This research bridges a gap between an important real world market and abstract models of economic theory is an important ingredient in any empirical work. The project develops new mathematical and statistical models to understand divisible good auctions. This kind of auction is typically employed to sell treasury bills or other government securities, electricity, and shares in initial public offerings (IPOs). Since about $4 trillion worth of treasury bills alone are sold annually worldwide, these auction mechanisms affect very important markets. All of these auctions share the common feature that bidders' strategies (bids) are restricted to take a particular form. In particular, they have to be characterized by a finite (and usually quite small) number of price-quantity pairs, which together describe the bid function (or demand function). This institutional feature of bids, which has been to a large extent ignored in the previous literature, has important implications for the performance of auction mechanisms. This project offers a new theory for auctions with constrained bidding and applies it on data from two different settings: treasury bill auctions of the Czech and Canadian governments respectively.
In the first application, this project analyzes treasury auction data from the Czech Republic. This country uses a uniform price auction format, which means that all bidders pay the same market clearing price for all treasury bills they win. This research offers a new method for recovering bidders' valuations from data on bids and using these estimates it demonstrates that ignoring the restrictions on the bidding strategies could lead to important problems when conducting counterfactuals or evaluating the performance of the auction mechanism.
In the second empirical application the project considers a novel statistical test for deciding between models with private and common values. In a model with private values, a bidder's own valuation for the auctioned good is independent of the information of his opponents. When values are common, on the other hand, a bidder's valuation might depend on rivals' private information. The test is applied to data from treasury auctions of the Canadian government, which have a discriminatory auction format, which means that every bidder has to pay his full bid for every treasury bill she wins. The results suggest that a private value model is appropriate in the case of 3-months treasury bills, but not for 12-months treasury bills. This research also analyzes the value of information that large primary dealers derive from customers' order flow and finds that this source of profits accounts for a significant share of primary dealer's overall profits in these treasury bill auctions.
The broader impacts of this research will benefit practitioners deciding on auction rules and how to design or adjust the institutional settings to improve auction performance. I
