jimf-empirical-design

Empirical Design (jimf-empirical-design)

When to trigger

• A cross-country panel is unbalanced, mixes incompatible series, or pools regimes that should be separated
• Frequency and alignment are unclear (daily FX vs. monthly flows vs. quarterly macro) and the design glosses the mismatch
• Key variables (exchange rate, capital flows, sovereign spread, pass-through) are measured in a way a referee will dispute
• Country coverage, the advanced-vs-emerging split, or sample period drives the result and is not justified
• Standard data quirks (USD vs. trade-weighted FX, gross vs. net flows, BoP vs. EPFR, nominal vs. real) are not pinned down

The JIMF data-design bar

International-finance referees scrutinize measurement and comparability as hard as identification, because cross-country data are heterogeneous and easy to mis-align. Three recurring fault lines: (1) which series — there are several defensible measures of every JIMF object, and the choice matters; (2) which countries and period — advanced vs. emerging, pre- vs. post-GFC, in-vs-out of a crisis window; (3) what frequency and alignment — mixing frequencies without saying how. Make each explicit and defend it before the result.

| JIMF object | Measurement choices to declare | Common referee objection |

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

| Exchange rate | bilateral USD vs. NEER/REER; nominal vs. real; end-of-period vs. average | "Your result is a dollar effect, not an exchange-rate effect" |

| Capital flows | gross vs. net; BoP (quarterly) vs. EPFR (high-frequency fund flows); by instrument (debt/equity/bank) | "EPFR is fund flows, not balance-of-payments flows" |

| Pass-through | import prices vs. CPI; aggregate vs. invoicing-currency level; horizon of pass-through | "Aggregate ERPT hides the dominant-currency margin" |

| Sovereign risk | CDS vs. EMBI/bond spread vs. rating; local- vs. foreign-currency debt | "LC and FC sovereign risk are different objects" |

| Global financial cycle | VIX vs. a factor (Miranda-Agrippino–Rey) vs. US shadow rate | "VIX is a proxy, not the GFCy" |

| Monetary stance | policy rate vs. shadow rate vs. surprise; domestic vs. foreign | "The ZLB period breaks your policy-rate measure" |

Design moves that read as JIMF-competent

1. Declare the country set and the split. State advanced vs. emerging, why each country is in, and report results separately if the mechanism differs — pooling AE and EM without a test invites rejection.
2. Anchor the sample period to the institutional history. Bretton-Woods break, euro introduction, GFC, ZLB, taper tantrum, COVID — say which regimes your window spans and whether you split them.
3. Resolve the frequency mismatch explicitly. If the shock is daily and the outcome quarterly, state the aggregation; if you use mixed frequency, justify it (MIDAS / local projections at the native frequency).
4. Defend the exchange-rate convention. Distinguish a dollar effect from a general exchange-rate effect; report NEER/REER alongside USD when the claim is about the exchange rate per se.
5. Document sources to the series level. BIS, IMF IFS/BoP, IMF AREAER (capital-account openness), EPFR, Datastream/Bloomberg, Lane–Milesi-Ferretti external positions, Ilzetzki–Reinhart–Rogoff regime classification — name the exact vintage and any splicing.

Execution bridge (StatsPAI / Stata MCP)

Run the asset-pricing battery, don't just specify it. Full map:

[execution-with-mcp](../../../shared-resources/empirical-methods/execution-with-mcp.md). JIMF is international macro-finance; cross-country panels + asset pricing — identification plus factor/Newey-West inference.

• Factor regressions / time-series alphas: feols with the right SEs (Newey–West /

clustered) — read the alpha and t off the return.

• Factor-zoo haircut: after disclosing how many signals were screened, apply

romano_wolf / benjamini_hochberg and report the alpha that survives.

• Fama–MacBeth + Shanken EIV are Stata-canonical — run via mcp__stata-mcp__stata_do

with the vendored resources/code/ (asreg / xtfmb).

• Exhibits: etable; hand formatting to the tables/figures skill.

Report the economic magnitude (bps/month alpha, Sharpe gain); full factor grid → appendix.

[JF execution walkthrough](../../../Journal-of-Finance-Skills/resources/worked-examples/02-execution-walkthrough.md).

Checklist

• [ ] Country set justified; AE/EM split reported or its absence defended
• [ ] Sample period mapped to international monetary history; regime breaks handled (split or controlled)
• [ ] Each JIMF object measured with a declared, defended choice (USD vs. NEER, gross vs. net, CDS vs. spread)
• [ ] Frequency alignment stated; mixed-frequency method named if used
• [ ] Data sources named to the series and vintage; splicing/cleaning documented for audit
• [ ] A dollar effect is not mislabeled as an exchange-rate effect (and vice versa)
• [ ] Sample-construction steps reproducible enough for the online appendix and Mendeley Data deposit

Anti-patterns

• Pooling advanced and emerging economies with no test for whether the mechanism is common
• Using EPFR fund flows and calling them balance-of-payments capital flows (or vice versa) without flagging the difference
• A "global financial cycle" measured only by VIX with no acknowledgment it is a proxy
• An unbalanced panel where entry/exit of countries correlates with the outcome (crisis countries dropping out)
• Spanning the GFC and ZLB with a single policy-rate measure and no regime control
• Hiding the exchange-rate convention so a dollar-specific result reads as a general exchange-rate result

Worked vignette (illustrative)

A draft regresses quarterly EM "capital flows" on a daily US surprise and finds a strong effect, but the flows are EPFR weekly fund flows aggregated to quarters and the panel drops three countries during their crises. The JIMF fix: state that EPFR captures benchmarked-fund flows (a leading indicator), not BoP flows, and either align the analysis at the native weekly frequency via local projections or report both; keep the crisis countries in with a balanced-panel robustness; and report whether the effect is a dollar phenomenon (USD bilateral) or survives in NEER terms.

Referee pushback mapped to the design fix

• "That's a dollar effect, not an exchange-rate effect." → Report NEER/REER alongside the USD bilateral; show whether the result is dollar-specific or holds for the effective rate.
• "EPFR is not balance-of-payments capital flows." → State what EPFR measures (benchmarked-fund flows, a high-frequency leading indicator) and either align at its native frequency or triangulate with BoP data.
• "Your panel is unbalanced in a way that correlates with the outcome." → Show a balanced-panel robustness; report whether crisis-driven entry/exit moves the estimate.
• "You pooled advanced and emerging economies." → Split AE/EM and test whether the mechanism is common before pooling.
• "The ZLB breaks your policy-rate measure." → Use a shadow rate or a monetary surprise over the affected window; show the result is not an artifact of the policy-rate floor.

A note on sources international-finance referees trust

Name the canonical datasets and their roles so the design reads as field-literate: IMF IFS/BoP for macro and flows; BIS for cross-border banking and FX statistics; IMF AREAER and the Chinn–Ito index for capital-account openness; the Ilzetzki–Reinhart–Rogoff classification for de facto exchange-rate regimes; Lane–Milesi-Ferretti for external positions; EPFR for high-frequency fund flows; Datastream/Bloomberg for prices, CDS, and yields. Using the wrong dataset for an object (e.g. a de jure regime classification when the question is de facto behavior) is a credibility tell referees catch quickly.

Pre-analysis design questions to settle first

Before estimating, lock five decisions and write the one-line justification for each — they are the questions a referee asks before reading Table 1:

1. Outcome and treatment frequency — at what frequency is each measured, and how are they aligned (aggregation, MIDAS, native-frequency local projections)?
2. Country set and entry/exit rule — which countries, why, and what happens to crisis-driven gaps in the panel.
3. Regime/period partition — which exchange-rate or policy regimes the window spans, and whether you split or control for the breaks.
4. The exchange-rate convention — USD bilateral, NEER, or REER, nominal or real, and whether the claim is about the dollar or the effective rate.
5. The flow/risk object — gross vs. net, BoP vs. fund flows, CDS vs. spread, and which the mechanism actually predicts.

Settling these up front prevents the most common revision loop, where a measurement choice the authors never justified turns out to drive the headline result.

Output format

【Journal】Journal of International Money and Finance
【Skill】jimf-empirical-design
【Country set / split】AE / EM / both → justified? [Y/N]
【Sample period / regimes】window + breaks handled
【Key measures declared】FX convention / flows type / risk measure / GFCy proxy
【Frequency】native / aggregated / mixed-frequency method
【Data sources】named to series + vintage; splicing documented? [Y/N]
【Next skill】jimf-robustness