Can we use clustse to produce wild bootstrap with xtreg (small # of clusters)

[Ryan Merrill]

Wizards, I am running xtreg on a small panel of company (14 clusters and 16 years). The xtset is thus xtset companyid year.
The vce(robust) option in basic xtreg will give incorrect standard errors following C & M 2015.
I have loaded the clustse package and attempted to run the regression as: clustse xtreg depvar indepvars, cluster (companyid).
The package appears to run but returns the heartbreaking summary below.
Thank you for advising on the following points:
1. can you use clustse with xtreg?
2. in a clean, panel with a handful of factors, what might be driving this error?

Thanks!!


. clustse xtreg production_log total_attacks_0_lag total_kills_0_lag wb_wdi_gdp_yoy, cluster(companyid)
Cluster variable is companyid
Number of clusters is 14
Starting CATs Procedure and attempting to run the model in each cluster
..............
The model ran successfully in 0 clusters out of a possible 14.
If the model failed in a high proportion of clusters, consider using another method.
This data has too few observations per cluster or too little variation
on the variables within clusters to run the model in each cluster.
The Ibragimov-Muller cluster-adjusted t-statistics procedure will not work without
using the force(yes) option to run the model in only successful clusters.
You can try using the wild or pairs cluster bootstrap-t procedures, or try to
use the force(yes) option. If forcing the procedure, pay attention to how many
clusters the model fails in.

Reference: Cameron, A. C., & Miller, D. L. (2015). A practitioner’s guide to cluster-robust inference. Journal of Human Resources, 50(2), 317-372.

[Carlo Lazzaro]

Ryan:
welcome to the list.
Since your cross-sectional dimension is lower than the time-series one, did you consider using -xtgls- instead of -xtreg-?

[Ryan Merrill]

- Carlo, thank you so much for writing, and I will discuss the question with my colleagues. meanwhile I am struggling to get the cgmwildboot program to run at all. I am attempting a simple specification, in lots of iterations, but get the error:

Initial estimtation does not work with cgmreg. Try
cgmreg by itself to see what the problem is.

This is an encouraging error message, but findit cgmreg returns nothing no results in Stata. Is cgmreg no longer in circulation?

Second, returning to my question, can anyone confirm that yes, cgmwildboot or clustse work with panel data?