Empirical Design & Inference (jfe-empirical-design)

When to trigger

• You sort on a characteristic but have not justified the variable, the breakpoints, or weighting
• You are choosing between Fama–MacBeth, panel regression, and GMM and unsure how to report
• Your standard errors are unclustered, or clustered on one dimension when two are needed
• You have an asset-pricing predictor but no out-of-sample or multiple-testing treatment
• Variable definitions are ad hoc and would not replicate

The JFE design bar

JFE is known for nuts-and-bolts methodological rigor. Referees scrutinize measurement, estimator choice, standard errors, and inference discipline line by line. The goal is a design that a skeptical expert cannot dismantle on technical grounds. This is the journal that published Fama & French (1993), "Common risk factors in the returns on stocks and bonds" (the three-factor model), Fama & French (2015), "A five-factor asset pricing model," and Banz (1981), the size effect — so an asset-pricing referee benchmarks your construction against that lineage directly. The best capital-markets paper each year wins JFE's Fama-DFA Prize; write to that standard. Code and non-proprietary data are mandatory at acceptance (Mendeley Data; see jfe-submission), so build a reproducible pipeline from the start.

Asset pricing

Factor / portfolio construction

• Justify the sorting variable economically and define it precisely (data source, lag, winsorization).
• State breakpoints (e.g., NYSE breakpoints vs. all-stock) and value- vs. equal-weighting, and show the choice does not drive the result.
• Report turnover and whether the strategy survives plausible transaction costs.

Cross-sectional inference

• Fama–MacBeth: report Newey–West / Shanken-corrected standard errors; state lags. (Note Fama-MacBeth itself originates in JPE 1973; the factor-model machinery it serves is JFE's home turf via Fama-French.)
• GMM / SDF: state moment conditions, weighting matrix, and over-identification (J-test).
• Report alphas against the Fama-French benchmarks — CAPM, FF3, FF5 — plus momentum/q-factor where relevant, and show your factor survives spanning regressions against them. A single-benchmark alpha will not satisfy a JFE asset-pricing referee.

Inference discipline

• Out-of-sample: show the predictor holds out of sample or in a holdout period.
• Multiple testing: when the predictor is one of many candidates, adjust (e.g., FDR / Bonferroni / data-mining-aware thresholds) and say so. Ignoring this is a known JFE red flag.

Corporate finance

• Define every variable with source, timing, and units; tabulate in a variable-definition table.
• Winsorize/trim consistently and state the rule; show results are not winsorization artifacts.
• Choose fixed effects deliberately (firm, industry-by-year, etc.) and justify what each absorbs.
• Standard errors: cluster at the level of correlation in the residuals (often firm and/or time); use two-way clustering when both matter; match the cluster level to treatment assignment for causal designs.
• Report economic magnitudes, not just significance — a coefficient is a number with units.

Execution bridge (StatsPAI / Stata MCP)

Run the asset-pricing battery, don't just specify it. Full map: execution-with-mcp. JFE is finance top-3 (with JF, RFS) — corporate-causal chain for corporate papers, factor-zoo haircut for asset pricing; attribute canon to the correct top-3 journal.

• Factor regressions / time-series alphas: feols with the right SEs (Newey–West / clustered) — read the alpha and t off the return.
• Factor-zoo haircut: after disclosing how many signals were screened, apply romano_wolf / benjamini_hochberg and report the alpha that survives.
• Fama–MacBeth + Shanken EIV are Stata-canonical — run via mcp__stata-mcp__stata_do with the vendored resources/code/ (asreg / xtfmb).
• Exhibits: etable; hand formatting to the tables/figures skill.

Report the economic magnitude (bps/month alpha, Sharpe gain); full factor grid → appendix. JF execution walkthrough.

Checklist

• Every variable is defined with source, lag, and transformation
• Winsorization/trimming rule stated and shown not to drive results
• Sorting/breakpoint/weighting choices justified and stress-tested
• Estimator matches the question (FM / GMM / panel FE) and is reported correctly
• Standard errors clustered/corrected appropriately (often two-way)
• Alphas reported against multiple benchmark models
• Out-of-sample and multiple-testing discipline applied for predictors
• Economic magnitudes interpreted, not just t-stats

Anti-patterns

• A new factor reported only in-sample, with no multiple-testing acknowledgment
• Standard errors that ignore cross-sectional or time-series correlation
• Breakpoints/weighting cherry-picked to maximize the spread
• Fama–MacBeth without Newey–West or Shanken corrections
• Variable definitions too vague to replicate
• Reporting t-statistics while never stating the economic size of the effect

Output format

【Field】asset pricing | corporate finance
【Estimator】FM / GMM / panel FE / portfolio sort
【SE treatment】cluster dims / NW lags / Shanken
【Benchmarks】[models alphas are measured against]
【Inference discipline】out-of-sample? multiple-testing adjusted?
【Magnitudes stated】yes/no
【Next】jfe-robustness