Dear Statalist members,

I use Stata 15.1 and I have several issues with my data. First I'd like to explain what my research is about as I think it's important to understand where I am heading to.

In my study I am looking at German companies from 2013 to 2018 (no. of obs 725). These companies are sorted into different stock exchange segments. From 2013 to 2016 all companies were within one segment. Due to regulatory changes this segment was closed and all companies had to - more or less (depending on wether they could fulfill certain KPIs) - decide in which of the two new stock exchange segments they want to be sorted into.


What I want to do now is to examine wether companies within one of the new segments benefit positive economic consequences such as higher liquidity and higher valuation due to the new segment they are in. (Because the segments differ concerning the regulatory level; from 2013 to 2016 it was at intermediate level and in 2017 & 2018 one segment is high(er) and one is low(er)). That is why I want to run regressions with the following structure. (segment=dummy variable, 1=2013-2016 segment (intermediate level), 2=2017 & 2018 segment (high level) & 3=2017 & 2018 segment (low level). I use the dummy variable as follows: i.segment (1, 2 or 3).


Liqit =α0 + α1SEGMENTit + α2LN_SIZEit + α3LN_VOLUMEit + α4LN_VOLATILITYit + α5GDPt + εit

1) Is there a test to determine wether I have panel data or fulfill the requirements of panel data?
2) Is there any reason why a pooled OLS with White standard errors (reg, robust) would be more suitable than a fixed effects regression (xtreg, fe)?

I am asking these questions because the results under using "reg, robust" better match my theoretical assumptions. I am aware of the fact, that i cannot just use the results that I like more instead of using the model that fits my data structure. But when I am using time-fixed effects the years 2017 and 2018 are omitted.

When I am using "xtreg, fe" just for the two new segments in 2017 and 2018, one of the segment also gets omitted (but only when I run the regression pairwise for segment 2 and 3).

This is why I am thinking, that the fixed effects model might not be suitable for my study as the segment variables might be time invariant (2013 - 2016 there is only segment 1; 2017-2018 there is either segment 2 or 3).

I hope my thoughts are fairly understandable.

Thanks a lot.
Martin.