³ÉÈË´óƬ

An important question in the auctions of divisible goods such as securities, emissions permits, and electricity is to determine whether a discriminatory auction yields higher revenues to the auctioneer than a uniform price or Vickrey auction. The question can be answered with information about the distribution of bidders' true marginal valuations for the good being auctioned. Since this information is not revealed directly in auction data due to strategic bidding, I model strategic behavior in a discriminatory auction. Based on this model, I propose an estimation method to reconstruct the distribution of marginal valuations using data on individual bids. I apply the method to data from 3-month Treasury bill auctions conducted by the Turkish Treasury between 1991 and 1993. I reconstruct the perfectly competitive outcome in this market using the estimated marginal valuations of the bidders. I find that the discriminatory auction yielded more revenue to the Turkish Treasury than in the competitive outcome. Since the auctioneer's revenue in the competitive outcome is an upper bound to revenue from a uniform price or Vickrey auction, I conclude that the Turkish Treasury was using the revenue maximizing mechanism among the alternatives. I also find evidence that bidders' demands become more inelastic when they face binding liquid asset reserve requirements; which supports the Treasury's choice of a discriminatory scheme for revenue maximization. I establish that my results are robust to the presence of a common-value component in bidders' utilities when comparing the revenue performance of the discriminatory and Vickrey mechanisms. I also show that the estimation framework can be generalized to allow for deterministic vs. uncertain supply, asymmetries, and stochastic participation by bidders. The empirical method I develop is exible and computationally straightforward, with potential applications to many other divisible good auction settings.