Product Level Trade Elasticities

The

The trade elasticity is a key parameter in all international trade models. Indeed, as demonstrated by Arkolakis, Costinot and Rodriguez-Clare (2012), welfare gains from trade (or losses when returning to autarky) are a function of the change in the share of domestic expenditure and the trade elasticity to variable trade costs. The

In estimating the structural gravity model for each of the 5,052 HS6 products, we obtain some non-significant coefficients on tariffs (i.e. when tariffs do not vary enough across exporters). We also find few positive and statistically significant trade elasticity (2.4% of all product lines) that we characterize in the working paper associated to the database (see Fontagné et al. 2019). Accordingly, in the published version of the database, each positive and non-significant HS6 trade elasticity has been substituted by the average elasticity of its HS4 heading (average across negative HS-6 specific elasticities within HS4). Concerned products are flagged.

The database therefore contains four variables:

**ProTEE**dataset (PROduct level Trade Estimated Elasticity) provides trade elasticities at product level. The trade elasticity is the reaction of bilateral import flows (in value) to a change in the applied import tariff for a given product (as defined by the WCO’s six-digit Harmonized System classification in revision 2007 - hereafter HS6). To build the dataset, we adopted a structural gravity approach and regressed bilateral import values on bilateral applied tariffs for each HS6 product. The estimated tariff coefficients represent the trade elasticity at product level.The trade elasticity is a key parameter in all international trade models. Indeed, as demonstrated by Arkolakis, Costinot and Rodriguez-Clare (2012), welfare gains from trade (or losses when returning to autarky) are a function of the change in the share of domestic expenditure and the trade elasticity to variable trade costs. The

**ProTEE**dataset, by providing such trade elasticity at product level, allows scholars and practitioners to obtain a more precise evaluation of the welfare gains from trade than using average homogeneous trade elasticity across products.In estimating the structural gravity model for each of the 5,052 HS6 products, we obtain some non-significant coefficients on tariffs (i.e. when tariffs do not vary enough across exporters). We also find few positive and statistically significant trade elasticity (2.4% of all product lines) that we characterize in the working paper associated to the database (see Fontagné et al. 2019). Accordingly, in the published version of the database, each positive and non-significant HS6 trade elasticity has been substituted by the average elasticity of its HS4 heading (average across negative HS-6 specific elasticities within HS4). Concerned products are flagged.

The database therefore contains four variables:

(i)

(ii)

(iii)

(iv)

**HS6**: the HS6 product category, in revision 2007(ii)

**Elasticity**: the value of trade elasticity(iii)

**Zero**: a dummy indicating whether the elasticity from the original estimation was a zero (i.e. non-significant)(iv)

**Positive**: a dummy indicating whether the elasticity from the original estimation was positive and significant Methodology: a Short Description

In estimating product level trade elasticity we applied a structural gravity approach on a panel of bilateral import flows (in values) and applied tariffs for a full matrix of 189 exporting and 152 importing countries. Bilateral import flows are from BACI database, while applied tariff data are from MAcMap-HS6. Since MAcMap-HS6 is available only for the years 2001, 2004, 2007, 2010, 2013 and 2016, our regression sample spans from 2001 to 2016 on a three-year windows base.

To address the heteroskedasticity in the error term (and the zero trade flows problem - missing information), we follow Santos-Silva and Tenreyro (2006) and adopt a PPML estimator as baseline (and preferred) estimator. In our estimations we always include exporter-year and importer-year fixed effects to fully control for exporter and importer multilateral resistance terms. By doing so, we exploit the variation in tariffs imposed on different exporters by a given importer at different points in time. We control for bilateral specific geographic related trade costs (as derived by a standard gravity model for trade), by including in all regressions: (i) distance (in logarithm), (ii) a dummy for common colony, and (iii) a dummy for common border. Please refer to Fontagné et al. (2019) for more details on the methodology.

To address the heteroskedasticity in the error term (and the zero trade flows problem - missing information), we follow Santos-Silva and Tenreyro (2006) and adopt a PPML estimator as baseline (and preferred) estimator. In our estimations we always include exporter-year and importer-year fixed effects to fully control for exporter and importer multilateral resistance terms. By doing so, we exploit the variation in tariffs imposed on different exporters by a given importer at different points in time. We control for bilateral specific geographic related trade costs (as derived by a standard gravity model for trade), by including in all regressions: (i) distance (in logarithm), (ii) a dummy for common colony, and (iii) a dummy for common border. Please refer to Fontagné et al. (2019) for more details on the methodology.