Replication
All numerical results in Lee (2026) can be checked using this repository or the accompanying journal replication archive. The repository contains the replication script, notebook, requirements file, and input data.
Setup
Clone or download the repository, create a virtual environment, and install the package and dependencies:
cd github_sw2023
python -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip setuptools wheel
pip install -r requirements.txt
pip install .
python -c "import sw2023; print(sw2023.__version__)"
In the journal submission archive, requirements.txt installs the submitted
source tarball. In the GitHub repository, the package is installed from the
checked out source tree with pip install ..
Run
# Tables only; quick Monte Carlo execution check (n_sims=20)
python replication.py --tables
# Tables and figures
python replication.py
# Manuscript-scale Monte Carlo table validation (n_sims=100)
python replication.py --full --tables
Reproducibility Notes
The Monte Carlo exercise is a focused implementation validation against selected Simar and Wilson Table F.1 entries, not a full replication of the entire Monte Carlo appendix. The original authors’ computational code, optimizer settings, tolerances, starting values, and exact random-number streams were not publicly available.
All bootstrap procedures accept a
seedargument. The replication script fixesseed=2023for all stochastic results.The wild bootstrap p-value (Section 6.4) uses
B=999draws; the test statisticTis deterministic and invariant to the seed.The full table command recomputes the manuscript Monte Carlo validation cells with
n_sims=100and writes*_full.csvfiles.Existing Monte Carlo output CSV files are deleted before each run; the submitted workflow cannot silently resume from earlier calculated output.
The default fresh Monte Carlo run is a shorter executable check with
n_sims=20. Its stochastic outputs are written to separate*_quick.csvfiles and are not expected to match the manuscript tables cell by cell.Figure generation uses
viz_rotation_3d.pyfor the direction-vector rotation figure and fresh fits tonorway_for_python.csvfor the Norwegian bandwidth-comparison figure; the figure run writesnorway_loocv_comparison.csvas an output.The Section 6 simulation reproduces the reported bootstrap test statistic
T=0.0397, p-value0.8819, and bootstrapTrange[0.0144, 0.2358]withseed=2023.