Methodological Shortcomings in Estimating Armington Elasticities

CGEMs are one of the most potentially powerful tools for simulating policies. However, a major restriction they face is them needing a huge number of parameters that are not always available, and even at times impossible to obtain. When CGEMs are applied to trade, Armington elasticities of substitution are one of those key sets. The common practice among CGEMs builders has been to impose these values, either at will or by using those stemming from other existing research. There is general consensus, however, that econometrically estimated parameters for each case study would substantially improve the robustness of results. Unfortunately, most of the so obtained elasticities are considered underestimates of both the real and theoretically expected parameters. This gave rise to a growing concern on the eventual role played by methodological and empirical issues in this matter. There are some findings that have generated a quite general consensus among researchers, related to the type of datasets used; the disaggregation levels at which goods are defined and the origins of imported varieties are considered; the frequency of data; among others. We here argue that many of these assessments, although mostly sensible, are not at all robust, as they are based on the exhaustive comparison of values that were, however, estimated using non comparable - and even statistically misspecified - econometric models. We focus on four particular issues: the spatial/temporal character of the datasets used; temporal frequency of the data; the aggregation level chosen to define a good; and the estimation methodologies used. We critically discuss the existing surveys and the papers therein reviewed, and use Uruguayan data as a means of testing for the validity of the different hypotheses