The trade elasticity from tariff-based regressions: what do we measure?

Non Technical Summary

A large share of empirical research on international trade relies on variation in import tariffs to estimate how strongly trade responds to changes in trade costs. Crucially, most of this tariff variation is generated by Preferential Trade Agreements, which allow importers to charge lower tariffs to certain partners. Because these agreements specify different tariff rates for different exporters, they create rich variation that economists routinely exploit to estimate the “trade elasticity” – a key parameter used in nearly all quantitative trade models.

However, a central but often overlooked fact is that firms do not automatically receive these preferential tariffs. To claim them, they must comply with rules of origin and other administrative requirements that can be costly or difficult. As a result, many eligible exporters continue paying the higher Most Favoured Nation (MFN) tariff instead of using the preferential rate. The share of imports that actually enters under the preferential tariff – the preference utilisation rate – is often far below 100%.

This paper demonstrates that ignoring these low utilisation rates leads to a fundamental mismeasurement of the trade elasticity. Standard regressions of imports on statutory preferential tariffs identify the responsiveness of trade to eligible tariffs rather than to the effective tariffs actually paid. In statistical terms, this is a classic partial compliance problem: tariff preferences are the treatment, but only some eligible firms comply by using them. The estimated coefficient therefore corresponds to an intent-to-treat effect, not the true treatment effect.

This paper shows that the commonly reported “tariff elasticity” from the literature is an eligible tariff elasticity, equal to the true structural elasticity multiplied by the average preference utilisation rate in the estimation sample. Because preference utilisation rates typically lie well below one, traditional estimates are attenuated. Moreover, these eligible tariff elasticities are not policy invariant: they depend on the specific utilisation patterns in the data – which vary widely across partners and agreements – and therefore cannot be validly applied to other policy environments, such as MFN tariff reforms or customs unions, where compliance is effectively complete.

To recover the correct effective tariff elasticity, the paper applies an instrumental variables approach inspired by the Local Average Treatment Effect (LATE) literature. Tariff reductions implied by EU Preferential Trade Agreements serve as the treatment assignment, while the tariff reductions actually realised – statutory cuts multiplied by the observed preference utilisation rate – represent the treatment received. Using this strategy on four major EU agreements (with Korea, Canada, Japan, and Singapore), the paper finds that the effective-tariff elasticity lies between 10 and 12 – around three times larger than eligible-tariff elasticity estimates of 3-4.

These findings have major implications for quantitative trade modelling. Because most models currently use eligible tariff elasticities, they effectively treat partial utilisation as if it reduced the responsiveness of trade rather than the size of the tariff change. This leads to systematic overprediction of the trade effects of FTAs. A general equilibrium illustration using the EU-Japan agreement shows that standard modelling approaches can overstate predicted trade increases by up to a factor of two.

The paper concludes that researchers and policymakers should treat partial utilisation as a feature of the tariff change – not as a modification of the elasticity – and that future empirical work should incorporate preference utilisation rates directly. Building comprehensive datasets on utilisation rates will be essential to ensure that trade policy is evaluated using structurally meaningful parameters.

Read the Working Paper