‘Vettes and Lemons on EBay - University of Virginia

`Vettes and Lemons on eBay

Page 1

`Vettes and Lemons on EBay

Christopher P. Adams, FTC Laura Hosken, FTC Peter Newberry, FTC

PRELIMINARY DRAFT

Do not distribution without authors' permission.

April 11, 2006

ABSTRACT

Using bid data from 8,000 new and used Chevrolet Corvettes sold on eBay, this paper empirically tests Akerlof's (1970) hypothesis that the used car market is characterized by low quality and informational problems. The hypothesis states that the used market has a higher proportion of low quality cars than the new market and buyers account for the difference by discounting their value for a used car relative to a new car. This is tested by comparing bids on new and late model used `Vettes. The paper finds little evidence of a premium for new `Vettes. The paper also considers a natural generalization of Akerlof's (1970) model which allows potential bidders to have different and incomplete private information about the quality of the used car. One implication is that a rational bidder will bid late in the auction in order to reduce the likelihood of other bidders observing his private information. A second implication is that a rational bidder will discount his bid in order to reduce the likelihood of winning the auction because his private information is incorrect (the winner's curse). We test the first implication by comparing bid times on new and used cars and by comparing bid times on newer and older used cars. Akerlof (1970) suggests that the quality and informational problems will be larger with used cars relative to new cars, and other empirical work suggests that quality and informational problems are larger for older used cars relative to newer used cars. Therefore, we expect to find more late bidding on used cars relative to new cars and older used cars relative to newer used cars. If anything, we find that the opposite is true for `Vettes. To test the second implication we compare bidding on used `Vettes for auctions with different numbers of expected bidders. Recent theory suggests that bidders facing a winner's curse problem will bid less as the number of expected potential bidders increase. Our analysis suggests bidders bid slightly more in auctions with a higher number of expected bidders. Overall, for used Chevrolet Corvettes sold on eBay there is little empirical support for the hypothesis presented in Akerlof (1970). It is not clear, however, whether these results generalize to other cars sold on eBay or cars sold in the off-line used market.

The views expressed in this paper do not represent the views of the FTC or any individual Commissioner. The authors would like eBay and staff at eBay Motors for providing the data. We also want to thank for information they provided us. We want to thank Jorge Roberts and David Yans, for excellent research assistance, Steve Berry, George Deltas, Avi Goldfarb, Debra Holt, Dan Hosken, Jim Lacko, Fiona Scott Morton, Mike Smith, Unjy Song, Steve Tenn, Bill Vogt, Pai-Ling Yin, Robert Zeithammer, participants at the 2005 IIOC meetings in Atlanta, 2005 INFORMS conference, Maryland IO seminar, FRB IO Seminar, DOJ's EAG seminar, and other FTC colleagues for helpful comments in regards to this research project. Any errors are our own. CONTACT: cadams@

`Vettes and Lemons on eBay

1. Introduction

Page 2

Akerlof (1970) states the market for used cars is riddled with quality and informational problems. Because only owners can observe the car's quality, the proportion of "lemons" in the used car market is much higher than the proportion of lemons in the new car market. Akerlof (1970) argues this "lemons problem" explains why there is such a large price difference between new cars and late model used cars. Akerlof (1970) further argues it is possible for the used car market to collapse because information asymmetries make trade difficult. Given these dire predictions, we may not expect used cars to become very popular in an on-line market like eBay. Yet in 2003 eBay Motors sold an average of one thousand used cars per day. eBay Motors has become the most successful unit of one of the most successful companies of the Internet era. Is Akerlof (1970) correct? Is there a lemons problem on eBay Motors? This paper presents a number of empirical tests of Akerlof's lemons hypothesis using bid data for new and used Chevrolet Corvettes sold on eBay between April 2001 and December 2003. Overall, the paper finds little empirical support for the hypothesis, at least in regards to used Corvettes sold on eBay Motors.

The paper's first test of Akerlof's hypothesis considers the implication that bidders are willing to pay a considerable premium for a new `Vette relative to a late model used `Vette. If there is a quality problem and the current owner has more information about the car's quality, then bidders will be wary of buying a used `Vette and will discount their bid to account for the increased likelihood that the car is a lemon. First, the paper compares bids on new and used `Vettes that are less than a year old. This simple comparison of means and medians suggests bids are higher for the late model used cars than for the new cars. Second, the paper presents results from regression analysis of bids and prices which account for observed characteristics of the cars, the sellers and the bidders. The results suggest there is little evidence of a premium for "new" cars relative to late model used cars.

The paper follows up on this simple test of a new car premium by looking for evidence of informational problems in eBay's used `Vette market. In Akerlof (1970) sellers of used cars know the exact quality of the car while buyers know only the population probability that the car is a lemon. A natural generalization allows buyers to observe "signals" of the car's quality. Such a model allows bidders to correspond with

`Vettes and Lemons on eBay

Page 3

the seller, observe detailed photographs of the car, obtain a Carfax Report, take the car for a test-drive or get an independent mechanic's assessment of the car's quality. If the car's quality is a characteristic that all bidders consider to be important and bidders have different signals of the car's quality, then the auction is a "common value" or "affiliated value" auction. This paper empirically tests whether bidders behave in a way that consistent with rational behavior when bidding in an affiliated values auction. In particular, the paper tests whether bidders bid late in the auction in order to hide their private information from other bidders and whether bidders discount their bids to account for the winner's curse (the probability the winner's private information over values the actual quality of the car).

Late bidding or "sniping" is something of a phenomenon on eBay and numerous explanations have been provided for why such behavior occurs (Bajari and Hortacsu (2003), Roth and Ockenfels (2002)). According to Bajari and Hortacsu (2003) if bidders have private information about the quality of the item then these bidders are made worse off if this private information becomes known to other bidders (Milgrom and Weber (1982)). In equilibrium, bidders will bid late in an eBay-type auction in order to reduce the chance their private information is revealed to other bidders. To distinguish this "common values" explanation from other explanations, the paper compares the propensity of bidders to bid late on new cars to their propensity to bid late on used cars. Further, the paper compares the propensity to bid late on newer used cars to the propensity to bid late on older used cars. If Bajari and Hortacsu (2003) are correct, the propensity to bid late should increase as quality becomes a more important characteristic of the car and the expected quality of the car decreases. That is, there should be more late bidding on used cars relative to new cars and more late bidding on older used cars relative to newer used cars. The paper's empirical results are not consistent with the "common values" explanation. The paper finds there is more late bidding on new `Vettes relative to used `Vettes and more late bidding on younger `Vettes relative to older `Vettes.

The winner's curse refers to the situation where the winner of the auction is the person that over values the item. For example, assume different bidders have different private information about the quality of a used `Vette. In this case, some bidders will have information suggesting the car is of high quality and these bidders will over value the car. Some bidders will have information suggesting that the car is of low value and

`Vettes and Lemons on eBay

Page 4

these bidders will under value the car. On average, bidders will be close to the correct valuation of the car, however it is not the average bidder that wins the auction, the highest bidder wins. Rational bidders will realize that there is a winner's curse problem and discount their bid accordingly. These bidders face a trade-off between over paying for the car and losing the auction. According to Haile et al (2003) the larger the number of expected bidders, the greater the amount each bidder will discount his or her bid. The paper tests the hypothesis by regressing bids on used cars on observable characteristics of the cars, sellers and buyers and on the expected number of bidders. As discussed below, it is necessary to use a proxy for the expected number of bidders and for this, the paper uses the number of days of the auction. The paper also presents results from an "order statistics" regression analysis which accounts for censoring of bids and bidders. The results suggest bids increase very slightly with the number of expected bidders. The paper finds that bidding behavior on used `Vettes is not consistent with the winner's curse and the existence of information problems in the market.

It is surprising that one of the most famous theoretical results in economics, Akerlof's lemons hypothesis, lacks substantial empirical support. Bond (1982, 1984) and Lacko (1986) use data on the actual repair costs of pickup trucks and cars (respectively) sold in the used market to determine if there is evidence of a lemons problem. Bond (1982, 1984) finds that for used trucks bought on the used car market when these trucks were less than ten years old, there is no difference between the distribution of repair costs of the trucks bought used and the trucks bought new. This result suggests the quality of late model trucks is similar whether or not the truck is sold in the new or the used market. Lacko (1986) analyzes the distribution of repair costs for used cars bought through a variety of channels including friends and family, new car dealers, used car dealers and newspaper classifieds. Lacko (1986) argues that friends and family are unlikely to hide important private information about the quality of the car. Lacko (1986) finds that for cars less than seven years old the distribution of repair costs is similar for all used cars, irrespective of the channel of distribution. Both Bond (1982, 1984) and Lacko (1986) find that the quality of vehicles sold in the used market is lower once those vehicles get older. This result is consistent with a lemons problem in the market for older used cars, but it is not consistent with Akerlof's (1970) explanation for the steep decline in the value of a car in its first year of ownership.

`Vettes and Lemons on eBay

Page 5

More recent work similarly finds evidence of a lemons problem for older used cars but not for newer used cars. Engers et al (2004) analyze ownership durations and argue that the model presented in Akerlof (1970) suggests ownership durations will be positively correlated. Lemons will be "hot potatoes" quickly passed from one owner to the next. The authors find no evidence that short durations for the first owner are associated with short durations for later owners. A result that is not consistent with Akerlof's lemons hypothesis. However, the authors do find that cars with very long initial ownership durations have subsequent ownership durations that are short and positively correlated. The authors argue this pattern is consistent with neglect by the first owner causing quality problems for later owners. Porter and Sattler (1999) develop upon work by Hendel and Lizzeri (1999) and Bulow (1982) and estimate a general model of the used car market which includes Akerlof's model of information asymmetries with buyer preference for used cars. Porter and Sattler (1999) argue the empirical pattern of trade is consistent with a preference for used cars and not consistent with asymmetric information. Other papers do find some empirical support for Akerlof's hypothesis.1

This paper presents the most direct empirical test of Akerlof's hypothesis, albeit in a relatively new market. The most straightforward test of Akerlof's hypothesis would be a survey of car buyers in which survey participants are asked to write down their value for a new car and a similar used car. If there is a premium for a new car, then the result would be consistent with Akerlof's lemons hypothesis. In this paper we come close to this test and to some extent we are able to do better because on eBay bidders must "put their money where their mouth is." The second price auction mechanism used by eBay suggests rational bidders will bid their expected value for the car, at least by the end of the auction (Vickery (1961), Bajari and Hortacsu (2003), Song (2003)). Below, we assume bidders do in fact bid their value for the car and so we are able to compare bids across new and used `Vettes to estimate the "new car premium." We are cognizant of various factors that may bias estimates of a car's value using bids on eBay including censoring of later bids, censoring of late entering bidders, discounting to account for

1 Genesove (1993) finds some evidence supporting Akerlof's lemons hypothesis by using data on dealer behavior and preferences. Emons and Sheldon (2002) analyze used car sales in Switzerland and find some support for the lemons problem but state that it is not a widespread problem. Most recently Wolf and Muhanna (2005) argue that faster depreciation of used cars on eBay relative to Kelley Blue Book is evidence of a lemons problem on eBay Motors.

 ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download