Currently, we rely on the bootstrap, which can be very slow for larger problems. See the derivation of analytical standard errors in the paper: [link](https://ideas.repec.org/a/ucp/jlabec/doi10.1086-683668.html)
