WEBINAR RECORDING · 2024-06
Quantitative Trade Models for Policy
In this conversation: Ben Shepherd; discussant Hildegunn Nordås; moderator Hanna Norberg.
Ben Shepherd tests whether a quantitative trade model’s tariff counterfactuals help explain observed trade changes. Hildegunn Nordås welcomes the accessible validation exercise but presses on model assumptions, bilateral omitted factors, adjustment timing, and how performance might change during structural upheaval.
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.
Evaluate a counterfactual against the right benchmark
Watch 18:03 ↗Shepherd compares modeled responses to 2017–2018 tariff changes with observed trade changes. Simple aggregate correlations are weak, while disaggregated regressions controlling for exporter-sector and importer-sector influences reveal a positive relationship. Tariffs explain only one component of changes in trade.
Policy interpretation: When commissioning model validation, require controls for other relevant changes and specify the level of aggregation. Do not interpret a tariff-only counterfactual as an unconditional forecast of everything that will happen.
Statistical usefulness does not establish exact prediction
Watch 22:03 ↗The preferred PPML exercise finds positive, statistically significant coefficients in 34 of 35 sectors. The pooled coefficient is below one, and a one-to-one relationship is rejected overall. In roughly a third of sectors it is not rejected at the stringent significance level reported, with results depending on the testing threshold.
Policy interpretation: Distinguish evidence that the model contains useful information from evidence of quantitatively accurate predictions. Non-rejection of a unit coefficient is not proof that simulated and realized changes coincide.
Parameter uncertainty is only one dimension of robustness
Watch 38:15 ↗Repeated simulations varying estimated trade elasticities generate narrow intervals in this application. Nordås points out that the exercise does not vary model assumptions and asks whether those assumptions become more consequential during structural change. Shepherd accepts the distinction and discusses possible comparisons across model structures.
Policy interpretation: Report what uncertainty intervals include, and separately test sensitivity to economic structure, policy measurement, and adjustment horizons. Narrow parameter intervals should not be presented as comprehensive confidence in a policy projection.
From the discussion
At 40:54–42:50, Nordås asks whether fixed effects absorb bilateral non-tariff influences and highlights anticipation, investment, and sourcing delays. At 52:03–53:50, Shepherd explains the one-year window and acknowledges that pandemic-era performance could differ. At 53:52–55:37, he explicitly says he has not run the same test on GTAP and invites replication. Their final exchange at 59:29–60:49 preserves the question of generalizing beyond the tested setting.
Recording references
- Recording: correlations and conditional performance — 18:03–24:49
- Recording: discussant’s robustness and timing critique — 38:15–42:50
- Recording: scope and comparative validation Q&A — 52:03–55:37; 59:29–60:49
