Model structure or data aggregation level: Which leads to greater bias of results?

The aim of this paper is to provide a first step toward a systematic sensitivity analysis of a system of linked models. We focus on two fundamental characteristics: the model structure and the data aggregation level. Employing the Global Trade Analysis Project (GTAP) framework, we combine the general equilibrium (GE) and partial equilibrium (PE) versions of the GTAP model, each of which is then run with a highly aggregated and a highly disaggregated version of the GTAP database. Based on this experimental setting, we quantify the biases resulting from the data aggregation, the model structure and the interaction of these two model characteristics. We conclude that data aggregation as well as the related false competition and tariff averaging influence the results significantly more than the model structure, whereas the bias stemming from the interaction of the two model characteristics is negligible.