jie-identification-strategy

Identification / Modelling Strategy (jie-identification-strategy)

When to trigger

• Your gravity regression is log-OLS and a referee will ask about zeros and heteroskedasticity
• A trade-policy effect is identified off staggered RTA/tariff timing with plain TWFE
• A shift-share (China-shock-style) design lacks a share- or shift-exogeneity argument
• An open-economy/structural model's mapping to the data is not disciplined
• You are unsure your design clears JIE's field-level credibility bar

The JIE identification bar

JIE judges credibility by the standards of international economics, where "identification" can be either a quasi-experimental design or the discipline a structural model imposes on the data. Both halves of the field have their own conventions; pick the branch that matches your paper.

Branch A: Gravity (the workhorse of empirical trade)

• Use PPML (ppmlhdfe, fixest::fepois), not log-OLS: it handles zeros and the heteroskedasticity that Santos Silva–Tenreyro showed bias log-linear gravity.
• Control multilateral resistance with importer×time and exporter×time fixed effects; for policy effects add country-pair fixed effects.
• Report the elasticity with an economic interpretation, not just a t-stat.

Branch B: Trade-policy shocks (tariffs, RTAs, NTMs)

• Staggered agreement/tariff adoption → move beyond TWFE: Callaway–Sant'Anna, Sun–Abraham, or de Chaisemartin–D'Haultfœuille; show pre-trends with an event study.
• Address anticipation and phase-in (agreements take effect over years).
• Falsification: effects absent for trade flows the policy should not touch.

Branch C: Shift-share / Bartik exposure (e.g., the China shock)

• State whether identification rests on exogenous shares or exogenous shifts.
• Use design-based inference: Adao–Kolesar–Morales standard errors or the Borusyak–Hull–Jaravel shift-level approach.
• Show the exposure measure is not proxying for a confounding local trend.

Branch D: Exchange rates, pass-through, international pricing

• Specify horizon (impulse responses / local projections) and control for currency of invoicing.
• Distinguish the exchange-rate movement's source (the "disconnect" problem); a raw correlation is not pass-through.

Branch E: Structural / open-economy macro (model is the identification)

• Map the model to targeted moments (spreads, cyclicality of the current account, pass-through elasticities) and report untargeted moments it also matches.
• For sovereign default (Eaton–Gersovitz / Arellano) or small-open-economy DSGE, state calibration vs estimation and show sensitivity to key elasticities (trade elasticity, risk aversion).
• Counterfactuals (often via exact-hat algebra in trade) must be robust to the elasticity you cannot pin down precisely.

Execution bridge (StatsPAI / Stata MCP)

Estimate and audit the design, don't only describe it. Full map: execution-with-mcp. JIE is international trade/macro; cross-country/firm panels — 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

• Design/branch named and matched to scope half (trade vs macro-finance)
• Gravity uses PPML with multilateral-resistance fixed effects (if applicable)
• Staggered policy timing uses a modern DID estimator with pre-trends (if applicable)
• Shift-share share/shift exogeneity argued with design-based inference (if applicable)
• Structural model disciplined against targeted + untargeted moments (if applicable)
• Inference matched to the design; counterfactuals shown robust to key elasticities
• The claim never exceeds what the design or model supports

Anti-patterns

• Log-OLS gravity that drops zeros and ignores heteroskedasticity
• TWFE on staggered RTA/tariff adoption with no heterogeneity-bias discussion
• A shift-share design with conventional standard errors and no exogeneity argument
• A calibrated model that matches one moment and is sold as validated
• A local trade/pass-through estimate oversold as a structural global parameter

Referee-pushback patterns with the venue fix

The four objections that sink identification at JIE, and the move that answers each: "gravity without multilateral resistance / on log-OLS" → PPML with importer×time and exporter×time fixed effects, zeros retained, elasticity reported; "trade-policy timing is endogenous" → pair fixed effects, a pre-trend event study, modelled phase-in, and a falsification on untouched flows; "shift-share exposure proxies a local trend" → an explicit share-vs-shift exogeneity claim with Adao–Kolesar–Morales or Borusyak–Hull–Jaravel inference; "the quantitative model is not disciplined by the data" → targeted and untargeted moments side by side, plus a counterfactual that survives a sweep over the un-pinned-down trade elasticity. Naming the branch (A–E) and pre-empting its signature objection in the design section is what separates a credible JIE identification from one a referee can puncture in a paragraph.

Output format

【Branch】gravity / trade-policy shock / shift-share / pass-through / structural
【Scope half】trade / macro-finance
【Identifying variation or model discipline】one sentence
【Diagnostics done】[PPML+FE, pre-trends, AKM inference, moment fit, ...]
【Diagnostics missing】[...]
【Counterfactual robustness】key elasticities varied? [Y/N]
【Next step】jie-data-analysis

Supplementary resources

• ../../resources/external_tools.md — ppmlhdfe/fepois, modern DID estimators, AKM inference, Dynare/Julia structural toolkits
• ../../resources/official-source-map.md — scope and editor-roster sources