jfi-identification-strategy
GitHub用于审计JFI论文核心分析引擎的Skill。实证方面压力测试因果设计，分离信贷供需；理论方面审查假设、均衡纪律及证明。不执行分析，仅评估设计严谨性，涵盖现代估计量、银行数据威胁及Khwa-Mian基准局限。
Trigger Scenarios
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill jfi-identification-strategy -g -y
SKILL.md
Frontmatter
{
"name": "jfi-identification-strategy",
"description": "Use when auditing the core analytical engine of a Journal of Financial Intermediation (JFI) paper — for empirics, the causal design that separates credit supply from demand in banking data; for theory, the assumptions, equilibrium discipline, and proof exposition. It pressure-tests the design; it does not run the analysis."
}
Identification Strategy (jfi-identification-strategy)
When to trigger
- Setting up or defending the empirical design of a banking/intermediation paper
- Setting up or defending the assumptions and propositions of a theory paper
Empirical track (applied banking / credit)
JFI referees are unforgiving on identification in bank data. Build a credible causal design and defend it:
- Source of variation: a regulatory change, supervisory shock, branching deregulation, a discontinuity in capital/eligibility rules, or a plausibly exogenous credit-supply shifter.
- Modern estimators: staggered DID with heterogeneity-robust estimators (Callaway–Sant'Anna, Sun–Abraham, de Chaisemartin–D'Haultfœuille), IV with weak-IV-robust inference, or RDD with the rdrobust toolkit.
- Bank-data-specific threats: bank selection into treatment, borrower–firm sorting, balance-sheet timing and mechanical reverse causality, and the lending-channel separation of credit supply from demand (firm×time fixed effects in matched lender–borrower panels).
- Inference: cluster at the level of treatment assignment (often bank or market); wild-cluster bootstrap when clusters are few.
Theory track (intermediation models)
When the contribution is a model, identification means analytical discipline:
- State assumptions transparently and motivate each economically (what friction it encodes).
- Make results precise as propositions/lemmas; keep proof exposition readable — sketch the mechanism in the text, full proofs in an appendix.
- Argue generality: which results survive relaxed assumptions, and where the boundary lies.
- A numerical example (see jfi-data-analysis) can illustrate the mechanism without claiming empirical estimation.
The within-firm benchmark, and when it is not enough
The Khwaja–Mian within-firm estimator is this community's default answer to demand confounds: with multi-bank firms, firm×time fixed effects difference out borrower demand and isolate the credit-supply channel. A JFI referee then pushes past the default:
- Multi-bank firms are larger and less bank-dependent — show what the design's external margin (single-relationship firms) does, or bound how far the within-firm estimate travels.
- "Equal demand across a firm's lenders" is itself an assumption: a firm may cut demand for one bank's specialized product. Address with loan-purpose controls or product-level fixed effects.
- Firm-level real outcomes cannot carry firm×time FE; aggregate the bank shock to the firm with pre-period exposure shares, and defend share exogeneity as in shift-share designs.
Design selection for common intermediation shocks
| Variation exploited | Default design | Venue-specific threat to pre-empt |
|---|---|---|
| Staggered regulation/deregulation across states or countries | Heterogeneity-robust staggered DID | Banks lobby for timing — show treatment is not predicted by pre-trend bank health |
| Capital- or size-threshold rule | RDD with density test | Banks bunch by managing the ratio; McCrary check is mandatory |
| Funding or deposit shock with differential exposure | Exposure (shift-share) design | Exposure shares correlate with local demand — balance on borrower observables |
| Run or crisis window | High-frequency event design | Mechanical balance-sheet timing; reverse causality from borrower distress |
Worked contrast: one estimate, two readings (illustrative)
A 1pp funding shock reduces bank-level lending by 2.8pp (bank panel, OLS). At JFI that is not yet a result: the same number is consistent with shocked banks happening to serve shocked borrowers. The within-firm version at 1.6pp (firm×time FE) is the publishable object — and the 1.2pp gap becomes evidence on borrower–bank sorting worth its own paragraph, not a nuisance to hide. JFI referees read the movement of the coefficient across fixed-effect columns as a diagnostic in itself; design the identification section so that movement is interpreted, not merely displayed.
Execution bridge (StatsPAI / Stata MCP)
Estimate and audit the identification claim, don't only argue it. Full map:
execution-with-mcp. JFI is banking and financial intermediation — typically corporate / bank causal designs built around regulation and shocks.
detect_design→recommend→ fit withas_handle=true→audit_resultto list the checks the design still owes.- Staggered DiD:
callaway_santanna/sun_abraham+bacon_decomposition+honest_did_from_result(the pre-trend test is low-power, Roth 2022). - IV:
effective_f_test+ ananderson_rubin_ci(valid under weak instruments), not a 2SLS t-stat alone. - RDD:
rdrobust(bias-corrected) +rddensity/mccrary_testfor manipulation. - 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. If StatsPAI/Stata are not connected, adapt the
vendored resources/code/ skeleton and flag any unverified number.
Anti-patterns
- OLS-plus-controls dressed up as identification on a bank panel
- Conflating credit supply and demand without firm×time absorption
- A theory whose key result silently depends on an unstated assumption
- Clustering at the wrong level, or ignoring few-cluster inference
Output format
【Track】empirical / theory
【Design or assumptions】<the variation, or the key assumptions>
【Top threat / boundary】<the main objection + answer>
【Inference / generality】<clustering, or which results survive>
【Next skill】jfi-data-analysis
Version History
- 1839142 Current 2026-07-05 13:39
