The structure of factor content predictions*

The last decade witnessed an explosion of research into the impact of international technology differences on the factor content of trade. Yet the literature has failed to confront two pivotal issues. First, with international technology differences and traded intermediate inputs, there is no existing definition of the factor content of trade that is compatible with Vanek's factor content prediction. We fill this gap. Second, as Helpman and Krugman (1985) showed, many models beyond Heckscher-Ohlin imply the Vanek prediction. Thus, absent a complete list of these models, we do not fully know what models are being tested when the Vanek prediction is tested. We completely characterize the class of models being tested by providing a familiar consumption similarity condition that is necessary and sufficient for a robust Vanek prediction. Finally, we reassess the performance of the prediction using the correct factor content definition and input-output tables for 41 countries. We find that the prediction performs well except for the presence of missing trade. Further, missing trade is not pervasive: it is associated entirely with 'home bias' in the consumption of agricultural goods, government services and construction.