LAST MINUTE BIDDING AND THE RULES FOR ENDING SECOND-PRICE AUCTIONS: THEORY AND EVIDENCE FROM A NATURAL EXPERIMENT ON THE INTERNET

This paper was written by Alvin E. Roth and Axel Ockenfels (of Harvard Economics/Business Administration, respectively). While the latter part of their paper is very technical, the qualitative remarks posed by Roth and Ockenfels shed light on the nature of online auctions.

The paper explores a phenomenon that is not accounted for in introductory models of auction analysis — the practice of last-minute bidding (or sniping). The existence of this practice seems to be at ends with the game theory behind second-price sealed-bid auctions; it should be a best strategy to submit your maximum valuation for some item at any point in the auction. Why, then, does sniping happen so often? Does it have a place in a stable equilibrium? Roth and Ockenfels purport that it does. They go on to justify sniping by pointing out some of the key differences in a canonical second-price sealed-bid auction and the auctions that happen online.

The most obvious thing that our model does not account for is the time limit on bid submissions. Despite what we’ve learned in class, this actually can make a difference in strategy. For example, theres this notion of a community valuation for a good. That is to say that a set of bidders watching the maximum bid on an object rise might think something along the lines of “Everybody else wants this item…it must actually be better than I had originally valued it.” In doing so they will bid higher, and the cycle will continue for some time. But sniping circumvents this potential price war — not revealing your true value until the end could increase your payoff.

However, the story gets more complicated when you consider some additional dynamics. For example, there are services that will enter an eBay bid for you at the very last minute. Furthermore, there is a probability that the server will get too trafficked towards the ending time of the auction and that your last minute bid will not go through (and this will decrease the payoff for the sellers). Furthermore, sniping tends to happen more on certain types of objects than for others (i.e. frequently on antiques, not so frequently on certain technologies). The aforementioned paper explores these phenomena and concludes that sniping has a place in an equilibrium.

This paper is relevant to the class because it provides insight into the extent to which our models are accurate. It essentially takes something well defined in theory (a second-priced, sealed-bid auction) and analyzes it in a real-world context.