The dramatic increase in outsourcing has led to a burgeoning theoretical literature that tries to explain the associated organization of production. So far the literature has focused solely on analyzing the determinants of decisions by outsourcing firms, but has ignored the firms to which production is outsourced. This dissertation bridges this gap in the literature by studying outsourcing decisions not from the point of view of the outsourcing firm alone, but as a joint process that actively involves the manufacturer to whom production is outsourced.

We focus on a particular form of international outsourcing, also known as the lohn system , in which the outsourcing firm provides the manufacturer with inputs needed to produce and then re-imports the final goods. We use an incomplete contracts framework to develop a theoretical model that shows that the lohn system is more likely to be adopted the lower the manufacturer's ability to find low-cost inputs; the lower the bargaining power of the manufacturer; and the lower the degree of relationship-specificity.

In order to test empirically the predictions of the theoretical model we exploit two unique firm-level databases with monthly data on physical production and balance sheet items for a large number of firms in Romania. We use the data sources to construct two data sets, at the firm and firm-product level, respectively. We present firm-level results for cross-sections for the years 2005 and 2006, while at the firm-product-level we provide results for both cross-section and panel data.

Our empirical findings support the main predictions of the theoretical model. For instance, measuring the bargaining power of the manufacturer as the ratio of domestic to export sales, we show that the lower this ratio is, the more likely it is that the manufacturer will adopt the lohn system. Similarly, we find that the lower the firm or product specificity, the higher the use of the lohn system. Using firm age as a measure of its ability to obtain the low-cost input provides mixed evidence for our theoretical prediction. Our results are robust to the use of different estimation procedures, measures, and samples.