TJuser / qrfe_bootstrap

Replication files for "Bootstrap inference for panel data quantile regression"

QRFE bootstrap

Replication files for "Bootstrap inference for panel data quantile regression"

Last update: 2023-01-28

Contents

This repository contains files that were used to conduct the Monte Carlo simulation experiments in the paper.

• The main file used for simulation is sim.R.
• The files run_simulations.sh, glue.R, missing.R and make_tables.R are specific to the distributed computing environment that was used (a SHARCNET/Compute Canada system). They were used to serially farm sim.R out to many CPUs, "glue" the parts back together, and then extract the information from the glued files.
• se_experiment.R extracted information about the performance of variance estimates computed using the bootstrap from the data files.
• data/ is a directory containing the results of the simulations for five different data generating processes considered in the paper.
• tables/ is a directory containing the Latex-format tables that appear in the paper.

Basic usage

If you are using R, then the following pseudo-code will compute one bootstrap coefficient vector. It assumes you have response vector y, n*T by p covariate matrix x and n*T by n matrix of unit indicators ind, assuming the panel is balanced (n observations with T time-series observations per unit). Suppose you would like to estimate a quantile regression at quantile level Tau. The commands below create nboot different bootstrap coefficient estimates for the p covariates and you can calculate whatever else you wish from there.

library(quantreg)
w <- rexp(n * nboot, 1)
U <- matrix(rep(w, each = T), n * T, nboot)
bcoef <- boot.rq.wxy(cbind(x, ind), y, U, tau = Tau)[, 1:p]