Hello everyone,

My question is not necessarily related to Stata codes but is broadly related to standard error clustering in panel data sets for finance research. Whenever I search information on how to select appropriate standard error clustering level, many people refer to Cameron and Miller (2015) - A Practitioner's Guide to Cluster-Robust Inferences (https://escholarship.org/content/qt1...qt1jq5d0pq.pdf). After going over this material, my takeaway is that you would like to have broad enough clusters to account for the possibility that regressors and the errors might be correlated within cluster. However, when increasing the size of clusters, you have to make bias-variance trade-off in that larger and fewer clusters have less bias but more variability and researchers want to be conservative and avoid bias.

My questions are as follows.

1) When we have the larger and fewer clusters that have less bias but more variability, what are the inferences related to significance of coefficient estimates? Does it mean that as variability within cluster is large, you have larger standard errors, making it more difficult to find significant results? I would like to learn about the implications of having more variability within cluster.

2) Is there any problems with defining clusters too broad that the correlation within cluster may be small? For example, let’s say that my clusters are too broad, and some of the residuals within my cluster are uncorrelated. Is there any problems from defining clusters too broadly?

3) In finance research, I commonly observe standard errors being clustered at the firm level. If broader clusters are more conservative way to model correlated standard errors, why do I not see a lot of studies that cluster standard errors at broader level (e.g., at the industry level)? Do you have a good explanation for my firm clustering would be better than industry level clustering?

All of your advice is greatly appreciated!

Thank you very much in advance.