Replication: Panel Local Projection

This repository hosts a replication package for

• Ziwei Mei, Liugang Sheng, Zhentao Shi (2026), "Nickell Bias in Panel Local Projection: Financial Crises Are Worse Than You Think," Journal of International Economics, 104210. [arxiv:2302.13455].

• paper_summary.pdf is a minimum non-technical summary of the paper.

The replication code is written in R.

Repository Structure and Replication Overview

• applications/: Empirical applications. Contains the data files (empirical_*.csv), the main R routine (LP_panel_all.r). The Jupyter notebook replication.ipynb reproduces the results and figures of all the four empirical application in the paper (Running time: <=30 seconds).
• simulations/: Monte Carlo simulations for the main text (Figures 1-3). See simulations/README.md for details about the scripts and figures.
• simulations_appendix/: Simulation code for Appendix C (Figures C1-C12). The workflow is explained in simulations_appendix/readme.md for instructions.

In short:

• Empirical results -> applications/
• Main-text simulations -> simulations/
• Appendix simulations -> simulations_appendix/

Packages

• Stata package

• R package provides functions to implement the estimation method.

• Python package is a replica of the R package

Docker Environment

DockerHub

An R environment is provided at DockerHub. It can run in Github's codespaces directly (Install a Jupyter Kernel extension in the remote vscode if popped up.)

Local Environment

To run the R notebook in a fully reproducible environment, build the Docker image from the repository root:

docker build -t plp .

Launch the container, exposing JupyterLab on port 8888 and mounting the repository so that changes persist:

docker run --rm -it -p 8888:8888 \
  -e JUPYTER_TOKEN=panel-lp \
  -v ${PWD}:/home/jovyan/work \
  plp

The container will print a URL containing the token (or the JUPYTER_TOKEN you supplied). Open that link in a browser to access JupyterLab, then open applications/replication.ipynb. All required R packages (ggplot2, reshape2, ggpubr) are pre-installed in the image.

Documentation Site

The GitHub Pages site in docs/ can be previewed locally with Jekyll:

cd docs
bundle install
bundle exec jekyll serve --livereload

This requires Ruby (>=3.0 recommended) and Bundler. Once the server is running, open the URL printed in the terminal (typically http://127.0.0.1:4000) to browse the site.

Contributors

• Pan Ji, Shen Shu, Shi Zhentao
• Please contact Shen Shu if you have any inquiries.

License

This work is licensed under the MIT License.