Using industry-fixed effects and a comparable industry dummy simultaneously

[Alyssa Jourdan]

Dear Stata experts,

I am currently doing research on institutional ownership in publicly traded US sin stocks using Stata 14.1. The dataset exists of about 3000 stocks over the time period 2003-2016. The main regression is presented below:

xtreg IO sin compind2 logMarketcap beta inverseprice SD AMR NYSE NASDAQ AMEX, i(time_industry)fe vce(cluster indcode)

As you can see I am using xtreg, time- and industry- fixed effects and cluster standard errors per industry. The dependent variable is institutional ownership, IO. The dummy variable sin indicates whether a stock is perceived a sin stock or not. Aside you find some standard control variables, such as size, company beta, inverse price, average monthly returns, standard deviation of daily return and dummies for the exchange a stock is trading on.

Now, my question regards the variable compind2, this dummy variable equals one if a stock resides in an industry that is similar to a sin industry. For example: the beer industry is seen as a sin industry and the dummy compind2 is one if a stock resides in the soda industry as they both are consumer goods. I intend to include this dummy in order to control for a possible preference of institutions for industries that are similar to sin industries.

This results in the following question: as I also include industry-fixed effects, is it unnecessary to include this dummy or do they correct for different aspects of industries?

Thanks in advance!

Kind regards,

Alyssa Jourdan

[Clyde Schechter]

If for any given industry, the value of compind2 is the same for all values of that industry, then it is completely redundant. Not only is it unnecessary to include it, Stata will automatically omit it from the model when you run it. If, however, two observations for the same industry can have different values of compind2, then it is new information and can be explicitly included in the model if you deem it relevant.

I don't understand your explanation of compind2 well enough to perceive which of these cases applies, but presumably you do.