技能 数据科学 学术研究可复现性与数据政策指南

学术研究可复现性与数据政策指南

v20260724
jbf-replication-and-data-policy
本指南详细介绍了根据《银行与金融杂志》(JBF)和爱思唯尔Option C要求,构建完整、可复现的研究数据包的最佳实践。它涵盖了数据可用性声明、代码结构化、以及处理专有或受限数据的要求,确保学术研究的透明度和可验证性。
获取技能
259 次下载
概览

Replication & Data Policy (jbf-replication-and-data-policy)

When to trigger

  • The paper uses CRSP, Compustat, BankFocus, Dealscan, TRACE, Call Reports, hand-collected data, or simulations
  • You need a data availability statement for Elsevier/JBF submission
  • You want a legal and useful replication package despite proprietary data

Policy stance

JBF follows Elsevier Option C research-data instructions. At submission, authors must state data availability. For shareable data, deposit the research data in a relevant repository and cite/link it in the article. If data cannot be shared, state why (for example, proprietary, sensitive, confidential, or licensing limits). JBF does not operate a dedicated journal-run replication archive, so the practical package should combine a repository deposit where rights allow, code, metadata, and a precise access route for restricted inputs.

Package structure

README.md
data_access/          # licensed-data instructions, not raw proprietary files
code/
  00_setup.*
  01_build_sample.*
  02_main_tables.*
  03_figures.*
  04_robustness.*
output/
  tables/
  figures/
requirements.*        # renv.lock / requirements.txt / Stata package list

Data availability statements

  • Public data: name source, URL, access date, and archive copy if permitted.
  • Proprietary data: identify vendor, product, access terms, and exact query or extraction instructions; do not redistribute restricted files.
  • Hand-collected data: share the dataset if rights allow, with coding rules.
  • No external data / simulation: state that no external data were used and provide simulation code and seeds.

Code expectations

  • One master script regenerates all tables and figures.
  • Random seeds and software versions are recorded.
  • Output files match manuscript numbering.
  • Confidential paths and author identity are removed before double-anonymized submission.

Proprietary-data audit

For licensed banking/finance sources, add a data_access/README.md that records:

  • vendor/product, query filters, identifiers, and date accessed;
  • variables needed to recreate the analysis;
  • any hand-cleaning or link-table rules;
  • whether a pseudo dataset can exercise the full code path;
  • who can legally access the raw data and under what license.

Source-by-source sharing matrix

Typical JBF source Redistribution What the package carries instead
US Call Reports / FDIC SDI public — share extracts pull scripts with vintage, date, and series mnemonics
Orbis Bank Focus / BankScope legacy licensed — do not share query filters, consolidation codes, variable list, access date
DealScan licensed — do not share extraction steps, facility/package level, link-table code
CRSP / Compustat (when used) licensed — do not share WRDS query description and merge keys
Hand-collected regulatory events usually shareable coding manual, per-event sources, inter-coder checks
Confidential supervisory data cannot leave the authority aggregate descriptives plus the access route for replicators

Worked statement draft (illustrative)

Data availability. Bank balance-sheet data come from US Call Reports (FFIEC,
public; pull scripts and vintages in /code/01). Loan-facility data are from
DealScan under license; we provide extraction filters and borrower link-table
code but not raw files. Hand-collected deregulation dates and coding rules are
deposited with the package. All tables and figures regenerate from run_all
with the licensed inputs in place; a synthetic sample exercises the full
pipeline without them.

Pre-submission package audit

  • run_all executes on a clean machine with only the licensed inputs mounted
  • every manuscript exhibit maps to exactly one output file, with matching numbering
  • the synthetic or masked sample passes every script end to end
  • vendor vintages and download dates recorded for Bank Focus, DealScan, and any WRDS pulls
  • no author-identifying paths or initials in code comments (double-anonymized review)
  • repository deposit linked and cited where rights allow, or non-sharing reason stated clearly

A package failing two or more boxes usually signals deeper pipeline debt; route back to jbf-data-analysis before drafting the statement.

Editor pushbacks on data

  • "The package cannot run without DealScan." → ship a synthetic or masked sample that exercises every script end to end.
  • "Which Bank Focus vintage?" → record vendor vintage and download date; bank coverage shifts across vintages.
  • "Confidential supervisory data make this unreviewable." → document the authority's access procedure and provide all code plus aggregate moments; the data statement must explain the access limits.

Output format

[Data status] public / proprietary / confidential / simulated / mixed
[Statement draft] ...
[Shareable files] ...
[Restricted files] ...
[Reproducibility gaps] ...
[Next step] jbf-submission
信息
Category 数据科学
Name jbf-replication-and-data-policy
版本 v20260724
大小 5.22KB
更新时间 2026-07-28
语言