Decomposing the Measure of Ignorance: TFP and Fundamental Productivity in the World Economy

Ricardo's classical example shows how countries realize welfare gains by specializing according to their comparative advantages. While England is relatively more productive in manufacturing cloth, Portugal has a comparative advantage in producing wine. When opening up to trade, England specializes in fabricating cloth and Portugal in wine making, yielding gains of trade for both nations. But if countries do specialize according to their comparative advantage, why not exploiting trade flows to estimate productivity? A lack of theory clarifying the link between observable trade flows and underlying fundamental productivity prevented such empirical exercises until recently. Introducing a probabilistic representation of technologies into a Ricardian model of trade in a continuum of goods (Dornbusch et al., 1977), Eaton and Kortum (2002) provided a Ricardian model allowing for more than two regions and showing how productivity and trade costs determine the flow of goods and services. Subsequent studies have outlined the link between fundamental productivity and measured TFP (Finicelli et al., 2013), introduced theoretically sound indicators of comparative advantage (Finicelli et al., 2013), or investigated how sub-national productivity shocks translate to the national level (Caliendo et al., 2014). A number of empirical studies employed Ricardian models to identify the productivity developments empirically (Chor, 2010; Fadinger and Fleiss, 2011; Levchenko and Zhang, 2013). This study exploits the World Input-Output Database to estimate the productivity of 34 sectors in 40 countries from 1995 to 2009, using a Ricardian general equilibrium model for parameter identification. We proceed in three steps. First, we adapt the reduced-form gravity equation estimation approach pioneered by Eaton and Kortum (2002) to the WIOD tables. The model's equilibrium conditions are used as side-constraints to a linear least squares estimator, ensuring that that parameter estimates form an equilibrium of the model and that counterfactual simulations are unbiased. Second, we employ the estimation approach on the WIOD, yielding both productivity and trade cost parameters. The results reveal a number of important trends of global productivity and trade costs. Third, we use the parameterized version of the model to estimate the welfare consequences of these trends. Our results indicate that average fundamental productivity grew by 1.6% across countries and sectors between 1995 and 2008. Differences between countries and sectors are substantial. While productivity in the electrical and optical equipment sector grew by 6.6% per year, it declined by 3.9% in the mining industry. The initial level of productivity exerted a significant influence on productivity growth, implying convergence of underdeveloped countries and sectors. China and the new EU members (those which joined from 2007 on) exhibited a particularly strong productivity growth. Even when controlling for convergence, the People's Republic still experienced a significantly higher productivity growth than the industrialized countries. Counterfactual simulations suggest that Chinese welfare would be 21% lower if it had been increasing its productivity half as strong. We find that services' productivity grows by about 1.8% per year. If it were to increase by only half that speed, welfare in most regions would be about 6 to 8% lower in 2008. In accordance with the literature, we find that the elasticity of distance falls slowly between 1995 and 2009. The median elasticity of distance across all sectors only fell by 3.7%. Consequently, the welfare effect of decreasing elasticities of distance is small. If 1995's numbers were still in place in 2008, welfare would be only between 0.5 and 1.5% lower in most nations.