Towards More Accurate Trade Model Forecasts

GTAP is an open-source model which can be customized by different users based on the expert inputs and specific situations being considered. However, it is at times applied without sufficient level of expert inputs and customization, leading to misleading policy conclusions. While GTAP is not specifically intended to be a forecasting tool, it is widely used that way to analyze free trade agreements and other trade policy proposals and the results of such analysis carry substantial weight in the political arena. In this article we suggest modifications to the model that can improve model results. Our GTAP model extension involves two key features. Firstly, we incorporate the country level variations in Armington elasticities by calibrations which are first estimated at country level using the detailed trade and tariff datasets. This is done by capturing these elasticities in the model through the variables in GTAP and iterating on the elasticities until the model-derived elasticities match the historically estimated elasticities. Secondly, we capture the empirically established result that 30% of the workers in transition do not get re-employed in other sectors as assumed by GTAP. We accomplish this by iterations: first we infer the employment changes by sector assuming unemployment in unskilled labor; second, we sum up the total job losses before transition and then multiply that by 30%, to derive the shock to employment. These model extensions enable us to produce results for the US and South Korea after the implementation of the KORUS trade agreement, which took effect in 2012. We compare our post-KORUS results with the results of other GTAP exercises done prior to KORUS and we compare the results to the actual outturn for the US economy in recent years.