The Choice of Structural Model in Trade-Wages Decompositions
Stanford King Center on Global Development Working Paper
Despite a large volume of literature, there has been relatively little numerical use of structural models when trying to decompose observed wage changes resulting from joint trade and technology shocks into portions attributable to each source. This paper highlights a series of difficulties with the numerical performance of widely used theoretical structures for decomposition experiments, which raise questions as to their suitability for this purpose. For small open economies, the conventional Heckscher-Ohlin-type model implemented with convenient functional forms (Cobb-Douglas, CES) reveals problems of specialization unless the price changes accompanying trade shocks are small. As a result, this model is typically unable to accommodate large international price changes in performing decompositions. The model can also yield wide ranges for decomposition from alternative parameterizations consistent with the same wage change. In addition, as Leamer (1994) has argued, the structure can only meaningfully accommodate sector-biased rather than factor-biased technical change in the small open economy case. A differentiated goods model which generalizes Heckscher-Ohlin removes problems of specialization and is able to accommodate large price and wage changes, but introduces demand side responses to trade shocks which greatly reduce the effect of trade on wages. It also performs in implausible ways in the presence of sector-biased technical change. Our conclusion is both that the choice of structural model seemingly matters for decomposing observed wage changes into trade and technology components, and that structural models based on standard trade theory seemingly have numerical properties that make them not wholly satisfactory for this purpose. We also argue that reduced-form methods that do not allow for discrimination between alternative model parameterizations may not be that informative for trade-wages decompositions.