Credit Default Swaps (CDS) are financial derivative products that insure bond investors against firm-default. Determining the payout, however, is complicated because the outstanding value of the insurance is larger than the debt outstanding. The value of a bond is also heterogeneous. CDS payouts are therefore determined in a two-stage auction. In the first stage dealers commit to either supply or purchase a fixed quantity at the unknown final price. Then, the excess supply or demand is announced and a multiunit uniform price auction is held to determine the market clearing price. Dealers have an incentive to bid strategically; in addition to the standard information rents in multiunit auctions, the two stage auction features (i) learning across rounds, (ii) pre-committment of quantities in the first round, and (iii) heterogeneous positions in CDS contracts. The paper develops and estimates a structural model of bidding behavior in these auctions and uses it to quantify the role of each of these channels in the dynamic auction process. I then consider counterfactual changes to the auction format, including a double auction design with step function bidding.