many questions. Studies have used these data to examine the effect of a seller’s reputation on
bidder behavior; see, for example, Lucking-Reiley et. al. (2007), Melnik and Alm (2002),
Resnick et. al. (2002) and Livingston (2005). Others have used eBay auctions to estimate the
demand for various goods, such as Adams (2007) and Song (2005). Finally, these data have
been used to test a variety of hypotheses from auction theory. For example, Katkar and Reiley
(2007) investigate how public or secret reserve prices affect seller revenues, and Hussein and
Morgan (2006) evaluate how higher shipping and handling charges affect the auction price and
how many bidders place a bid in the auction.
Each of these types of studies is typically based on a model of seller and/or bidder
behavior that relies on a standard theoretical assumption in auction theory: that the number of
bidders who participate in the auction is both known and exogenous. Unfortunately, this is likely
not the case in Internet auctions, where there are frequently many auctions of the same item
active at any given time that are in competition with each other. Buyers can choose which
auctions they find the most attractive, forcing sellers to compete for bidders. The number of
bidders in a given auction therefore is endogenous, not exogenous, and the number of bidders in
each auction is not independent.
The analysis supports the central insight of the model of Peters and
Severinov – that auctions on eBay are not independent, and that bidders compare the prices
available in various auctions of a good when determining which auction they want to bid in. As
the authors note, the goal of their model is not to model the eBay market specifically, but to show
that when auctions compete against each other for bidders, a simple mechanism exists that results in an
ex-post efficient outcome, . However, there are differences between the setting studied by Peters and
Severinov and the design on the eBay market, and these differences result in a divergence between
the predictions of the model about how bidders will behave, and the choices that they are actually
revealed to make in the data. The model predicts that bidders should always choose to bid in the
auction that offers the lowest current price. The data, however, reveals that bidders only select this
auction if it also is about to end, and if the seller has a suitably established reputation for honest behavior.