Perla, Tonetti, and Waugh: Corrigendum Replication

JAX replication of the corrected model in Perla, Tonetti, and Waugh (AER): SMM calibration, balanced-growth-path counterfactuals, transition dynamics, and every table and figure the corrigendum includes. The repository is self-contained: the published-model overlays and SMM data targets are frozen CSV artifacts in data/.

See Corrigendum: Equilibrium Technology Diffusion, Trade, and Growth for a description of the new quantitative results and the numerical algorithm.

Setup

Install uv:

• curl -LsSf https://astral.sh/uv/install.sh | sh on MacOS or Linux
• powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex" on Windows

uv run installs the pinned dependencies on first use; uv sync installs them explicitly from uv.lock.

Full replication

Replicate all results and generated figures, tables with

bash replicate.sh

Results are saved output/. Timings on an Apple-silicon laptop are about 1 minute. Individual drivers, from the repository root:

uv run python replication_estimation.py     # SMM at the published AER theta and sigma: fit and welfare
uv run python estimation_robust.py          # independent closed-form-versus-grid verification of that fit
uv run python replication_results.py        # transition dynamics for the 10% trade-cost cut: welfare and result artifact
uv run python replication_tables.py         # the corrigendum's tables as \input-ready fragments, the eq-(60) welfare decomposition, and the published-path welfare re-aggregation
uv run python replication_figures.py        # manuscript figures and an auxiliary counterfactual sweep
uv run python bgp_grid_convergence.py       # discrete-BGP-to-analytic convergence table across grid refinements
uv run python numerics_macros.py            # numerics-appendix macros sourced from the artifacts above

The estimation writes estimation_theta4.98898_sigma3.16692.json; the downstream drivers load their parameters from it.

Outputs

A full run leaves in output/:

• estimation_theta4.98898_sigma3.16692.json — the accepted SMM fit
• verification_theta4.98898_sigma3.16692.json — the independent verification summary
• transition_results.json and transition_arrays.npz — the auditable transition result artifact
• bgp_grid_convergence.json — the grid-convergence table inputs
• inline_results.json — the rounded numbers used in the corrigendum's prose
• table_params.tex, table_moments.tex, table_persistence.tex, table_results.tex, decomposition_equation.tex — \input-ready table fragments
• inline_macros.tex, figure_macros.tex, numerics_macros.tex — \input-ready macro files
• the eight manuscript figures corr_density.pdf, ovl_trade.pdf, ovl_omega.pdf, ovl_fixed_costs.pdf, ovl_log_consumption_inset.pdf, ovl_growth.pdf, corr_gbm_chi.pdf, corr_delta_chi.pdf, plus the auxiliary ovl_counterfactual.pdf

Variations

Every driver exposes its options on the CLI (--help for the full list):

# results-only transition solve on a coarser preview grid (the table and
# figure drivers intentionally reject it because they require the production
# AlgorithmSettings; the defaults are the production grids)
uv run python replication_results.py \
  --alg_set.z_segment_1_points 720 --alg_set.z_segment_2_points 960 \
  --alg_set.z_segment_3_points 480