Replication Package (ecta-replication-package)

When to trigger

• You are preparing the deposit required under the Data and Code Availability Policy
• Monte Carlo tables cannot be regenerated bit-for-bit from a clean checkout
• An empirical paper has no documented data provenance or access path
• You are at acceptance / final-files stage and the Data Editor will verify reproducibility

Econometrica (and the other Econometric Society journals) enforce a single ES-wide Data and Code Availability Policy. Concrete, Econometrica-specific specifics that differ from the AER/AEJ (AEA Data Editor + openICPSR) pipeline:

• The policy applies to papers with empirical, experimental, and/or simulation results. A pure-theory paper with no such results is effectively exempt — but any requested exemption or limitation on data/code availability must be stated at initial submission (the handling Co-Editor decides; exemptions are not considered later).
• Reproducibility is verified before final acceptance by the Econometric Society Data Editor team. In practice you submit final files at Conditional Acceptance to a separate Editorial Express account for the Data Editor's checks and correspondence.
• The package must include a README in PDF; the Social Science Data Editors' README template is recommended and covers every required item.
• For packages conditionally accepted after July 1, 2023, the replication package is deposited at the Econometric Society Journals' Community at Zenodo (you may reserve a DOI in advance). Another trusted open repository with a permanent DOI can satisfy the requirement with Data Editor approval — but not the AEA/openICPSR route by default.

Verify the current policy text and deposit location on the official ES Data Editor site before finalizing — the specifics evolve.

Build the package as you go, not the night before final files. A package assembled from memory at the end almost never reproduces — and here a real human Data Editor will run it.

What the package must contain

Reproducibility discipline for Monte Carlo

• Seed everything and record the seed alongside each table. Re-running must reproduce the exact numbers, not merely "similar" ones.
• Master script runs all simulations and writes outputs to named files that map to table numbers.
• Runtime honesty. State how long the full simulation takes and on what hardware; if it is days, provide a smaller smoke-test path that runs quickly and a way to verify the full run.
• No manual steps. No "then copy the number into the table by hand" — exhibits should be generated programmatically where feasible.

Empirical data provenance

• Public data: include it (or a script that downloads a fixed version) plus the citation.
• Proprietary / restricted data (e.g., licensed firm-level, confidential admin data): you generally cannot redistribute it. Document the exact source, version, access procedure, required licenses/fees, and contact, so a replicator can obtain the same data. Provide all code, and where allowed, a synthetic or example dataset that exercises the pipeline.
• Confidentiality: strip personal identifiers; respect data-provider agreements; state any approvals obtained.

Recommended structure

replication/
  README.md            # provenance, environment, runtime, exhibit↔script map
  run_all.{do,R,py,jl} # master script: one command rebuilds everything
  code/                # numbered scripts (setup → simulate/estimate → tables → figures)
  data/                # public data or a synthetic example; provenance for restricted data
  output/              # generated tables/figures (regenerable, not hand-edited)
  env/                 # Dockerfile / renv.lock / environment.yml / Project.toml

Checklist

• Every table, figure, and number (paper + Supplemental Material) regenerated by code
• Single master script reproduces everything end to end
• Every random draw seeded; Monte Carlo tables reproduce bit-for-bit (simulations are in-scope)
• Environment pinned (versions + dependencies)
• README in PDF (Social Science Data Editors' template) documents hardware, runtime, data sources, and exhibit↔script map
• Public data included or downloaded by script with a fixed version
• Restricted data: provenance + access path documented; synthetic example provided if allowed
• Any exemption/limitation requested at initial submission (theory paper with no empirical/experimental/simulation results may be exempt)
• Deposit plan: Econometric Society Journals' Community at Zenodo (DOI reserved); Data Editor checks pass at conditional acceptance
• Confidentiality and data-use terms respected; identifiers removed
• Verified against the current ES Data and Code Availability Policy and Data Editor site

Anti-patterns

• Unseeded simulations, so tables only reproduce "approximately"
• A pile of scripts with no master file and undocumented run order
• Numbers transcribed into tables by hand, untraceable to any script
• "Data available on request" with no provenance, version, or access procedure
• An environment that only runs on the author's machine (unpinned versions)
• Redistributing proprietary data in violation of the license
• Leaving package assembly to the final-files deadline

Output format

【Package status】complete / gaps
【Master script】present / missing
【Seeds recorded】yes/no — bit-for-bit reproducible: yes/no
【Environment pinned】yes/no (tool: ...)
【Data】public-included / restricted-provenance-documented / theory-exempt / MISSING
【Exhibit↔script map】complete / gaps: [...]
【Deposit】Zenodo (ES Journals' Community) DOI reserved: yes/no
【Policy check】verified against current ES Data Editor policy: yes/no
【Next step】ecta-referee-strategy