Julia PoissonFE.jl Package

I’ve released a Julia package, PoissonFE.jl, to estimate Poisson regression with fixed effects. Similar to my R package, poisFErobust, it computes robust standard errors from Wooldridge (1999).

The Julia package is quite a bit faster than R. Aside from the raw speed improvement of Julia, it implements the likelihood function without the fixed effects parameters, so it should be faster than GLM.jl for models with many fixed effect levels (thousands). The estimates should match Stata’s command, xtpoisson y x, fe vce(r).

I continue to be impressed by Julia’s ecosystem. Estimating the coefficients only required implementing the likelihood function. Automatic differentiation via ForwardDiff.jl enables a simple function call to Optim.jl which implements efficient line search algorithms. There was no need to implement the gradient or Hessian by hand.