jfqa-identification-strategy

JFQA Identification Strategy (jfqa-identification-strategy)

Use this skill to make the research design defensible for JFQA, an empirical and quantitative finance journal. JFQA referees press hard on whether a correlation is causal (or, in asset pricing, whether a premium is robust and not data-mined).

Empirical finance designs (the common case)

Pick the design that matches the question and defend it:

• Cross-section of returns — portfolio sorts and Fama-MacBeth regressions; Newey-West / clustered SEs; control for standard factors; report economic magnitudes (return per one-SD change), not only t-stats; guard against data snooping (out-of-sample, multiple-testing awareness).
• Corporate finance panels — firm and time fixed effects, two-way clustering; show the variation that identifies the coefficient.
• Policy / regulatory shocks — staggered DID with a modern estimator (Callaway-Sant'Anna, de Chaisemartin-D'Haultfœuille), event-study leads/lags, and parallel-trends evidence; avoid naive TWFE on staggered timing.
• Natural experiments / IV — instrument relevance (first-stage F), exclusion logic backed by an economic story, weak-IV-robust CIs.
• Thresholds / index reconstitution — RDD with manipulation/density tests and bandwidth robustness.
• Announcements — event study with CARs/BHARs, a defensible market model, and attention to calendar clustering.

What referees demand

• The right standard errors (clustering dimension justified, two-way where needed).
• Economic significance reported alongside statistical significance.
• Endogeneity confronted explicitly, not waved away with "controls."

Theoretical submissions

JFQA also publishes theory. If your paper is a model, pivot this skill to: stating assumptions transparently, deriving results/propositions, clean proof exposition, and testable implications a finance reader can take to data. Keep generality matched to the question.

Threat-to-remedy matrix for the JFQA referee report

| Endogeneity threat | How it surfaces in the draft | JFQA-grade remedy |

|---|---|---|

| Reverse causality | outcome plausibly drives the regressor | timing structure, a shock that moves only the regressor, or an IV with an economic exclusion story |

| Omitted firm-level variation | "we control for size and B/M" | firm FE plus a within-firm variation count showing the coefficient is still identified |

| Selection into treatment | treated and control firms differ pre-event | matching or entropy balancing plus pre-trend evidence, not either alone |

| Anticipation of regulation | effects appear before adoption | shift the event date, drop the anticipation window, show announcement-date returns |

| Data-mined anomaly | one sort, one sample, large t-stat | sub-period splits, out-of-sample evidence, multiple-testing discussion |

| Bad controls | post-treatment variables on the RHS | re-specify; report with and without, and explain which is the estimand |

Worked vignette: staggered adoption done the JFQA way (illustrative)

Suppose 23 states adopt a disclosure rule between 2008 and 2016 and the outcome is the credit spread of in-state issuers. A naive TWFE regression gives -4.1%; the Callaway-Sant'Anna group-time ATT gives -2.6% because late-vs-already-treated comparisons inflated the TWFE number. The JFQA presentation: CS estimator as the headline, TWFE relegated to the appendix with the discrepancy explained, an event-study figure whose lead coefficients are jointly insignificant (p = 0.42), and — with only 23 clusters — wild cluster bootstrap inference (p = 0.03) instead of leaning on asymptotics. That package answers the three referee questions (estimator, pre-trends, inference) before they are asked.

The anomaly-credibility bar in asset pricing

• Acknowledge the multiple-testing problem head-on: the post-2016 factor-zoo literature argues for materially higher t-hurdles for new predictors; state how many specifications were examined.
• Show tradability: turnover, transaction-cost drag, and whether the premium survives value-weighting and the exclusion of microcaps.
• Run spanning tests against the standard factor models in current use; a new "factor" that the existing ones price is a robustness row, not a contribution.

Execution bridge (StatsPAI / Stata MCP)

Estimate and audit the identification claim, don't only argue it. Full map:

[execution-with-mcp](../../../shared-resources/empirical-methods/execution-with-mcp.md). JFQA is empirical finance (asset pricing + corporate) — the DiD / IV / RDD chain for corporate causal claims, the factor-zoo haircut for cross-sectional pricing.

1. detect_design → recommend → fit with as_handle=true → audit_result to list

the checks the design still owes.

1. Staggered DiD: callaway_santanna / sun_abraham + bacon_decomposition +

honest_did_from_result (the pre-trend test is low-power, Roth 2022).

1. IV: effective_f_test + an anderson_rubin_ci (valid under weak instruments),

not a 2SLS t-stat alone.

1. RDD: rdrobust (bias-corrected) + rddensity / mccrary_test for manipulation.
2. OVB: oster_delta / sensemakr — how strong a confounder would have to be.

Report the economic magnitude; route the full battery to the appendix; keep every

number reproducible. A run end-to-end (synthetic data, real returns) is in the

[JF execution walkthrough](../../../Journal-of-Finance-Skills/resources/worked-examples/02-execution-walkthrough.md). If StatsPAI/Stata are not connected, adapt the

vendored resources/code/ skeleton and flag any unverified number.

Output format

【Design】sorts/FMB / panel FE / staggered DID / IV / RDD / event study / theory
【Identifying variation】what makes it credible
【Inference】clustering / weak-IV / multiple-testing handling
【Economic magnitude】effect size in finance units
【Next step】jfqa-data-analysis