Suppose I have the following data:
loan_id borrower_id loan_amount lender_id var time
xxxx xxxx xxxx xxxx xxxx 20xx
Suppose that one borrower can have multiple lenders.
If I want to run the following regression:
loan_amount = var + \epsilon
Do I need to cluster by borrower_id, loan_id as well as time? Because a lender/borrower's characteristic might be correlated with the loan amount, and the characteristics might also be correlated through time. If so, am I looking at a three-way clustered robust standard error here?
Do you think the multiway clustered robust standard error in R will work in this case? For example:
vcov <- cluster.vcov(model, cbind(borrower_id, lender_id, petersen$year))
and then use coeftest(m1, vcov)
to get the clustered robust standard errors.
Do you think this makes sense?