Applying Roy-model counterfactual methods to determine the direction of industrial use: raw versus processed exports in Uzbekistan

Determining the direction of industrial use, i.e. whether output is exported raw or processed, requires counterfactual returns: what a firm that exports raw output would earn by processing, and what a processing firm would earn by exporting raw. We apply Roy-model counterfactual methods to this question for Uzbekistan. We first measure the direction of industrial use in observed data. Using broad economic categories, the processed share of goods exports rose from 0.54 in 2010 to 0.85 in 2022. HS chapter 71, dominated by non-monetary gold, inflates the level of this share but not its growth: excluding it, the share rose from 0.40 to 0.79. Within food and beverages, the processed share rose from 5.4% to 31.8%. We then model the firm's choice as a Roy model of comparative advantage and use two counterfactual methods: a switching regression that recovers average returns to processing for processors (ATT) and non-processors (ATU), and a bounding argument due to Gardner that identifies a lower bound on group differences in returns among processors. We state the conditions under which each method is valid. A simulation calibrated to Uzbek aggregates shows that naive comparisons of processing and raw exporters can strongly overstate the return to processing, that the switching regression recovers counterfactual returns when correctly specified, and that the Gardner bound fails when processing rates exceed one half or when cost shifters vary within cells. Firm-level customs records linked to cluster registries are required to apply these counterfactual methods to Uzbek exporters