Repository navigation
PPML with high-dimensional fixed effects? #143
Description
Activity
Hi Gabriel,
You're right, calculating the PPML and friends in the end boils down to a series of pseudo de-meaning operations followed by OLS, which is what FixedEffects.jl is doing. I've implemented a version of this in GLFixedEffectModels.jl, maybe this is of use to you. I should add that bias correction for the incidental parameter problem (which is a nice way to kill off annoying seminar questions) is not yet implemented, and that I also haven't done a lot of testing.
I think in the end it may make sense to put this into FixedEffectModels.jl, depending on how Matthieu and Erik feel about this.
Reacted by Erik Loualiche and DanielReacted by Gabriel E Kreindler and azev77A quick update (I cannot believe that >2 years have passed!!) that I have been using a lot of the following, and it's been incredibly useful:
using GLFixedEffectModels myformula = term(:myy) ~ term(:myx) + GLFixedEffectsModels.fe(:myfe1) + GLFixedEffectsModels.fe(:myfe2) nlreg(df, myformula, Poisson(), LogLink(), GLFixedEffectsModels.Vcov.cluster(:a, :b))Reacted by Johannes BoehmGlad to hear it's been useful!
I believe it's closed by @jmboehm package
Thank you for the fantastic package!
I would like to ask whether there are any plans to implement a Poisson pseudo-maximum likelihood estimator with high-dimensional fixed effects, similar to STATA's PPMLHDFE?
http://scorreia.com/software/ppmlhdfe/
Alternatively, is it possible to call FixedEffectModels from another package that estimates PPML?