Home | Business News | Browse by Publication | E | Economic Inquiry

Some evidence on late bidding in eBay auctions.

Article Excerpt
I. INTRODUCTION

Bidding in the last minutes or seconds of an auction is a common strategy pursued in online auctions with fixed end-times. For instance, Roth and Ockenfels (2002, 2006) observed bids in the last minute and last 10 sec in 37% and 12% of eBay auctions, respectively. Bajari and Hortacsu (2003) documented a similar pattern.

Roth and Ockenfels (2002) put forth several explanations of late bidding. In their study of eBay auctions with fixed end-times and Amazon auctions with flexible end-times, (1) they find evidence consistent with the following three explanations of late bidding.

1. Last-minute bidding constitutes an optimal response to the presence of a bidder or bidders who submit multiple bids in one auction. (2) Bidding late is an efficient strategy that deprives the incremental bidder of sufficient time to respond.

2. Late bidding might be an optimal strategy for well-informed bidders (experts) who want to protect their private information concerning the value of a particular item. Assume that only experts can recognize the true resale value of the auctioned item (e.g., antique furniture). When the expert bids early in the auction, her bid might be a signal for other bidders that the object is unusually valuable. Bidding just before the end of a fixed end-time auction allows the expert (who can be recognized by her frequent participation or high feedback number) to profit from her information without leaving other bidders enough time to closely examine the item and bid.

3. Last-minute bidding could result from implicit collusion among bidders against the seller giving higher payoff to the successful bidder. For instance, assume that you want a new computer. It is worth $1,000 to you, and you believe one other bidder is willing to pay $1,000. If both of you use the proxy bidding system, (3) the price quickly rises to $1,000. Even if the tie is resolved in your favor, this is no bargain. But suppose you bid $300 early on and your competitor bids $500 in the last minute. You then bid $600 but take a chance that your bid might not be transmitted before the auction closes. Even if you get the computer for $600 only half the time (assuming that there are many auctions for the same computer running approximately at the same time), it is better than paying $1,000.

The aim of this paper is to test the explanations of late bidding proposed by Roth and Ockenfels (2002). After describing and exploring the primary data set in Sections II and III, respectively, Section IV tests whether late bidding creates collusive gains. We examine whether the prices in auctions with late bids are systematically lower than the prices in the remaining auctions. Section V proposes a duration model to analyze the effects of multiple bidding (Hypothesis 1) and expertise (Hypothesis 2) on the timing of the last or winning bid (the number of seconds the last or winning bid arrived before the end of the auction). (4) Section VI concludes.

II. DATA

This paper uses two data sets collected from eBay by a "spider" program. The primary data set contains all eBay auctions in categories reported in Table 1 that were listed on eBay on a particular day. Out of the 140,000 auctions downloaded, we exclude those that did not receive any bid (46.8%), auctions with the "Buy it Now" option (5) (17.4%), auctions in currencies other than U.S. dollars (7.9%), Dutch auctions (5.9%), auctions in which the reserve price was not met (6) (3.5%), and those where identities of bidders were not disclosed (1.2%). The number of remaining auctions for each category is given in Table 1. This paper employs the following variables:

* TPRICE, the total price in dollars paid by the winner (i.e., the winning price in the auction plus shipping and handling cost) (7)

* BIDCOUNT, the number of bids per auction

* BIDDERS, the number of bidders who placed their bids in the auction

* DMULTBID, dummy variable that identifies the occurrence of multiple bidding

* SELLERRAT, seller's rating (feedback score), in thousands (8)

* AUCTIONLEN, auction's duration in days (with the precision of seconds)

* LASTBTEND, the number of seconds the last bid was received before the auction closed (also referred to as the duration of the last bid)

* LASTFEEDB, the rating (feedback score) of the last bidder, in thousands.

The second data set mimics the structure of the primary data set, and the additional information it carries will be described in detail in Section V, Hypothesis 2.

III. DESCRIPTIVE STATISTICS

The basic descriptive statistics of the variables defined above for the primary data set are presented in Table 2. Table 1 breaks down the data set into eBay's categories and reveals the differences of the key variables across the product categories.

Approximately half of the primary data set comes from the computer category. Another one-quarter is formed by decorative arts and antiquities that belong to the antiques category, and the remaining auctions were taken from two other main categories on eBay--stamps and coins.

The average total price in the complete data set is $90.76, which is considerably more than the median price and indicates that the distribution of price is skewed to the right. The items selling for the lowest prices were, according to expectation, stamps. This was also the reason for the extremely low minimum value of TPRICE in the whole data set.

The median number of bids per auction was five, which exceeds the median number of bidders by two. This is a result of multiple bidding--bidders submitting more than one bid were present in nearly 60% of the auctions.

Table 1 documents relatively big differences among the individual categories. In general, computer auctions attract more bidding activity, measured by the number of...

View this article FREE - Now for a Limited Time, try Goliath Business News
Free for 3 Days!