Rof Data Analysis

Review of Finance Data Analysis

Use this when the finance result is not yet credible enough for RoF. Reopen the current author guidelines and code-sharing policy before final submission.

Audit

• Map every finance claim to a table, figure, model result, identification test, or robustness check.
• For empirical work, document sample construction, variable definitions, filters, winsorizing or trimming, identifiers, timing, and economic magnitudes.
• For theory work, connect assumptions to equilibrium, comparative statics, or pricing implications that a broad finance audience can evaluate.
• Use strong benchmarks: standard asset-pricing factors, corporate-finance controls, market microstructure alternatives, or banking/household-finance baselines as relevant.
• Prepare code and data documentation early; RoF publication is conditional on receiving replication programs when the policy applies.

Finance-result stress test

For each headline result, record:

Result | Economic magnitude | Identification/model threat | Benchmark | Replication file

Then ask whether the claim would survive a top-three-finance referee:

• Does the magnitude matter economically, not only statistically?
• Is the benchmark the right one for asset pricing, corporate finance, banking, household finance, or market microstructure?
• Are timing, sample filters, and variable construction strong enough to rule out mechanical effects?
• For theory, do assumptions generate finance implications rather than only mathematical existence?
• Can the code package reproduce the exact result and the robustness check that protects it?

If the answer is weak, repair the analysis before rewriting the introduction.

European and international data discipline

RoF's EFA readership expects cross-market evidence handled with the same care US-only papers give CRSP. Known traps an RoF referee will catch:

• Datastream/Refinitiv equity returns: remove padded post-delisting observations, stale repeated prices, and non-trading-day zeros; report screen-by-screen sample counts.
• Delisting and survivorship: merge dead-firm lists back into Datastream samples and delisting returns into CRSP samples; a performance result built on survivors only is a mechanical effect, not a finding.
• Bankscope/Orbis (Bureau van Dijk): consolidated and unconsolidated statements duplicate the same banking groups — filter on consolidation codes; record the database vintage, because BvD overwrites history and a later re-pull will not rebuild your sample.
• Currency and fiscal timing: convert at observation-date exchange rates; align heterogeneous fiscal-year ends; lag accounting data enough to rule out look-ahead.
• ECB SDW, Eurostat, and national-supervisor series: store exact series codes and download dates; revisions silently change merged regressors.

Estimation defaults an RoF referee assumes

Put one defining specification in the body and route the grid of variants to the internet appendix — RoF editors prize clean identification and economic magnitudes over kitchen-sink regressions.

Worked vignette — a euro-area bank-margin paper

Illustrative numbers only. The headline: after the 2014 negative-policy-rate cut, high-deposit banks lowered lending margins 28 bp more than low-deposit banks; sample of 412 euro-area banks from Bankscope, 2010–2019.

• Magnitude: 28 bp on a 180 bp mean margin is roughly 16% — state it that way, not as a t-statistic.
• Mechanical check: consolidation duplicates double-count the largest groups; rerun on unconsolidated statements. If 28 bp falls to 9 bp, the result is a filtering artifact.
• Benchmark: contrast with matched non-euro banks to separate the negative-rate channel from the post-crisis trend; that contrast belongs in the body.
• Replication hook: the Bankscope extract cannot ship; queue the pseudo-dataset and run logs now (see rof-replication-and-data-policy).

Pipeline discipline from raw pull to exhibit

One reproducible path, because RoF can hold publication until programs arrive:

raw/        immutable vendor pulls (Datastream, Bankscope, CRSP), dated
build/      cleaning scripts: screens, delisting merges, winsorize 1/99
analysis/   one numbered script per manuscript table or figure
out/        exhibits regenerated end-to-end by run_all; diffs reviewed

• Winsorize or trim ratio variables and state the percentile in every table note.
• Fix seeds for any bootstrap or simulation inference and log them.
• Maintain a variable dictionary mapping each manuscript symbol to its construction line.

Execution bridge (StatsPAI / Stata MCP)

Run the battery, don't just enumerate it. Full map: execution-with-mcp. Review of Finance is the EFA flagship — corporate-causal chain for corporate papers, factor-zoo haircut for asset pricing.

• Many outcomes / specifications: romano_wolf (step-down FWER) or benjamini_hochberg.
• OVB sensitivity: oster_delta / sensemakr.
• Inference: wild_cluster_bootstrap (few clusters), twoway_cluster / conley.
• Re-fit off one handle: audit_result(result_id) lists missing checks + the exact suggest_function for each.
• Exhibits: etable / did_summary_to_latex from the handle — no retyped numbers.

Decisive checks in the body, exhaustive battery in the appendix. JF execution walkthrough.

Output format

[Analysis readiness] strong / adequate / weak
[Claim -> evidence] <claim: table, figure, model, or robustness>
[Top-three-standard gap] <one issue>
[Replication blocker] <data, code, pseudo-data, or logs>
[Next analysis] <single task>