Year and Industry Fixed Effects for non-panel data

[Rick de Leest]

Dear all,

I have a question regarding a regression I am trying to run on some M&A transactions. I am trying to include both YR and Industry which I have coded by year and given a numerical value respectively. I want to verify whether my intuition is correct and I am using STATA the way I should. I can find a lot of threads on FE and panel data, but unfortunately nothing really conclusive on something comparable to my own research.

The code I am running is as follows:

reg indpvar depvar depvar ... i.year i.industrycode, robust

My problem is this leads to serious indications of Multicollinearity (VIFs for year dummies over 100). Furthermore, as the industry dummies sometimes only have 2-5 observations their significance seems less trustworthy. I am not extremely confident whether what I am doing is correct, any help would be very much appreciated.

Kind regards,


Rick

[Carlo Lazzaro]

Rick:
- the only way to reduce multicollinearity issue is to remove from the set of predictors one of the variables causing the problem; when multicollinearity creeps up in interactions, one suggested approach is to center around a meaningful value (i.e., the mean) the variable for which you are investigating the conditional effect;
- however, if you have panel data, why don't rely on -xt- suite of commands?;
-eventually, your chances of getting helpful replies are conditional on posting what you typed and what Stata gave you back (as per FAQ). Thanks.