards

xtreg index l.index wlnta wtier wcinc wnim wimnp wnlta i.year, fe r

.

[Carlo Lazzaro]

Simona:
welcome to this forum.
If you include among the set of predictors the lagged dependent variable, -xtreg- wil be biased, and you should consider -xtabond-.
Please note that panel data analysis regression is far from being trivial: hence, it's recommended that you get yourself (really) familiar with the building blocks of (linear) panel data analysis by studying a decent panel data econometrics textbook (see, for Stata users, https://www.stata.com/bookstore/micr...metrics-stata/).
As far as your questions are concerned:
1) improving your model means giving a fair and true view of the data generating process. For instance, you do not say if your dataset was checked for omitted variable bias (which can reveal some source of endogeneity 2);
3) any claim about robustness should specify: robustness to what?

[Simona Galletta]

Dear Carlo,
Thanks for your reply. I am truly sorry for my lack of knowledge.
Currently I am studying "Econometric analysis of panel data" (Badi H. Baltagi) fifth ed., but I am going to buy the manual you suggest.
I will check for omitted variable, once I'll figured out how to do it and try xtabond as well. About robustness I mean if there are tests that will help me to find and correct the problems in the estimation.(e.g. Hausman test)

Kind regards,
Simona

[Carlo Lazzaro]

Simona:
you should not be sorry about Learning something from scratch because, as the most prolific contributor of this forum(Nick Cox) said once (in an English only a bit better than mine !!): "We are all beginners. Some of us are only more experienced".
That said:
-hausman- test can help you out in deciding which one, between -fe- and -re- specification, fits your data better. Unfortunately, it allows default standard errors only. In your case, provided that you can stick with -xtreg- after revising your model but keeping -cluster()- standard error, you can use a cousin of -hausman-, the user-written programme -xtoverid- that allows non-default standard errors;
-worked-out examples on Ramsey specification test for detecting non-linearity among predictors and dependednt variable (which is usually sold with the promising but misleading definition of -omitted variable bias test-) are reported in another valuable (although a bit specific) Stata Press handbook (https://www.stata.com/bookstore/heal...-using-stata/: pages: 60-62). Although it does not cover panel data analysis it deploys some interesting (health) econometrics issues.
As an aside, you cannot consider -xtreg- and -xtabond- as aliases,because the underlying theoretical requirements are different: I would recommend you to discuss model issues with your teacher/supervisor/professor and/or a more experienced colleague.

[Simona Galletta]

Dear Carlo,
thanks for being supportive and for the advice. I will try to implement them and maybe I will give you some good news.

Kind regards,
Simona