impex-machina/trade-elasticities

v0.3.0 (2026-07-08). Full regeneration on the six-patch fix series (GitHub tag v0.3.0). Gamma re-levels onto Soderbery (2018) Table 2 benchmarks (gamma/(1+gamma) = 0.405 vs his 0.408): gamma median rises 0.238 -> 0.680 and the implied median export-supply elasticity is now 1.47 (the prior-scale bug in v0.2.0 biased gamma down and inflated the implied elasticity). Sigma is essentially unchanged (median 2.878). Standard errors are corrected (sigma_se was understated, rho_se overstated); the weak-IV screen now uses the minimum-eigenvalue Cragg-Donald statistic (Stock-Yogo pass 59% -> 17% -- the honest number); sigma_robust passes on 10.6% of rows. Details: docs/methodology/v020_v030_comparison.md in the GitHub repo. v0.2.0 remains available pinned at revision 7e598f6cb98e -- do not mix versions within one analysis.

2026-07-10. Added exporter-cluster bootstrap SE benchmark outputs under validation/ (bootstrap_se_summary.csv, per-cell bootstrap_se_cells.csv, 736 cells): real-data calibration of the analytic sigma_se by identification stratum and estimator branch. See docs/methodology/validation_section_draft.md (Section 4) in the GitHub repo for results and interpretation.

Trade Elasticities — BACI HS92 V202601

Importer-product-exporter trade elasticity estimates: heterogeneous import-demand elasticities (sigma) and inverse export-supply elasticities (gamma) estimated from CEPII BACI bilateral trade data, following Soderbery (2018) and Grant & Soderbery (2024).

This dataset holds the published outputs of the estimation pipeline. The code that produces them, full methodology, and replication instructions live in the GitHub repository:

https://github.com/impex-machina/trade-elasticities

License

Data in this dataset: CC BY 4.0. The pipeline code (in the GitHub repo) is licensed separately under MIT.

What's here

Outputs are organized by pillar. The authoritative index — including SHA-256 checksums and provenance — is data/manifest.csv in the GitHub repo; the table below mirrors its human-readable view.

Path	Pillar	Description
stage1/baci_hs92_v202601_elast_country_hs4_feenstra_sigma.rds	1	Stage 1 sigma estimates (HLIML primary, Step 2 fallback)
stage2a/baci_hs92_v202601_elast_regional_hs4_fixed_sigma.rds	1	Stage 2a regional gamma with fixed sigma
stage2b/baci_hs92_v202601_elast_country_hs4_fixed_sigma.rds	1	Stage 2b country-level gamma with shrinkage + SE
stage2b/..._summary.rds / .txt	1	Country-pair summary table (binary + human-readable)
validation/liml_validation_tier1a.csv	2	Synthetic recovery: Tier 1a sigma grid
validation/liml_validation_tier1b.csv	2	Synthetic recovery: Tier 1b sample-size convergence
validation/se_calibration_mc_summary.csv	3	SE calibration Monte Carlo (4 regimes x 3 formulas)
validation/se_calibration_mc_per_param.csv	3	Per-parameter calibration detail

The three pillars: (1) the BACI HS4 empirical core, (2) synthetic recovery of the estimator, (3) standard-error calibration. See the repo's docs/methodology/ for details.

Raw BACI data is NOT here

The raw CEPII BACI HS92 V202601 trade data is not redistributed in this dataset (it is CEPII's to distribute, and it is large). Download it directly from CEPII:

https://www.cepii.fr/CEPII/en/bdd_modele/bdd_modele_item.asp?id=37

Place it under data/raw/ in your clone of the repo. The pipeline reads it from there.

Loading the outputs in R

The recommended path is to clone the GitHub repo and use the bundled loader, which reads the manifest and verifies checksums:

# from the repo root, after renv::restore()
source("R/load_outputs.R")
load_outputs()                       # downloads all manifested files to data/derived/
x <- readRDS("data/derived/stage2b/baci_hs92_v202601_elast_country_hs4_fixed_sigma.rds")
head(x)

To pull a single file directly from this dataset without the repo:

url <- paste0("https://huggingface.co/datasets/impex-machina/",
              "trade-elasticities/resolve/main/",
              "stage2b/baci_hs92_v202601_elast_country_hs4_fixed_sigma.rds")
tmp <- tempfile(fileext = ".rds")
download.file(url, tmp, mode = "wb")
x <- readRDS(tmp)
head(x)

Citation

If you use these data, please cite the paper (DOI to be added on publication) and the underlying sources:

• Soderbery, A. (2018). Trade elasticities, heterogeneity, and optimal tariffs. Journal of International Economics, 114, 44-62.
• Grant, M. & Soderbery, A. (2024). Heteroskedastic supply and demand estimation: Analysis and testing. Journal of International Economics, 150, 1-23. https://doi.org/10.1016/j.jinteco.2023.103817
• CEPII BACI World Trade Database, HS92 V202601.