Final Report Summary - HETMAT (Heterogeneity That Matters for Trade and Welfare)
HETMAT is a research project intended at studying conditions under which firm-level heterogeneity matters for trade patterns and welfare gains. There are three parts of the project.
The first part starts by showing that the distributional assumption on firms’ heterogeneity is critical in assessing whether there are new insights that trade economists gain when adding heterogeneity to their models. One particular distribution, the Pareto distribution has been used extensively because it yields simple and usual functional forms for equations describing the welfare gains from trade and the bilateral trade volumes (gravity), two fundamental equations in modern trade analysis. HETMAT shows in the context of the gravity equation that this Pareto assumption is not innocuous and its predictions seem partly at odds with empirical patterns, in particular concerning the invariance of the response of trade flows to trade costs. An even more important result is that, when distributed non-Pareto, micro-level heterogeneity does matter, i.e. aggregate statistics are not sufficient to evaluate welfare changes and predict changes in trade patterns. One needs micro-level data and more advanced estimation and calibration methods developed in this part of HETMAT.
The second part of the project focuses on specifying which type of firm-level heterogeneity matters. The literature has almost entirely concentrated on differences in productive efficiency and on its impact on exports. In many cases, product quality might be more relevant in the competition on international markets. Furthermore, firms are highly heterogeneous in the organization of their multinational production network. Of particular importance, HETMAT shows that i) the quality and brand appeal of products can be at least as important as productive efficiency, and ii) that the consequences of trade policy changes are critically influenced by choices that go well beyond exports. In particular, the decisions of where to source production for each market served, and which varieties to offer should be integrated in the model if one is to understand fully the consequences of signing regional agreements or increasing tariffs. The ERC grant has enabled me to buy extremely rich datasets for cars at the model level to investigate those two aspects of heterogeneity and provide quantified answers to the questions raised by current trade policy questions such as the Brexit, or changes in the US trade policy stance.
Extending the model so that firms are able to produce and export multiple products, the third part of the project shows that heterogeneity inside the firm also matters for welfare changes following trade shocks. Following trade liberalization, the change in the relative sales of different products sold by the firm, alters its measured productivity. Importantly, this mechanism is only present if the model deviates from the constant elasticity (CES) demand system traditionally used in the field. HETMAT also has an important stream of investigation studying the micro and macro-level consequences of adopting more flexible demand systems.
The first part starts by showing that the distributional assumption on firms’ heterogeneity is critical in assessing whether there are new insights that trade economists gain when adding heterogeneity to their models. One particular distribution, the Pareto distribution has been used extensively because it yields simple and usual functional forms for equations describing the welfare gains from trade and the bilateral trade volumes (gravity), two fundamental equations in modern trade analysis. HETMAT shows in the context of the gravity equation that this Pareto assumption is not innocuous and its predictions seem partly at odds with empirical patterns, in particular concerning the invariance of the response of trade flows to trade costs. An even more important result is that, when distributed non-Pareto, micro-level heterogeneity does matter, i.e. aggregate statistics are not sufficient to evaluate welfare changes and predict changes in trade patterns. One needs micro-level data and more advanced estimation and calibration methods developed in this part of HETMAT.
The second part of the project focuses on specifying which type of firm-level heterogeneity matters. The literature has almost entirely concentrated on differences in productive efficiency and on its impact on exports. In many cases, product quality might be more relevant in the competition on international markets. Furthermore, firms are highly heterogeneous in the organization of their multinational production network. Of particular importance, HETMAT shows that i) the quality and brand appeal of products can be at least as important as productive efficiency, and ii) that the consequences of trade policy changes are critically influenced by choices that go well beyond exports. In particular, the decisions of where to source production for each market served, and which varieties to offer should be integrated in the model if one is to understand fully the consequences of signing regional agreements or increasing tariffs. The ERC grant has enabled me to buy extremely rich datasets for cars at the model level to investigate those two aspects of heterogeneity and provide quantified answers to the questions raised by current trade policy questions such as the Brexit, or changes in the US trade policy stance.
Extending the model so that firms are able to produce and export multiple products, the third part of the project shows that heterogeneity inside the firm also matters for welfare changes following trade shocks. Following trade liberalization, the change in the relative sales of different products sold by the firm, alters its measured productivity. Importantly, this mechanism is only present if the model deviates from the constant elasticity (CES) demand system traditionally used in the field. HETMAT also has an important stream of investigation studying the micro and macro-level consequences of adopting more flexible demand systems.