Fast Estimation of Linear and Poisson Models with High-Dimensional Fixed Effects in Python : The FastHDFE package

Titelangaben

Larch, Mario ; Schoenfeld, Mirco ; Shikher, Serge:
Fast Estimation of Linear and Poisson Models with High-Dimensional Fixed Effects in Python : The FastHDFE package.
Bayreuth, Germany , 2026 . - 70 S.

Volltext

	Format: PDF Name: fasthdfe-techreport.pdf Version: Veröffentlichte Version Verfügbar mit der Lizenz Creative Commons BY 4.0: Namensnennung
	Download (579kB)

Abstract

We present FastHDFE, an easy-to-install Python package providing the commands reghdfe and ppmlhdfe for estimating linear and multiplicative (Poisson pseudo-maximum-likelihood, PPML) models with high-dimensional fixed effects. They rest on the method of alternating projections and, for PPML, iteratively reweighted least squares, and provide homoskedastic, heteroskedasticity-robust, and multi-way cluster-robust standard errors, detection and removal of separated observations via the iterative rectifier, and singleton handling, with the demeaning step implemented in compiled C code. The commands closely replicate the Stata packages of the same name, reproducing their point estimates to at least six decimal places and the standard errors to at least five decimal places. On the ITPD-E gravity dataset, with roughly 83 million observations and more than four million fixed effects, they run about six to forty-four times faster than Stata while integrating naturally into the Python ecosystem.