Causal Identification (jf-identification)

When to trigger

• The paper makes a causal claim ("X causes Y") resting on a research design
• You rely on an instrument, a shock, a discontinuity, or a diff-in-diff and a referee will attack the exclusion/parallel-trends assumption
• Endogeneity (reverse causality, omitted variables, selection) threatens the headline result

Scope: corporate / empirical causal effects. For cross-sectional asset-pricing tests use jf-empirical-design.

JF's bar for identification

JF is the AFA flagship, general-interest, with a ~5% acceptance rate and ~33–45% desk rejection (afajof.org editor reports, accessed 2026-05-30). For a corporate/empirical paper, credible identification is usually the binding constraint — a clever question with a weak design is a classic JF desk reject. The design must convince a broad AFA readership, not just specialists.

Design audit

• State the source of variation in one sentence in the introduction (JF rewards a clearly named shock or instrument).
• Show the identifying assumption is testable where possible (pre-trends, first-stage F, McCrary test) and put the full battery in the Internet Appendix (bundled in the same PDF; see jf-internet-appendix).
• Report economic magnitude, since JF writes for a general-interest reader.

Worked vignette — a staggered-regulation natural experiment

Illustrative numbers. A paper claims a disclosure regulation, rolled out across states in 2011–2016, causes treated firms to cut leverage; the DID shows book leverage falling 4.2 pp (t = 3.4). Walk it through JF's bar:

1. Name the variation in one sentence: "Staggered state-level adoption of Rule X gives treated firms a plausibly exogenous shock to disclosure costs" — the introduction's credibility hook for a broad-readership editor.
2. Defend the assumption: an event-study plot shows flat pre-trends before adoption; the full coefficient panel goes to the Internet Appendix.
3. Fix the staggered-adoption bias: a naive two-way fixed-effects estimate (4.2 pp) is contaminated by already-treated controls. Re-estimate with a modern estimator (Callaway–Sant'Anna or Sun–Abraham); the clean estimate lands at ~3.1 pp — report it and flag the TWFE bias.
4. Pre-empt anticipation: show no effect in the year before the law as a placebo.
5. Report magnitude: 3.1 pp on a ~30% mean is a ~10% relative move — say so, since JF prizes economically large effects over bare significance.

The editor sees a named shock, a defended assumption, the right estimator, and a magnitude that matters to the AFA readership.

Referee-pushback patterns and the JF-specific fix

Calibration anchors for JF identification

• For a corporate/empirical paper, identification is typically the binding constraint: a first-order question with a fragile design is a classic flagship desk reject, while a less novel question with airtight identification can survive.
• JF expects the identification battery visible but not bloating the body — pre-trends, first-stage F, McCrary density, balance tables go to the Internet Appendix, with one or two decisive plots in the main text.
• Weak-instrument and modern-DID standards evolve; confirm the expected diagnostics against recent issues and current author guidelines.

Execution bridge (StatsPAI / Stata MCP)

Do not stop at advising the right estimator — run it and report the number. Full map: shared-resources/empirical-methods/execution-with-mcp.md. JF-specific instantiation:

1. detect_design → preflight → recommend on the data; fit with as_handle=true.
2. Staggered DiD: estimate with callaway_santanna / sun_abraham (not bare TWFE); run bacon_decomposition to expose the bad-comparison weight you are correcting — this is the "TWFE is biased" pre-emption, executed. Put the clean estimate in the body; the event-study/pre-trend panel goes to the Internet Appendix.
3. IV: report effective_f_test and an anderson_rubin_ci (weak-IV-robust), not a 2SLS t-stat alone.
4. RDD: rdrobust for the bias-corrected estimate; rddensity/mccrary_test and bandwidth sweep (rdbwselect) → Internet Appendix; one density/RD plot in the body.
5. audit_result(result_id) to enumerate what the design still owes; then honest_did_from_result to bound a pre-trend violation. Cite methods only via bibtex.

The JF body shows one or two decisive exhibits with the economic magnitude; the full diagnostic battery lives in the bundled Internet Appendix (see jf-internet-appendix). If StatsPAI/Stata are not connected, adapt the vendored resources/code/ skeleton and say which number is unverified.

See this run end-to-end on synthetic data — every number an actual tool return — in resources/worked-examples/02-execution-walkthrough.md (TWFE −0.0227 vs clean CS −0.0272, pre-trends p = 0.155, honest-DiD breakdown point).

Checklist

• Source of identifying variation named in one sentence
• Exclusion / parallel-trends / continuity assumption explicitly defended
• First-stage strength (IV) or pre-trend evidence (DID) shown
• Modern estimators used where staggered adoption applies
• Confounders and anticipation effects addressed
• Magnitude interpreted, not just significance

Anti-patterns

• A causal verb ("increases", "causes") with only conditional correlations behind it
• An instrument with a hand-waved exclusion restriction
• Two-way fixed-effects DID on staggered adoption with no modern correction
• A clever question whose design no broad-readership editor would send out

Output format

【Design】NE / IV / DID / RDD
【Source of variation (1 sentence)】...
【Key assumption + how defended】...
【Main threat pre-empted?】yes / no
【Magnitude】...
【Next step】jf-robustness