Industry fixed effects omit the independent variable

[Maaike Koese]

Dear STATALIST community,

In conducting my research, I encounter the following problem. My dataset consists of around 6,500 M&A deals. My research question is what effect the delta biodiversity risk exposure (deltaHoldingBDR), which represents the difference between the acquirer and target industry-level holding-based biodiversity risk exposure, has on the CARs of the acquiring firm. In my view, it is essential to include year, acquirer industry and target industry fixed effect. However, STATA omits my independent variable then.

Does anyone has some tips to resolve this problem?

Thanks a lot in advance!

[Nils Enevoldsen]

Perhaps absorb(Year AcGIC_cat TarGIC_cat) is sufficient for your purposes, instead of absorb(Year##AcGIC_cat##TarGIC_cat)? Are your GICs at the sector level, so there are 11 possibilities? Let's say you have twenty years of data. Then you've added 11*11*20 = 2420 fixed effects (most of which don't appear even a single time in your data). But perhaps 11+11+20 = 42 fixed effects is enough.

Also, you say you have 6,500 deals in your dataset, but the regression you showed has 800 observations. Is this what you expect?

[Maaike Koese]

Thanks a lot Nils for your reply! The GICS are on industry group level (24 categories) and I have 12 years of data. Further, my apologies for the confusion. The screenshot that I included is a regression of a subset of my whole sample which encounters the same problem.

When running the regression without the interaction effects between the industries, the regression unfortunately still shows an omitted independent variable. Do you happen to know if there is something else I am doing wrong?

[Andrew Musau]

In my view, it is essential to include year, acquirer industry and target industry fixed effect.

If the variable does not vary within acquirer industry, then its coefficient is not identified in the regression. Since you have a firm level panel, assuming that the variable varies within firms, you can include it and have firm and year fixed effects. If not, look at the correlated random effects (Mundlak) regression as an alternative.