Optimal Auctions, an eBay experiment

In this project, we seek to explore optimal auctions by using eBay data. The purpose is to apply the theoretical framework to real world data and try to evaluate the way individuals bid in reality, compared to how the theory as applied to the data would suggest they should. The project is set out in two parts. The data is drawn from eBay auctions referring to the same watch. First, an analysis of the real-world data and second an application of the theory to this data.

In the first part we seek to compare how different parameters in the auctions affect price as well as how the characteristics of the auctions interact with the price and other auction parameters. This includes, both looking at descriptive statistics of auctions, grouped by certain characteristics, such as their duration or type as well as regression analysis looking at how specific aspects of the auction, such as bidder ratings, duration, starting bid and others interact with the final price.

In the second part we apply the theory. We use the bidding data and the bids, and we deduce each bidder’s valuation. Here we use normalized data to do so, grouping actions by number of bids and duration. Then we apply the theory. First, we apply a VCG model in an attempt to maximize bidder welfare. We also deduce the reserve price and then we try to understand how a revenue maximizing auction would operate both compared to the real-world data as well as normal second price auction, assuming the same bid vector.

In all we reach the following conclusions. First, there is a direct relation between the auction type, the bidders it attracts and the final price of the good. Second, bidders in reality do not bid in a welfare maximizing way. Third, eBay starting prices are usually lower than they should be and last, that the bids on eBay, subject to some caveats are very close to revenue maximizing.

• final report (including deliverables)
• proposal
• code files