| name | imfer-identification |
| description | Use when the identification argument is the bottleneck for an IMF Economic Review (IMFER) manuscript — cross-country panel, high-frequency policy-surprise, crisis event study, narrative, or open-economy structural identification. Stress-tests the data-to-object mapping to IMFER's policy-relevant bar; it does not build the model (imfer-theory-model) or run the robustness suite (imfer-robustness). |
Identification Strategy (imfer-identification)
When to trigger
- A cross-country causal claim rests on OLS-plus-controls or TWFE on staggered policy adoption
- A high-frequency policy-surprise series (monetary, FX intervention) needs its exclusion defended
- A crisis "event study" cannot separate the policy from the macro shock that triggered it
- A capital-control or program effect is plausibly endogenous to the crisis it is meant to address
- An open-economy model's parameters are estimated but it is unclear what in the data identifies them
The IMFER identification bar
IMFER referees read as both frontier econometricians and policy analysts, so the mapping from data to the policy-relevant object must be explicit and the object must be the one a policymaker cares about. International-macro data make this harder than a clean single-country RCT: small N of countries, endogenous policy adoption, global common shocks, and spillovers that violate SUTVA across borders. Name the variation, defend it against the macro confounder, and report inference that respects cross-country dependence.
Branch A: Cross-country panel
- Move beyond TWFE under staggered policy adoption: Callaway–Sant'Anna, Sun–Abraham, de Chaisemartin–D'Haultfœuille; show clean event-study leads.
- Global common shocks (the global financial cycle, US monetary policy) confound country panels — saturate with time effects, or use a shift-share / external-instrument design.
- Spillovers break SUTVA: a control imposed in one country affects its neighbors. State the interference assumption or model the network explicitly.
- Inference: cluster on country and allow cross-sectional dependence (Driscoll–Kraay); with few countries, wild-cluster bootstrap.
Branch B: High-frequency policy-surprise
- Build the surprise from a tight window around announcements (intraday futures/yields); show it is unpredictable from the prior information set.
- Defend the exclusion: the surprise moves the outcome only through the policy channel, not contemporaneous news. Use a poll-based or sign-restriction purge if needed.
- For spillover designs, identify the foreign transmission (US-shock-to-EM-flows) and rule out the domestic-news confounder.
Branch C: Crisis event study / narrative
- Define the event window and the counterfactual path; the trigger shock and the policy response are nearly simultaneous — argue what isolates the policy.
- Use narrative classification (IMF program dates, intervention episodes) with documented coding rules; report sensitivity to window length and event definition.
- Address selection into crisis/program: countries adopt programs precisely when fundamentals deteriorate.
Branch D: Open-economy structural / DSGE
- Name what identifies each parameter from a data moment (an impulse response, a comovement, an external-finance premium) — not "the estimator converged."
- Report a sensitivity matrix (which moment moves which parameter) and Monte Carlo recovery of known parameters.
- Argue policy-invariance for the counterfactual (Lucas critique), since IMFER counterfactuals are read as policy advice.
Execution bridge (StatsPAI / Stata MCP)
Estimate and audit the design, don't only describe it. Full map:
execution-with-mcp. IMFER is international macro/finance; cross-country panels with confounded policy — emphasize identification and clustering.
detect_design → recommend → fit with as_handle=true → audit_result.- Observational causal claims: staggered DiD (
callaway_santanna / sun_abraham +
bacon_decomposition + honest_did_from_result); IV (effective_f_test +
anderson_rubin_ci); RDD (rdrobust + mccrary_test).
- Experiments: randomization-based inference +
romano_wolf for many-outcome control.
- Sensitivity:
oster_delta / sensemakr for observational claims.
Report the magnitude in interpretable units; route the full battery to the appendix. A
run end-to-end (synthetic data, real returns) is in the JF execution walkthrough.
Checklist
Anti-patterns
- TWFE on staggered capital-control / program adoption with no heterogeneity-bias discussion
- A country panel that ignores the global financial cycle / US-policy common shock
- Treating a CFM or IMF program as exogenous when it is adopted in response to the crisis
- A "policy surprise" still predictable from prior macro news
- Clustering only on time, or ignoring spatial/cross-country dependence
- "The model converged" presented as structural identification
The international-macro confounders to name explicitly
International-macro identification fails in characteristic ways the referee pool knows by heart; name the ones in play.
- The global financial cycle. A common push factor (US monetary policy, global risk appetite) drives flows, spreads, and prices in many countries at once. Time effects help only if the loading is homogeneous; otherwise use an explicit factor or external instrument.
- Reverse causality with the crisis. CFMs, interventions, and IMF programs are adopted because conditions deteriorated; the policy and the outcome share a cause.
- Cross-border spillovers (SUTVA). A control unit is contaminated by the treated country's policy (contagion, portfolio rebalancing); the "untreated" counterfactual is not clean.
- Anticipation. Markets price an expected regime change before the announcement, contaminating pre-periods.
- Small N of countries. Asymptotics in the country dimension are unreliable; report finite-sample-robust inference.
Worked vignette (illustrative)
A panel claims capital-flow-management measures cut the volatility of inflows. A weak version runs TWFE with country and year effects. An IMFER version: (i) uses Callaway–Sant'Anna for staggered adoption with flat pre-trends; (ii) instruments adoption with a regional-contagion shift-share to break endogeneity to the country's own crisis; (iii) saturates with the global financial cycle (a VIX/US-shock factor) to kill the common shock; (iv) reports Driscoll–Kraay SEs for cross-country dependence. The estimand — the effect on the adopting country's inflow volatility — is stated, and the spillover to neighbors is flagged as a separate object.
Referee pushback mapped to the identification fix
- "Staggered TWFE is biased here." → Re-estimate with Callaway–Sant'Anna or Sun–Abraham; show flat event-study leads.
- "This is the global financial cycle, not your policy." → Add the common-shock factor / US-shock control and show the result survives.
- "The policy is endogenous to the crisis." → Bring an external instrument or narrative timing; defend selection-into-program.
Output format
【Journal】IMF Economic Review
【Skill】imfer-identification
【Branch】cross-country panel / policy-surprise / crisis event / structural
【Data-to-policy-object mapping】one sentence: ___
【Common-shock / spillover handling】___
【Selection / endogeneity defense】___
【Inference】clustering + cross-sectional dependence; SEs not asterisks: ___
【What it does NOT identify】___
【Next skill】imfer-theory-model
