Imfer Robustness

Robustness Suite (imfer-robustness)

When to trigger

• The main estimate is in hand and you must show it is not an artifact of one specification
• A referee asks "is this driven by a few countries / a particular crisis episode / the sample window?"
• The result depends on a sample-selection rule, a country grouping, or a deflator/exchange-rate convention
• You suspect a global common shock or one influential economy is doing the work
• You want to pre-empt specification-search and country-cherry-picking concerns

The IMFER robustness bar

IMFER referees probe whether the headline number is stable, honestly inferred, and not the product of country-composition or specification search — and, distinctively, whether it would still guide policy under reasonable alternatives. Robustness here is a targeted set of checks, each tied to a specific threat to an international-macro design, not a wall of regressions. The persuasive object is that the point estimate barely moves when you drop the obvious country, switch the window, or change the inference.

Inference under cross-country dependence (the common slip)

Standard errors are where IMFER robustness most often fails, because country panels violate the assumptions behind default clustering. Flows, spreads, and prices are cross-sectionally dependent (a global shock hits everyone), so country-clustered SEs alone understate uncertainty — use Driscoll–Kraay or a two-way (country and time) cluster. The country dimension is usually small (30–60 economies), so cluster-robust asymptotics are unreliable — report a wild-cluster bootstrap. Serial correlation within country is the norm — do not assume i.i.d. errors. State which of these you address and why; a referee who suspects the SEs are too tight will discount the whole result.

Robustness craft

1. Lock the primary specification first. Everything else perturbs it; do not present five co-equal panels and let the referee guess the preferred one.
2. Lead with the country-composition checks. Leave-one-country-out and drop-the-dominant-economy are the first things an IMFER referee runs in their head — show them in the main paper.
3. Stress the global common shock. Demonstrate the effect is not the global financial cycle masquerading as your policy variable.
4. Match inference to cross-country structure. Driscoll–Kraay or two-way clustering, and wild-cluster bootstrap when the country count is small — wrong SEs are the most common IMFER robustness failure.
5. Report stability, not just significance. Show the point estimate range across checks; a check that moves it is information — bound the implication for policy rather than hiding it.

Execution bridge (StatsPAI / Stata MCP)

Run the battery, don't just enumerate it. Full map: execution-with-mcp. IMFER is international macro/finance; cross-country panels with confounded policy — emphasize identification and clustering.

• Many outcomes / specifications: romano_wolf (step-down FWER) or benjamini_hochberg.
• OVB sensitivity: oster_delta / sensemakr.
• Inference: wild_cluster_bootstrap (few clusters), twoway_cluster / conley.
• Re-fit off one handle: audit_result(result_id) lists missing checks + the exact suggest_function for each.
• Exhibits: etable / did_summary_to_latex from the handle — no retyped numbers.

Decisive checks in the body, exhaustive battery in the appendix. JF execution walkthrough.