WEBINAR RECORDING · 2022-01
Gravity Modeling Tips & Traps for Practitioners
In this conversation: Ben Shepherd; discussant Silvia Sorescu; host Hanna Norberg.
Ben Shepherd connects a policy question, structural gravity specification, estimation, and counterfactual simulation. An estimated bilateral coefficient is not a complete policy impact. Discussion with Silvia Sorescu examines domestic shipments, global value chains, missing data, and the limits of mechanical extensions.
Session information draws on the recording and its original description. Affiliations and policy context refer to the time of the webinar.
FROM THE CONVERSATION
Three takeaways.
What they mean in practice.
Prepared from the session recording using automatic transcription and editorial review. Follow the timestamp links for the original wording and context.
Start with the question and the estimand
Watch 11:03 ↗Shepherd recommends checking whether gravity actually answers the research question, with policies affecting bilateral trade costs particularly natural applications. He separates estimating a parameter from asking what trade and income would look like after a policy change. The latter requires a counterfactual exercise, including effects on third countries.
Policy interpretation: Specify the policy shock and desired outcome before choosing a familiar regression framework.
Estimation must respect the model and the data
Watch 16:15 ↗The presentation favors PPML, theoretically appropriate importer-time and exporter-time effects, and usually pair effects in panels. Domestic shipments support important applications. The later selection discussion qualifies any blanket estimator recommendation: PPML cannot repair an incorrectly specified conditional mean or omitted determinants simply by accommodating zero trade.
Policy interpretation: Treat estimator choice, identifying variation, fixed effects, and the meaning of missing observations as one research-design problem.
Extensions require economic structure
Watch 32:26 ↗For GVC questions, Shepherd recommends multisector models with input-output linkages rather than substituting value-added trade into the standard gross-trade equation. Similarly, separate price and quantity questions require returning to the underlying derivation. Counterfactuals must explain their assumptions and distinguish estimated partial responses from general-equilibrium outcomes.
Policy interpretation: Choose extensions that represent the relevant mechanism and explain exactly what the simulation holds fixed.
From the discussion
Sorescu asks how country-level policies can be identified (31:27–35:53); Shepherd discusses international-versus-domestic shipment interactions under the relevant exclusion assumption. On missing production observations, he rejects automatically turning absent data into zeros (42:37–43:31). Her monthly-data question prompts a warning about policy variables observed only annually, correlated errors, and matching frequency to actual variation (43:31–45:45). The final estimator exchange emphasizes specification and diagnostic testing over mechanically choosing PPML (51:12–54:46).
Recording references
- Recording: five pieces of modeling advice — 11:03–25:58
- Recording: applications and econometric questions — 32:26–55:41
