Estimation Precision and Robust Inference in Archival Research

by Joachim Gassen and David Veenman

This repository contains the replication code for our study "Estimation Precision and Robust Inference in Archival Research", published in the Journal of Accounting & Economics, as well as the links to our R and Stata packages for multidimensional fixed effect estimation and multi-way standard error clustering for robust regression estimators.

Overview

• The code folder contains all the programming code to replicate the analyses presented in the paper and Online Appendix. The 00_data_wrds.do file in the code/stata folder provides details on the data sourced used in the paper.

• The packages ferols [R] (https://github.com/joachim-gassen/ferols) and robhdfe [Stata] (https://github.com/dveenman/robhdfe/) combine Huber M-estimation with high-dimensional fixed effects (FE), which was previously not feasible with other robust regression packages.

• The package robcluster2[Stata] (https://github.com/dveenman/robcluster2/) produces multiway cluster-robust standard errors for robust regression estimators. It computes standard errors for robust estimators based on equation 3.11 of Croux, Dhaene, and Hoorelbeke 2003, adjusted for multi-way clustering following Thompson (2011) and Cameron, Gelbach, and Miller (2011).

• The package roboot [Stata] (https://github.com/dveenman/roboot/) produces bootstrapped standard errors for MM-estimators using the fast bootstrap procedure described by Salibian-Barrera and Zamar (2002). The robust estimator in roboot is the MM-estimator defined by robreg mm. The package produces MM-estimates with bootstrapped standard errors without cluster-adjustment, one-way cluster-robust standard errors, or two-way cluster-robust standard errors. Two-way cluster-robust standard errors are obtained using the formula from Thompson (2011) and Cameron, Gelbach, and Miller (2011). The cluster-robust versions rely on the computational efficiency of drawing low-dimensional matrices per cluster as described by MacKinnon (2022), which speeds up the bootstrap procedure.

• The replication code uses the following external Stata packages:

  • robreg, available via SSC [version used: 2.0.8 18sep2021 by Ben Jann]

  • xtrobreg, available via SSC [version used: 1.0.5 29aug2021 by Ben Jann]

  • coefplot, available via SSC [version used: 1.8.8 22aug2025 Ben Jann]

  • reghdfe, available via SSC [version used: 6.12.3 08aug2023 by Sergio Correia]

  • ftools, available via SSC [version used: 2.49.1 08aug2023 by Sergio Correia]

  • gtools, available via SSC [version used: 1.8.1 27aug2021 by Mauricio Caceres Bravo]

  • winsor2, available via SSC [version used: 1.1 16dec2014 by Lian Yujun]

  • outreg2, available via SSC [version used: 2.3.2 17aug2014 by Roy Wada]

  • robhdfe, available via https://github.com/dveenman/robhdfe/ [version used: 1.0.0 08apr2026 by David Veenman]