firm-year observation with macroeconomic time series data?

[River Huang]

Dear All, I am confused with the results obtained from, for example, this (kind of) paper (Oil price uncertainty, CSR and institutional quality: A cross-country evidence). This data is here (excel: pttn21ene.xlsx , Stata 16 dta: pttn21ene.dta ). Take the Table 4 as an example

The trick is that the paper controls for "year" fixed effects (FE), along with firm fixed effects (two-way FE model). However, the key (macroeconomic) variable "OILVOL" is the same for all firms for each year. Clearly, it is "perfectly correlated" with year FEs. It can be seemed by the following code (ssc install reghdfe):

// Table 4, column (1)
reghdfe esg oilvol size i.year, a(id) vce(cl id year) 
reghdfe esg oilvol size, a(id year) vce(cl id year)

From

reghdfe esg oilvol size i.year, a(id) vce(cl id year)

It seems to me that Stata drops the last variable (see 2017 year dummy variable above) when there are many variables perfectly correlated.
However, from

reghdfe esg oilvol size, a(id year) vce(cl id year)

You can see that the key variable "oilvol" is omitted due to multicolinearity.

Thus, my question is that do the results in Table 4 make sense?

[Andrew Musau]

You are correct River. The way to see it is to change the base category of the year dummies and see if the coefficient on oil volume also changes. This should be the case basing on your second reghdfe results. Therefore, the coefficient on oil volume is not identified. If the paper is based on this result, send in a comment to the journal.

reghdfe esg oilvol size ib2003.year, a(id) vce(cl id year)
reghdfe esg oilvol size ib2004.year, a(id) vce(cl id year)

[River Huang]

Dear Andrew, Thank you for the comments and suggestions.