Data Analysis (jie-data-analysis)
When to trigger
- Assembling a bilateral trade panel or a cross-country macro/finance panel
- Choosing the gravity estimator and fixed-effect structure
- Deciding which robustness checks an international-economics referee will demand
- Disciplining a structural open-economy or trade model against the data
Empirical norms at JIE
JIE referees come from international trade or international macro/finance and apply that field's conventions. Match them.
Data construction
-
Trade: build bilateral product-level flows from UN Comtrade or the cleaned BACI (CEPII); merge gravity covariates from the CEPII Gravity database (distance, contiguity, common language, RTAs); tariffs/NTMs from WITS/TRAINS or WTO. Keep zeros — they are informative and PPML uses them.
-
International macro/finance: assemble panels from IMF IFS/BOP/DOTS, BIS, the External Wealth of Nations (Lane–Milesi-Ferretti), Penn World Table; sovereign-debt/default histories from established crisis datasets.
Estimation
-
Gravity → PPML (
ppmlhdfe, fixest::fepois) with importer×time, exporter×time, and pair fixed effects; report the trade elasticity.
-
Cross-country macro panels → fixed effects with care for dynamic-panel bias (
xtabond2/xtdpdgmm where lagged dependent variables matter); cluster appropriately.
-
Dynamics → local projections (
lpirf) for pass-through and impulse responses; VAR/BVAR where warranted.
-
Structural → solve and calibrate/estimate (Dynare for DSGE; exact-hat algebra for trade counterfactuals); report targeted and untargeted moments.
Robustness suite (referees will ask)
- Alternative fixed-effect structures and samples; alternative trade-cost proxies.
- Sensitivity to the trade elasticity (and, for macro, to risk aversion / discount factor).
- Modern DID estimators if any staggered policy timing is involved.
- Design-based inference for shift-share exposure; multi-way clustering for dyadic data.
- Counterfactual robustness to the elasticities that cannot be pinned down precisely.
Execution bridge (StatsPAI / Stata MCP)
Run the battery, don't just enumerate it. Full map:
execution-with-mcp. JIE is international trade/macro; cross-country/firm panels — 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.
Checklist
Anti-patterns
- Log-OLS gravity dropping zeros (biased per Santos Silva–Tenreyro)
- Single-clustered standard errors on dyadic trade data
- A structural model matched to one moment and declared validated
- Reporting stars without the trade/welfare/pass-through magnitude
- Ad hoc analysis with no script trail, breaking the mandatory replication deposit later
Referee-pushback patterns and the JIE-specific fix
The objections an international-economics referee raises most, with the move that resolves each. Each maps to a concrete change in the empirical section.
- "Gravity estimates without multilateral resistance / still on log-OLS." Fix: re-estimate with PPML (
ppmlhdfe/fepois), importer×time and exporter×time fixed effects, pair fixed effects for policy effects; keep zeros; report the elasticity, not just significance.
- "Trade-policy variable is endogenous — RTAs form where trade is already growing." Fix: pair fixed effects to absorb time-invariant selection, pre-trend event study, phase-in/anticipation modelled, and a falsification on flows the policy should not touch.
- "Shift-share exposure may proxy a local trend." Fix: state whether identification rests on exogenous shares or shifts, use Adao–Kolesar–Morales or Borusyak–Hull–Jaravel inference, and show exposure is uncorrelated with pre-period trends.
- "Quantitative model is not disciplined by the data." Fix: tabulate targeted vs untargeted moments, report counterfactuals under a range of trade elasticities, and show the headline result is not an artifact of one un-pinned-down parameter.
- "Inference ignores dyadic dependence." Fix: multi-way (exporter and importer) clustering, not a single cluster dimension.
Worked micro-example (illustrative numbers)
A gravity panel of bilateral manufacturing trade, 60 countries × 20 years, asking whether an RTA raises members' trade. Log-OLS on positive flows returns an RTA coefficient of, say, 0.55 (about a 73% increase). Re-estimating in PPML with importer×time, exporter×time, and pair fixed effects — retaining the ~30% of dyad-years with zero trade — the coefficient falls to roughly 0.20 (about 22%). That drop is the textbook Santos Silva–Tenreyro pattern: log-OLS over-weighted small positive flows and discarded informative zeros. Multi-way clustering on exporter and importer widens the standard error enough that the honest takeaway is "a modest, precisely-bounded RTA effect," not the inflated log-OLS number. (Figures illustrative, chosen to show the mechanism.)
Output format
【Data】sources + unit (dyad-year / country-year / firm-year)
【Estimator】PPML+FE / dynamic panel / local projection / structural
【Inference】multi-way / design-based / cluster-robust
【Magnitudes】[elasticity / welfare / pass-through reported]
【Robustness done】[FE structures, elasticity sensitivity, modern DID, ...]
【Robustness missing】[...]
【Next step】jie-contribution-framing or jie-tables-figures
Supplementary resources
