fixed effects by group and individual

[Julia Lanzani]

Hi everyone,

I want to run a regression with several fixed effects. Specifically, I want to include fixed effects at the company level (including total assets, net income, and other firm characteristics) and at the shareholder level (including number of shares/ holdings, ...). However, I am not sure how to tell Stata that I have fixed effects at the company and the shareholder level. I have read about the group fixed effects, but I am not sure if this is the correct way of doing it. At the moment it would be sufficient to understand the example provided here (found at http://scorreia.com/help/reghdfe.html#examples):

reghdfe citations funding, a(year inventor_id) group(patent_id) individual(inventor_id)

Now for my usecase I need to find out how I can absorb additional variables that belong to the group patent_id and inventor_id. Can I just add them to the absorb function above? Additionally, I want to know if I need to add the comand vce(cluster patent_id) if I group by patent_id.

Thank you in advance,
Julia

[Jesse Wursten]

In reghdfe you can include as many variables in the "absorb" option as you want. So if you want company and investor fixed effects, you can specify absorb(companyID investorID). How to cluster your standard errors is a more complicated question, I recommend reading Cameron & Miller 2015: http://jhr.uwpress.org/content/50/2/317.short

[Julia Lanzani]

Hi Jesse, thank you very much. One follow-up question: So if I have company and investor fixed effects with IDs for both company and investor AND additionally I have several characteristics for both company (e.g., size, net income, ...) and investors (e.g., investment style). How do I tell STATA that these characteristics should be absorbed at the company and investor level too? Is it as simple as absorb(companyID size net_income investorID investment_style)?

[Andrew Musau]

I have several characteristics for both company (e.g., size, net income, ...) and investors (e.g., investment style). How do I tell STATA that these characteristics should be absorbed at the company and investor level too?

You call such variables control variables, i.e., variables that are not of direct interest to your research but whose effects must be taken into account in your analysis. reghdfe (from SSC) absorbs indicators, so the control variables must be included in the same fashion as your independent variables in the regression. Whether they vary over firms or investors matters not in terms of how they should be included. Of course, if investment style is time invariant, it will be collinear with the absorbed dummies and will drop out. But that is nothing to worry about as it is a control variable, and its effects are already captured by the dummies.

reghdfe citations funding size netincome investment_style, a(company year investor) cluster(investor)

[Julia Lanzani]

So just plugging the controls in the absorb function would not work?
Like this:
reghdfe citations funding, a(size netincome investment_style) cluster(investor)

[Andrew Musau]

You could technically absorb any indicators, but not continuous variables. In general, you should absorb the fixed effects and include everything else as shown in #4.