The Effect of Culture on the
Winner’s Curse in Internet Auctions:
An Empirical Study using eBay
Wilco Alberda
S1256408
May, 2007
Rijksuniversiteit Groningen Faculty of Economics
Department of International Economics & Business Supervisors: Dr. K.S. Mühlfeld
ABSTRACT
TABLE OF CONTENTS
INTRODUCTION... 4
THEORETICAL BACKGROUND ... 7
METHODS... 16
Data and Sample... 16
INTRODUCTION
In 1995 the online auction site AuctionWeb was founded by computer programmer Pierre Omidyar. It was part of a personal website which contained information on an internet virus. The first item he placed for sale on AuctionWeb was a broken laser pointer that he sold for $13.83. He was completely surprised that he was able to sell his broken laser pointer, but the buyer claimed that he collected broken laser pointers. This is how AuctionWeb started and grew until 1997, when it became eBay, after Omidyar’s consulting firm Echo Bay Technology group (Cohen, 2002).
The importance of internet auctions grew rapidly during the last decade. eBay, for example, the largest consumer to consumer online auction site, had in the third quarter of 2006 net revenues on the marketplace of $1.049 billion, a growth rate of 22% over the $858 million reported in the third quarter of 2005. The eBay platform confirmed registered user base at the end of the third quarter 2006 totalled 212 million, representing a 26% increase over the 168 million users reported at the end of third quarter of 2005. The total value of all successfully closed items on the eBay trading platforms was $12.6 billion, representing a 17% year-over-year increase from the $10.8 billion reported in the previous year (eBay’s third quarter 2006 financial results).
These figures show the increasing use of internet auctions by consumers (and businesses), as almost everything can be auctioned on the site. The diversity of products being sold is immense: from food to cars, coins to iPods, comics to antiques and even art.
An auction is defined by McAfee and McMillan (1987) as: “a market institution with an explicit set of
rules determining resource allocation and prices on the basis of bids from the market participants.”
Auctions incur transaction costs as the buyers and sellers are brought together by a third party; a traditional auction house such as Christie’s or Sotheby’s and nowadays internet auctions such as eBay. However, transaction costs in auctions have been reduced with the upcoming of internet auctions, as buyers and sellers are brought together more efficiently with extensive listings and powerful search technologies. Furthermore, the costs of running a website are much lower than the costs of a live auction. Live auctions contain auctioneers, travel costs and costs of delay. Costs of delay occur in live auctions as the object cannot be sold immediately. The seller has to wait until the next live auction, whereas sellers on the internet can place their product immediately on the web and sell it. However, it faces a trade-off, since less information is available for online bidders. A bidder cannot experience the object in a live setting; one has to deal with the information available on the site, such as pictures and descriptions (Kazumori and McMillan, 2005). eBay has created a feedback mechanism to help potential buyers and sellers evaluate each other, as to increase trust on the marketplace.
being auctioned is valued the same by each participant, due to lack of information. Sometimes products in auctions lack information on quality or features. This especially occurs on internet auctions, as bidders can not experience the object in a live setting and all bidders have the same information that is provided on the website. All bidders value the product using the same available information. However, each bidder has private signals. Private signals can be different or additional information on the object or different interpretation of the existing information. Therefore there is no such thing as a pure common value auction.
The winner is cursed as he or she overpays for the product. Bidders in common value auctions are subject to winning the object and paying too much, since they use the available, incomplete and often asymmetric information and therefore eventually ‘lose’ the auction. Every bidder has private signals that provide an estimate of the true value, on which he or she bases her bid. Winning the auction means that the winning bidder has valued the object the highest. The real true value, the common value, of the item is unknown. Therefore the winning bidder could well have placed a bid above the real true value, as all other bidders had lower private signals (risk neutrality assumed). In these types of auctions a bidder is subject to the winner’s curse; however participants can adjust for the winner’s curse by bid shading (bidding less than their true value). Buyers must bid more conservatively the more bidders there are, because the chance that there is someone with a lower value (signal) is greater, which increases the chance of overpaying (Kagel and Levin, 1986). A good example of the winner’s curse is the famous ‘Jar Experiment’ that will be explained in the theoretical background section. The influence of culture on the winner’s curse will be investigated. The focus will be on whether culture influences bidder’s adjustment for the winner’s curse. Adjustment for the winner’s curse means that bidders take into account the likelihood of a winner’s curse in a common value auction with incomplete information and therefore bid lower or less. Hofstede’s (1980, 1991 and 2001) uncertainty avoidance index will be used as a cultural measure. Do bidders from high uncertainty avoidant cultures adjust more for the winner’s curse? This is the question that will be attempted to be answered in this research. The reason for the choice of this particular research question is that the winner’s curse has never been linked to culture in any research. This research is therefore the first to attempt to find a relation between the winner’s curse and culture (UAI).
THEORETICAL BACKGROUND
The field of online auctions has increasingly been covered by academic research the last decade. One of the topics discussed is ‘sniping’, which is a much discussed topic. Sniping will be discussed briefly, since it is an important phenomenon occurring in online auction and it can be a mean to protect oneself for the winner’s curse.
eBay uses a ‘hard close’ or a fixed end time. The length of the auction, which can be 3, 5, 7 or 10 days, is predetermined and will not be extended for a certain amount of time after the last bid is placed (ebay.com). This created a phenomenon called ‘sniping’, last-minute bidding. Bidder’s on online auctions, especially eBay with its’ hard close, tend to hold its bid up to the last minutes or even seconds of the auction. Several explanations are given (Roth and Ockenfels, 2002). First, implicit collusion against the seller, as bidders withhold their bid until the last minute, they cannot react on each other’s bid, and therefore no ‘price wars’ exist. Another reason can be that sniping is used as a protection measure against naïve English-auction bidders. Some bidders on eBay might not understand the proxy bidding mechanism and therefore bid incrementally, like in the traditional English auction. By placing a snipe bid the incremental bidder cannot react to that bid, and as a result, the sniper might win the auction at the incremental bidder’s initial low bid, assuming no other players participate in the auction.
have to give a phone number, credit card number or home address in order to obtain a verification number, to be able to register at eBay. However, the measures against fraud paradoxally are open to fraud as well.
In a seminal theoretical paper, Akerlof (1970) argues that when asymmetric information is present in market settings, potential buyers rationally lower their willingness to pay, resulting in a suboptimal number of exchanges: market failure (lemon market). Without a mechanism for sellers to develop a reputation, dishonest sellers might drive out honest ones.
Solely low quality products on internet auctions would then be the result. Sellers can falsely describe product or omit important details on the quality of the product. Also, buyers and sellers can both defect by not paying and not delivering, respectively. These problems concerning trust and reputation reduce the efficiency of the internet auction. To reduce the risk of a ‘lemon-market’ (Akerlof, 1970), eBay has created a feedback mechanism. This is a platform on which every buyer and seller can leave feedback on each other after a transaction has been made.
Whether the feedback mechanism of eBay is beneficial for both buyers and sellers can be inferred from the massive amount of transactions that is taking place daily on eBay. On the other hand, Resnick and Zeckhauser (2001) show that only 0.3 percent of all feedback by buyers about sellers was negative or neutral. This can be interpreted that fraud does occur only rarely and that most are satisfied with the transaction and product. Another interpretation is that players are afraid of retaliation. Cabral and Hortaçsu (2004) have shown that 40 percent of negative feedback is retaliated.
Most studies, however, focus on the probability of selling and the premium paid for good reputations. For example, Ba and Pavlou (2002) studied four items on eBay: music CDs, modems, Windows Server software CDs and digital camcorders, and found that buyers valued a good reputation, especially for higher priced items. Dewan and Hsu (2001), researched a generalist (eBay) versus a specialty (Michael Rogers, Inc.) online auction and showed that buyers were willing to pay a premium on the specialty online auction for collectible stamps, since the stamps get a reliable value assessment by a third party. However, Eaton (2002) studied guitars on eBay and found that negative feedback reduces the probability of sale, but not the price of sold items.
Vishwanath (2004) compared the effect of seller feedback rating on bidder participation in three different cultures (Canada, France and Germany). He took interpersonal trust by Inglehart (1997) as a measure for cultural difference. He finds a significant interaction between interpersonal trust, feedback rating and bidder participation. Cultures with low interpersonal trust have relatively low bidder participation in online auctions with sellers having a relatively high negative feedback rating. The opposite goes for cultures with high interpersonal trust.
Japan and the US) using eBay. He used Hofstede’s (1980) Uncertainty Avoidance Index (UAI) as a cultural measure. He finds that information has a significant influence on the attractiveness of an auction in Japan (with a high UAI). However, information has a minimal effect on the number of bidders and the final price in the US and Germany (with lower UAIs). His study finds that the threshold for ambiguity differs among cultures. Bidders in Japan are more sensitive to information than American bidders.
Both studies of Vishwanath focus on cultural differences, which will be the focal point of this research. The first study is on trust and reputation among cultures, while the second is on the information effect. For the first research interpersonal trust is the cultural measure and for the second the uncertainty avoidance index by Hofstede (1980).
Hofstede (1980, 1991, and 2001) has made a comprehensive study on the influence of culture on
work-related values during 1967 to 1973, while working as a psychologist at IBM. He collected data from questionnaires from over a 100,000 employees of IBM in 53 countries. From this study he developed the initially four and later five cultural dimensions. These dimensions are:
• Power Distance Index (PDI): is the extent to which less powerful members of a society accept
the fact that power is distributed unequally. In societies with high power distance, everybody has his/her rightful place in society and status is important to show power. In societies with low power, status is less important.
• Individualism (IDV): In individualist societies, people look only after themselves and close
relatives. Identity is placed on individuals. In collective cultures, people belong to groups that look after each other in exchange for loyalty. Here identity is more determined by the group or network to which one belongs.
• Masculinity (MAS): In masculine cultures the dominant variables are achievement and
success. The dominant features in feminine cultures are caring for others and quality of life. Within masculine cultures there exist large role differentiation between males and females.
• Uncertainty Avoidance Index (UAI): Is the extent to which people feel threatened by
uncertainty and risk, and try to avoid them. There is great need for rules and structure. They tend to be less innovate and entrepreneurial.
• Long-Term Orientation (LTO): Is the extent to which a society exhibits a future-oriented
rather than a conventional historic or short-term perspective.
research. It is too strong an assumption that bidders from masculine cultures see auctions solely as ‘games’ or that status can be inferred from winning an auction. However, it is pure speculation on this behaviour without being able to obtain a reference in this field. Therefore a research on power distance and masculine behaviour could be executed, but it is a too an extensive research to execute for this thesis, as questionnaires are needed. Obtaining data from questionnaires is very difficult and time consuming, as the response rate is generally very low. Furthermore, a bidder remains anonymous during and after the auction and direct communication is needed for both dimensions to become visible for both players, as to be able to show status or masculinity. For these reasons masculinity and power distance are left out of this research.
Long-term Orientation can be applicable in the field of online auctions, when bidders show the thrift and perseverance to pursue their desired needs. In auctions this means that each bidder should bid until it attained its need. Here it is expected that the need is to get an object at a desired price; accordingly a (smart) bidder from an LTO culture will participate in many online auctions for that particular item with its maximum bid equal to the desired price and persevere until his or her need is fulfilled. Here it is assumed that LTO does not mean that the bidders’ need is to obtain the item and bid until the item is obtained, and therefore is likely to overpay (winner’s curse). It is difficult to measure LTOs influence on bidder behaviour with respect to the winner’s curse. The only possible way to measure this is to look at each bidder individually, to see whether each bidder bids in multiple auctions in the way described above, but this is beyond the scope of this research. However, in this light, bidders from cultures with an LTO should be less receptive to the winner’s curse.
Individualism is the extent to which bidders strive only for self-enhancement. Bidders from collectivist cultures prefer face-to-face business and are therefore less likely to participate in online auctions. Provided that bidders from collectivist cultures are unlikely to participate in online auctions, there is unlikely to be a useful online database from these cultures. Indeed, when looking for data from these culture countries little was found. This will be explained in further detail below.
Lastly, uncertainty avoidance affects the tolerance of ambiguity and the trust in ‘opponents’. The Uncertainty Avoidance Index (UAI) will be used as a cultural measure. The UAI will likely affect online bidder behaviour. It is expected here that bidders from an uncertainty avoidant culture will, for example, do more research before placing a bid or will bid more cautiously. The UAI has already extensively been used in a wide field of academic researches. For example, Mueller and Thomas (2001), link the UAI and entrepreneurial potential, Money and Crotts (2003) test the effect of uncertainty avoidance on information search, planning, and purchases of international travel vacations, while Lim et al. (2004) research the effect of uncertainty avoidance on internet shopping.
suffered unexpected low profit rates in the 1960’s and 1970’s on lease sales. Not surprisingly, many economists greeted this claim with scepticism, because such a claim implies repeated errors and a departure from rationality, or at least from equilibrium predictions.
The empirical literature does not reach a clear and unambiguous conclusion about the existence of the winner’s curse, perhaps because there are many obstacles that complicate a study based on ‘real life’ data. Therefore the first, and famous, laboratory experiment was conducted in 1983 by Bazerman and Samuelson. The ‘Jar Experiment’ is a well known experiment that showed the implications of the winner’s curse. Bazerman and Samuelson (1983) conducted 48 first-price sealed-bid auctions with jars containing coins worth $ 8.00 to classes of MBA students, with the value unknown to the subjects (common value auction). While the average estimate of the jar’s value was $ 5.13, the mean winning bid was nevertheless an overbid, $10.01. Therefore, the winners lost on average $ 2.01 of wealth. Kagel and Levin (1986), Dyer et al. (1989) and Kagel et al. (1989) also performed laboratory experiments with sealed bid auctions that provide evidence of “judgmental errors” which leads the authors to question the applicability of existing auction models. They find the existence of the winner’s curse. According to Lind and Plott (1991), the difference between real life data and experimental data is that they behave differently, because subjects of the experiments face limited liability. Therefore, the internet, providing a large and easily accessible (real life) database, has offered new opportunities for testing the winner’s curse.
However, there has not been much research on the winner’s curse in online auctions. The first to investigate the existence of the winner’s curse in online auctions were Mehta and Lee (1999). The winner’s curse was found, and depending on bidder experience and age of the product. Less experienced bidders tend to be more subject to the winner’s curse as are products that are relatively new on the market, due to lack of information.
The most well known study is from Bajari and Hortaçsu (2003). They have tried to find whether the winner’s curse exists and whether participants adjust for the curse. According to the Milgrom and Weber (1982) model, bidders will rationally lower their bids in a common-value auction as to prevent a winner’s curse from happening. Bajari and Hortaçsu (2003) create a structural model of bidding and use this model to find, among others, whether bidders adjust for the winner’s curse. They do find that participants adjust for the winner’s curse. On average bidders bid 10 percent less than their private value signals and the entry of an additional bidder causes them to decrease their bid by 3.2 percent. Jin and Kato (2002) conducted a field experiment with baseball cards on eBay. They bought baseball cards on online auctions to sell them later on eBay. They find that buyers do adjust for the winner’s curse but insufficiently.
measure directly whether a particular bidder will adjust for the curse, but all bidders on average. They find that bidders do not fully adjust for the winner’s curse, mainly because bidders lack information on the number of bidders; the price varies as the actual numbers of bidders exceeds or falls short of the number expected.
This research fills the gap between the researches of Dewally and Ederington (2004) and Vishwanath (2003). Both research the number of bidders and the final price, even if it is for a different reason. Vishwanath (2003) investigates the information effect on the final price and the number of bidders. More information should lead to more bidders and higher prices. Dewally and Ederington (2004) investigate the relation between the number of bidders and the final price, whether the number of bidders influences the final price. In this research the winner’s curse investigated by Dewally and Ederington (2004) and the cultural dimension (UAI) researched by Vishwanath (2003) will be combined, to find whether bidders from uncertainty avoidant cultures do adjust their bid for the number of bidders. In other words, this research tests whether cultures with a high UAI are less receptive to the winner’s curse than cultures with a low UAI.
All researches used different methods in finding the winner’s curse or the adjustment thereof. In field experiments it is easier to encompass all variables describing the winner’s curse. A field experiment uses an internet auction (e.g. eBay) as a platform, but in a controlled setting. It is an experiment in a naturally-occurring environment rather than in a laboratory. A structural model gives a detailed description of the variables included, but the more variables included, the more difficult the model becomes.
The framework from Dewally and Ederington (2004) will be used for this research, because it is from the methods mentioned above the most feasible and practical framework. It is easily applicable for cross-cultural studies, since it is not as time consuming or costly as the others researches. Field experiments need investments as items are to be bought before they can be sold on the net. Structural models are quite time consuming as for creating a model much experience is necessary. This research lacks the means to create a structural model or execute an experiment.
FIGURE 1
Framework by Dewally and Ederington (2004)
The lower box is called “Positive Relationship between Number of Bidders and Final Price”, but could also be called “Winner’s Curse” in the cases of “Failure to Recognize the Winner’s Curse” and “Lack of Information”. It is called differently, because the first box “Private Value Component” does not lead to a winner’s curse and therefore a more general term is necessary. Hence, throughout this research positive relationship between the number of bidders and final price is used interchangeably with winner’s curse, as the former is a measure for the latter.
The first indicated reason is the private value component. This means that bidders attach a private value to a product and thus it does not have a pure common value. An increase in the number of bidders increases the expected price since it increases the sample of private values from which the highest is drawn. The product chosen (iPod) for this research falls in the consumer electronics category. Consumer electronics tend to have a common value and are little subject to a private value, which will be discussed below. In this paper it is assumed that consumer electronics do have negligible private value.
The second reason is that bidders fail to recognize the winner’s curse and therefore do not adjust for it. Bidders from uncertainty avoidant cultures are more likely to recognize the winner’s curse for two reasons. First, it is expected that they are more likely to correct their price for the number of bidders
participating, as uncertainty avoidant bidders are more careful bidders. They will likely monitor the bid pattern and the number of bidders before placing a bid, against the less uncertainty avoidant bidders who are more likely to just place a bid, without taking into account the number of bidders. The hypothesis is composed into two parts: the first is the relationship between the number of bidders and the final price and the second is the uncertainty avoidance. The hypothesis has been set up by using the articles of Dewally and Ederington (2004) and Vishwanath (2003). Dewally and Ederington’s study found that the final price increases with the number of bidders for the US, which is a culture low in uncertainty avoidance. Then the direction of the relationship for uncertainty avoidant cultures has been generalized using Vishwanath (2003) to state the following hypothesis:
Hypothesis 1a: The relationship between the number of bidders and the final price of
an online auction is positive for cultures low in uncertainty avoidance, while the relationship is insignificant for cultures high in uncertainty avoidance.
is expected that bidders from cultures high in uncertainty avoidance are likely to always bid lower than the lowest price available on the internet, while bidders from low uncertainty avoidant cultures place bids around that price, as they do less research. Their bids will show a lower average deviation than bidders from cultures higher in uncertainty avoidance, as they will both have positive and negative deviations (price at national electronic shop − price at eBayi), which decreases the average deviation.
Bidders from higher uncertainty avoidant cultures will mainly show positive deviations that increase the average deviation.
Hypothesis 1b: Cultures high in uncertainty avoidance will have a larger positive
deviation between the final price at eBay and the price at the lowest national electronic shop than less uncertainty avoidant cultures.
Finally, assuming that consumer electronics are common value items and that bidders do not fail to recognize the winner’s curse, they do not have full information on the number of bidders (Dewally and Ederington, 2004). Participants will expect a number of bidders on certain estimations and place a bid accordingly. They will fail to adjust for the winner’s curse when there are unexpected bidders (number of actual bidders – expected number of bidders). This only occurs in cultures with high uncertainty avoidance, as these bidders are more likely to do research on the expected number of bidders and place a bid accordingly. High UAI cultures do not tolerate uncertainty and will put much effort into reducing it (Hofstede, 2001). Bidders from low uncertainty avoidant cultures are more likely to just place a bid, without taking the expected number of bidders into account, and a positive deviation between the actual and expected number of bidders will not influence the final price in low UAI cultures. The winner’s curse in this hypothesis is explained by the lack of information.
Hypothesis 2: A positive deviation between the actual and expected number of bidders
METHODS
Data and Sample
The countries used for this research are the United States, United Kingdom and France. Three countries are chosen as to take a large spread of UAIs. The country economic indicators and indexes are shown in table 1.
TABLE 1
Macro Economic Data on United States, France and United Kingdom, and Hofstede’s (1980, 1991, 2001) Cultural Dimension Indexes
United States France United Kingdom Population (2006) 301,013,0001 61,045,0002 60,536,0003 GDP (PPP) $ per capita (2005)4 41,399 29,187 30,436 Internet users (2006)5 232,057,067 30,837,592 37,600,000 Percentage Internet Users (Population 2006) 77,1% 50,5% 62,1% Hofstede (2001) long-term orientation index 29 -- 25
Hofstede (2001) masculinity index 62 43 66
Hofstede (2001) uncertainty avoidance index 46 86 35 Hofstede (2001) individualism index 91 71 89 Hofstede (2001) power distance index 40 68 35
1. Source: US Census Bureau
2. Source: Institut National de la Statistique et des Études Économiques (INSEE) 3. The UK Office of National Statistics
4. Organisation for Economic Co-operation and Development (OECD) 5. Internet World Stats
As can be seen from table 1, the uncertainty avoidance indexes of the United Kingdom, United Stated and France are, respectively 35, 46 and 86. This sample is taken from a larger range of uncertainty avoidance indexes; from 8 (Singapore) to 112 (Greece). Even though the countries in the sample have a quite diverse UAI, all three are developed countries according to the OECD. They all have a relatively high GDP per capita, at least 50% of the population has internet and they all share a ‘western’ culture. Therefore, it could be likely that the bidder behaviour on the internet will be more or less the same among these countries. However, this study controls for the potential differences in bidder behaviour that result from different levels of economic development by taking a sample of countries that includes only developed countries. Even though the differences are not as clear as between Asian and European nations, other studies (e.g. Vishwanath, 2003, 2004) do find differences in bidder behaviour, regarding information effects.
the auction immediately by bidding the required amount for that option, which is stated by the seller). The importance of relationships in collectivist cultures is higher than in individualistic cultures (Hofstede, 1980). Hence, it is expected that buyers from collectivist cultures, such as Singapore, attach much value to face-to-face business and for that reason are weary to bid online. Consequently almost no iPods are sold online, which limits the available data. This also applies to data on other Asian countries that have a local eBay site like Malaysia, India and China.
eBay has acquired in September of 2001 19.5% of MercadoLibre, the Latin America's number-one auction site and eBay´s exclusive partner for the Latin American region. However, MercadoLibre.com does not have the same features as eBay.com, even though they are quite similar. One feature that they do not have in common is the ‘completed listings’ feature. As a result, data from South American countries is unobtainable. Therefore, only western countries were left too choose from. The countries were chosen as to maximize the spread of the UAI.
The iPod Video 30 GB is chosen as the object for this research. A product with a common value should be chosen, because the winner’s curse only occurs in common value auctions. Technological products have a common value as they can easily be resold, which is a feature of common value products (Bajari and Hortaçsu, 2003). Furthermore, it is unlikely that someone will value the iPod much more than another, because he or she attaches emotional value to the iPod, or is part of someone’s collection. Individuals valuing the iPod less will not buy an iPod. Emotional value is only attached to the iPods once it has been bought, because the owner puts its own music on the iPod. This gives the iPod a private value component, but a unique one, since it has been ‘custom made’ for just one owner. Once the iPod is going to be resold the music can easily be erased to be of use for the new owner. Therefore, iPods on the market are assumed, for this research, to have negligible private value. Additionally, the features of electronics can easily be specified, which decreases the complexity and errors of data collection. An iPod video 30 GB is an object which cannot be classified in much other than new and used. The iPod Video 30 GB is still relatively new (about 1 year), so most used iPods are still in good condition. Damaged and/or scratched iPods were left out of the sample.
The value of an iPod can easily be inferred from other trading channels, such as from traditional and electronic shops. Therefore, the information asymmetry is low and should prevent the winner’s curse from happening, or at least limit its occurrence. This depends of course also on the time spent searching by participants in online auctions. When seeking thoroughly, one is more likely to find a lower priced iPod and will lower its’ maximum bid (Mehta and Lee, 1999).
The iPod Video 30 GB was chosen, particularly, because it is an iPod that is sold best on eBay, and consequently gives the largest database.
in the sample. The UK also sells a lot of iPods online and for France a census is taken; all data available on eBay.fr during the months October, November and December will be used for this research.
The data collected from eBay includes: the final price, number of bids, the number of bidders, the feedback rating of the seller, number of bids in the last ten minutes, whether the iPod is new or used, a reserve price/starting bid dummy and whether the iPod is sold by a Power Seller.
A delayed start was made with collecting information on the reserve price and/or starting bid. Data was already being collected when the reserve price was included as a variable and could not be retrieved from the online database, since it is available only for about thirty days. This means that the sample including reserve prices is smaller than the whole sample.
The used iPods are conditioned on that they are in good shape, which they generally are. Unsold iPods and sold through the buy-out option were not included in the sample. iPods sold through the buy-out option are not sold in a traditional auction system and therefore are not testable on the winner’s curse. Unsold iPods are of no interest, because most of the auctions with unsold iPods had no or unreasonably low bids. Also reserve prices have obstructed actual sales.
The data collection period was from the 1st of November until the 21st of December. The data collected
includes iPods placed on eBay from the 15th of October until the 19th of December, a time span of two
months. Data collection started on the 1st of November by collecting data of iPods from the 15th of
October, the ‘oldest’ iPods in the database. From then on the data was collected daily for France and every other day for the UK and the US. This was done because too many iPods are sold daily on eBay, and hence it was impossible to keep up with the daily number of sales. Data was collected until the 21st
of December with iPods that were placed on the 19th of December on eBay. Time limitations restricted
the collection period to two months. The samples were taken conveniently, taken in an arbitrarily and unstructured manner from the websites.
The sample for the UK includes 1000 iPods, from a population of about 2300. This is an average of the population size, as the number changes continuously. The rate at which iPods are sold is never quite the same that is why the range of the database on eBay is between about 1800 and 3000 iPods. With a confidence level of 95%, the confidence interval is about 2.33.
The US has a sample size of 936 iPods, from a population of about 5200 iPods, which is again an average of the population size during the two months. The confidence interval is about 2.9, with a confidence level of 95%.
The census for France contains 439 iPods. This includes all iPods Video 30 GB sold from the 16th of
October till the 19th of December. A census means that the confidence interval is zero, as the whole
may show, since a census does not have unselected cases that a sample does have. A sample can have cases unselected that influence the results. However, the sample sizes for the UK and the US are quite large and therefore it is expected that there is only limited bias.
Measures
For each hypothesis a distinction will be made between new and used iPods. New iPods tend to be generally higher-priced than used iPods (with a few exceptions) and therefore might show different bid patterns. For example, bidders for used iPods could be more subject to the winner’s curse, as the information on quality and condition is more open to ‘fraud’ than the information for new iPods. New, unused iPods should still be in the box without any damage, whereas the used iPod could have undescribed damages or software damages that can not be (or is not) shown in the picture. Therefore it is likely that the information for used iPods is more often incomplete as there are more possibilities for fraud. Left out information can lead to the winner’s curse as bidders value the object more than it is worth. So each hypothesis will test the used and new iPods separately. Also, the sample of iPods sold by power sellers is being tested separately, because uncertainty avoidance can also affect the number of bidders in auctions with power sellers. Uncertainty avoidant bidders may be attracted to auctions with power sellers, as they have more experience selling iPods and are therefore more likely to be reliable. They have built up a good reputation and consequently will not have a reason to cause that reputation damage by selling faulty iPods or not delivering. While relatively new sellers can easily default, because they have not built up a reputation, they can easily re-enter under a new identity (Cabral and Hortaçsu, 2004). Some buyers know this and therefore tend to buy from power sellers.
Dependent Variables
The dependent variable for hypothesis 1a and 2 is the Final Price. The final price is the last and winning bid of an internet auction. The ‘proxy bidding’ system on eBay ensures that the bidder with the highest valuation pays one increment above the second highest bid. This is the winning bid and is therefore the price paid for the auctioned item. On eBay the final price is clearly stated in the completed listings.
The deviation between the lowest price available on an electronic (national) shop and the final price on eBay, for each iPod, will be used as the dependent variable for testing hypothesis 1b (Price at National Electronic Shop − eBay Pricei). It is assumed for this hypothesis that bidders keeping their bids below
these prices are likely to have done more online research.
are: pcworld.co.uk for the UK, AlwaysLowest.com for the US and Rebelio.com for France, as shown in table 2. The price of apple.com is shown as a benchmark for these prices in each country; generally prices on eBay are lower.
TABLE 2 Electronic Shop Prices
US UK FR
Apple.com $ 249,00* £ 189,00* € 300,96*
pcworld.co.uk -- £ 158,97 --
AlwaysLowest.com $ 234,99 -- --
Rebelio.fr -- -- € 264,00
* price including delivery
Independent Variables
The Number of bidders is the independent variable in hypothesis 1a. The relationship between the
number of bidders and the final price (dependent variable) will be tested. As the number of bidders
increase the final price will also increase for cultures low in uncertainty avoidance, as the bidders fail to adjust for the winner’s curse, according to hypothesis 1a. One should note that the number of
bidders is not the same as the number of bids. The number of bids is what is stated on the website of
eBay, whereas the number of bidders is to be counted manually.
The independent variable for hypothesis 1b will be a Country Categorization Variable. Here will be tested whether uncertainty avoidant participants will do more research than less uncertainty avoidant bidders. A larger average deviation (price at national electronic shop − eBay pricei) is expected for
cultures high in uncertainty avoidance (France) than for low uncertainty avoidant cultures (US and UK), as bidders from these cultures are likely to bid closely around the lowest price available. To be able to compare the deviations between the countries, all prices are converted into one currency, the dollar. The mid-market rates for the Euro and Pound Sterling where taken at the 22nd of November,
around the centre-date of the data collection period. The rates were, at closing time, 1.297 Euros and 1.971 GBP per dollar. The rate was taken as an estimate of the average over that period. Even though it is not the exact rate for the whole collection period, it is only used to give an indication of the value of the deviations for comparison.
for the price expected, as it is also the only information that the bidders have and therefore are likely to base their expectation on that same average. However, note that the average calculated for this research is not from the same sample (or census) as available to the bidders within this sample (or census). Data is only available for thirty days after the auction has been completed, which means that the database changes continuously. Therefore, the average number of bidders changes continuously too. Nevertheless, the number of bidders for each country is generally about the same for each auction. Hence, the average does not vary much.
The number of unexpected bidders will be calculated by subtracting the number of expected bidders from the number of actual bidders, for every auction. Here will be tested whether bidders lack full information and therefore a positive relationship between the number of bidders and final price is shown. A large positive deviation (actual – expected bidders) will lead according to the hypothesis to a higher price. Unexpected bidders are not included in the available information and hence bidders are unable to anticipate to these entrants, which lead to an increase in the final price. This means that a positive influence from the number of bidders on the final price is partially explained by a lack of information. However, it is expected that bidders from cultures low in uncertainty avoidance just place a bid, without taking into account the number of expected bidders. The influence on the final price in these cultures will only come from the number of bidders, and not from the unexpected number of
bidders.
A less than expected number of bidders is not included in this analysis, since it does not explain the winner’s curse, which is the focus of this research.
Control Variables
Control variables will be included to see what portion of the regression procedure that determines the variation in the dependent variable is explained by the control variables versus the independent variable, to reduce omitted variable bias.
The Feedback Rating and the Dummy of the Reserve Price/Starting Bid will be used as control variables for hypothesis 1a. The feedback rating might influence the final price, as a low feedback rating should indicate an unreliable seller, which decreases the price bidders are willing to pay, as the probability of default increases.
learning experiences with reserve prices. True sellers with auctions including a reserve price would not want to ‘lose’ their iPod for an unreasonably low price and hence place a reserve price, whereas sellers selling iPods with a lower quality, while indicating a new condition, gamble on getting as much bidders as possible to increase the final price and therefore do not use a reserve price.
The range of bids becomes smaller with a reserve price. When a reserve price is set, for example, at 100, the bid range from 0-100 has been replaced by a reserve price, which likely reduces the number of
bidders. Generally, most auctions do have bargain hunters, who bid at the lowest prices possible.
These bidders are now eliminated from the auction and therefore fewer bidders remain. For this reason, caution is necessary when using the reserve price as a control variable, as it also has a direct influence on the number of bidders, which is the independent variable. Therefore, tests for correlation and multicollinearity will be executed.
Note that the probability for success of an auction (the item is sold) decreases with a reserve price as the chance that a bidder has a value above the reserve price decreases. However, unsuccessful auctions are not included in the sample.
For this research it is assumed that a reserve price has a positive effect on the final price, as price is an extrinsic signal for quality (Tellis and Gaeth, 1990). Even though there are other explanations of the effect of the reserve price, as discussed earlier, here it is expected that when a reserve price is included, only serious bidders are left to bid and bargain hunters are left out. These serious bidders see in the reserve price a signal for high quality and hence start ‘serious’ bidding. This relates to both new and used iPods. The signal for high quality for new iPods is that the it has really not been used and still wrapped in its package. Using a reserve price creates trust for the bidders that the iPod is still unused. Since bidders infer high quality from the reserve price, they are willing to bid more, which increases the final price. However, there are some data limitations that complicate the collection of the reserve price.
The first data limitation is that no distinction can be made between the reserve price and starting bid. A starting bid is: “The price at which a seller wants bidding for an item to begin in an auction-style
listing. The starting price may not necessarily be the lowest price they are willing to sell their item for.” (ebay.com). Whereas a reserve price is: “The lowest price at which a seller is willing to sell an item in an auction-style listing” (ebay.com). After the auction, in the ‘completed listings’ of eBay, no
“may not necessarily be the lowest price they are willing to sell an item” and therefore only starting bids that are used in a reserve price kind of way, when they are used as a real minimum bid they are included in this research. This reduces the differences between the two.
The second data limitation is that no information is available after the auction on the amount of the reserve price or starting bid (ebay.com). Hence a dummy of the reserve price and/or starting bid has to be made, which does not make a distinction between the two. It is difficult to establish the exact level of the reserve price, but it is easier to determine that a reserve price has been used. Looking at the first bid generally gives an idea whether a reserve price or starting bid is used. When browsing through running auctions on eBay it was found that the first bid for most iPods start under 50 (in the local currency), which generally means that there is no reserve price and a negligible or no starting bid. Some iPods are sold with a high first bid, which indicates that the iPod had a reserve price or starting bid. This does not include the whole picture of differences between the levels of reserve prices, or the difference between a reserve price and starting bid. However, no difference between the reserve price and starting bid in bidder behaviour is assumed, as the information on the difference is unavailable. It is expected that it will influence the outcomes somewhat, but the low starting bids are left out by including only first bids of 50 and higher, which decreases the number of starting bids in the sample. For that reason there will be only looked at reserve price and high starting bids, using a dummy, of which is expected that they influence the final price positively.
For hypothesis 1b the control variables are the Number of Bids, the Dummy of the Reserve Price and the Feedback Rating. The reserve price and feedback rating are included for the same reasons as mentioned above. Also, the number of bids will be accounted for. Uncertainty avoidant bidders are likely to bid less aggressively, whereas bidders from cultures low in uncertainty avoidance bid more aggressively. Bidders from the latter culture are less careful in bidding and thus likely to place more bids (Vishwanath, 2003). This might influence the deviation between the final eBay price and the price of the electronic shop, as more bids generally lead to a higher final price, which influences the
deviation. This could partially explain the deviation, while it should be explained by the categorization
variable and therefore the number of bids will be included as a control variable. However, note that it is not always the case that more bids lead to a higher price; it also depends on the bid increments (the difference between two bids). It was too time consuming to include the bid increments, as this needs a separate and extremely large database on every bid placed in all 2375 auctions in this research and it is not the focal point of interest of this research.
Sniping will be controlled for in the second hypothesis, because more sniping leads to less available
the number of bids in the last ten minutes. It is possible to place a counter bid in the last ten minutes, but it is rarely seen on eBay. The data shows that most sniping is done between the last five and ten minutes of the auction. After the last five minutes almost no sniping occurs. This can be explained by the fact that bidders are afraid that their bid will not be transmitted, as this is the risk that comes with last minute bidding (Roth and Ockenfels, 2002). It seems that bidders from the chosen countries choose to snipe, but do not wait until the last minutes or seconds, because of the fear of a ‘bid failure’. Therefore, for this research, sniping is defined as the number of bids in the last ten minutes.
Statistical Techniques
First, a test for correlation will be executed between the independent variable and control variables. A test for correlation measures the strength and direction of the relationship between the independent variable and control variables. When there is a significant relationship between these variables, the results of the regression analysis that will be used to test the hypotheses will be biased, as the independent and control variables are related. The regression model will be influenced as high correlations cause problems when trying to draw inferences about the relative contribution of each independent variable. In that case the results should be interpreted with care. However, a test for multicollinearity will be used to test for the validity of correlation matrix. This will be discussed below.
A multiple linear regression analysis will be used to test for the hypotheses, because this shows the direction of the relationship between the dependent and independent variable. It goes beyond correlation by adding prediction capabilities in the form of a model. The regression models for the hypotheses variables look like this:
(1a) Y = 0 + 1 N + 2 F+ 3 RP+ (1b) YD = 0 + 1 C1 + 2 C2 + 3G+ 4 F + 5 RP + (2) Y = 0 + 1 U + 2 S+ where,
Y = final price RP = dummy of the reserve price
YD = eBay pricei - lowest electronic shop price C = country categorization variable
0 = the Y intercept of the regression line G = number of bids
1 , 2 , 3 and 4determine the slope of the model U = the unexpected number of bidders N = number of bidders S = number of snipes in the last ten minutes
In a multiple linear regression model, with five independent variables, Y is a linear function of X plus
.
0, 1 , 2 , 3 , 4 and 5 are referred to as the parameters of the model and is a random variable
referred to as the error term. The error term accounts for the variability in Y that cannot be explained by the linear effect of the independent variables. For
5 a split sample is used, as the data collection for
the reserve price had a delayed start.
For hypothesis 1a the model is explained by the independent variable (number of bidders) and the control variables (feedback and dummy of the reserve price). It is expected that the sign of
1, which
shows the relationship between the number of bidders and the final price, is positive for cultures low in uncertainty avoidance, as bidders fail to adjust for the winner’s curse.
2 is also expected to have a
positive sign, as a higher feedback should lead to a higher final price. It could partially explain the relationship between the number of bidders and the final price and hence is expected to weaken the relationship between them.
3 is expected to have a positive sign, as a reserve price increases the final price. The reserve price could partially explain the relationship between the number of bidders and the final price and hence is expected to influence the relationship between them. The number of bids (G)
is not included as control variable, as is done for hypothesis 1b, because the numbers of bidders influences the number of bids and therefore are too highly correlated (see correlation in tables 3, 4 and 5). A control variable should not be related by too much to another independent variable and therefore will be looked at the Variance Inflation Factor (VIF) of the dummy reserve price, to see whether it shows multicollinearity. Multicollinearity exists when there is a strong correlation between two or more of the independent variables in a multiple regression model. When there is multicollinearity the reserve dummy is redundant, which leads to overfitting of the regression model. The variable should then be left out of the analysis, as it only should explain the dependent variable and not other independent ones. As a rule of thumb, a VIF of more than 10 indicates a multicollinearity problem. The regression model for hypothesis 1b has a country categorization variable (C) as independent variable. This dummy is used to indicate whether there is a difference among the countries in the
deviation between the prices at the lowest available electronic shop price and eBay. Two country
dummies are included to analyze the three countries, as with k levels, k-1 dummy variables should be included in the analysis. The values of C1 and C2 are as follows:
Country C1 C2
France 1 0
UK 0 1
Here, C1 is one if iPods are sold in France and C2 is zero if otherwise. C2 is one if iPods are sold in the
UK and C1 is zero when sold in France or the US. When iPods are sold in the US, both C1 and C2 are
zero. Entering these values in the equation (1b) gives the following results:
(3) E (YD US) = 0 + 3G+ 4 F + 5 RP + (4) E (YD France) = 0 + 1 C1+ 3G+ 4 F + 5 RP + (5) E (YD UK) = 0 + 2 C2 + 3G+ 4 F + 5 RP + Thus,
0 is the expected value of the deviation for the US, while 1 is the difference between the deviations of the US and France and
2 is the difference between the deviations of the US and UK.
Additionally to the feedback rating and the dummy of the reserve price, which have the same expected relationship as under hypothesis 1a, is included the number of bids as a control variable.
3 shows the
relationship between the number of bids and the deviation. It is expected that
3 is negative for France
(high UAI) as the average deviation decreases with an increase in the number of bids. The average price at eBay and the electronic shop price converge. While, for the US and UK (low UAI)
3 has a
positive sign as the deviation increases. The average price at eBay increases and therefore diverges from the electronic shop price.
The last regression model is for hypothesis 2 with the unexpected number of bidders as the independent variable.
1 shows the relationship between the unexpected number of bidders and the final price. It is expected that unexpected bidders, in cultures with a high UAI have a positive effect on
the final price,
1 is positive. A 1close to zero is expected for cultures low in uncertainty avoidance.
The number of snipes is controlled for, where
2 shows the relation between the control variable and
the final price. The sign of
2 is expected to be positive as snipe bids do increase the final price.
Once the regression analyses are conducted with SPSS 15.0, the adjusted R2 will be looked at, to see to
what extent the model explains the variance. Adjusted R2 is a modification of R2 that adjusts for the
number of independent variables in a model. When adding independent variables, some of the increase in R2 would be simply due to chance variation in the sample. Adjusted R2 corrects for this and hence
will always be lower than R2. Low percentages of the adjusted R2 mean a poor fit.
standardized coefficient , whether it is positive, negative, large or small. The t-test will be used to find out whether is significantly different from zero, or, in other words, the relationship is statistically significant. The results will be tested for significance with one-tailed tests for the low uncertainty avoidant cultures (France) in hypothesis 1a, all cultures in hypothesis 1b and high uncertainty avoidant cultures (US and UK) in hypothesis 2, as the relationship is indicated in one direction in the hypotheses. Two-tailed tests will be used for high uncertainty avoidant cultures in hypothesis 1a and low uncertainty avoidant cultures in hypothesis 2, with unspecified relationships between the dependent and independent variables in the hypotheses.
Robustness Analysis
A robustness analysis is used to check whether the assumptions of the regression model hold. To check whether the regression models for hypothesis 1a are robust, the data for all countries is pooled and a country dummy is included. A new regression model is made that includes an interaction effect between a country dummy and the number of bidders. The interaction effect measures the joint effect of the country dummy and the number of bidders on the final price. The model is shown below:
(6) Y =
0 + 1 N + 2 C1+ 3 C2 + 4 N*C1+ 5 N*C2 + 6 F+ 7 RP +
where,
Y = final price N = number of bidders
0 = the Y intercept of the regression line C = country categorization variable 1, , 2 , 3 , 4 , 5 , 6 and 7determine the slope
of the model
N*C = interaction effect of number of bidders and
the country categorization variable
F = feedback rating RP = dummy of reserve price
= error term
And the country dummies are organized as follows:
Country C1 C2
France 1 0
UK 0 1
US 0 0
The pooled data is used to check whether the effect of the number of bidders on the final price is different between countries, which is measured by the interaction effects N*C1 and N*C2. The US is
the ‘reference’ country, with both country dummies equal to zero. France (C1) and UK (C2) are
compared with the US.
and the UK, respectively, on the final price is different than the effect in the US. Again the feedback rating and the dummy of the reserve price are controlled for. A t-test will be used to test for its significance. The interaction variable can be especially subject to multicollinearity and therefore will be tested using the Variance Inflation Factor.
RESULTS
First some descriptive statistics and correlations will be given.
In table 3 the descriptive statistics of the sample of the iPods sold in the UK are given. The final price is between 92, - and 195,-, which include both new and used iPods. The mean price is about 141,-. The number of bidders participating in an auction is between 1 and 20, with an average of 8.45. The average feedback rating is about 99% and there is an average of about three snipers in the last ten minutes. The sample of the reserve price is much smaller, due to the delayed start of collection. Out of the sample size of 495 of which the reserve price is collected, only 147 sellers actually used one. Furthermore, 776 iPods in this sample are new and 224 are used.
The matrix shows significant correlation between the dummy of the reserve price and the number of
bidders (-0.657) for the UK. This is not an extremely high correlation, but could cause problems
interpreting the model. To verify the results from the matrix a multicollinearity test will be executed after the regression analysis. Also, the number of bids and the number of bidders are significantly highly correlated (0.813) as was discussed earlier. For that reason they are not included in the same analyses.
The final price of US iPods is between $ 127.5 and $ 350,-, which include many used iPods (table 4). Only 389 iPods are new in this sample and 547 are used. The average price that American iPods are sold is about $ 202. The number of bidders reached up to 39, with an average of about 9.7. The average feedback rating is about 98.6% and the average number of snipers in the last ten minutes is about 4. Data collected including the reserve price amounts 371 that about half (183) the sellers actually used.
The table shows significant correlation (-0.475) between the number of bidders and the dummy of the reserve price, which is not very high. The correlation between the number of bids and the number of bidders is high with 0.750 and significant and therefore are not included in the same analyses.
TABLE 3
Descriptive Statistics and Bivariate Correlations UK
Variables Mean SD Cases Correlations
1 2 3 4 5 6 1. Final Price 141.0301 15.64704 1000 1 2. Bids 17.29 8.850 1000 0.144** 1 3. Bidders 8.45 3.776 1000 0.215** 0.813** 1 4. Feedback 99.047 2.9293 1000 0.088** 0.002 0.012 1 5. Snipe 3.02 2.832 1000 -0.023 0.403** 0.398** 0.034 1 6. Dummy Reserve .30 .457 495 0.149** -0.612** -0.657** -0.016 -0.032 1
** Correlation is significant at the 0.01 level (2-tailed).
TABLE 4
Descriptive Statistics and Bivariate Correlations US
Variables Mean SD Cases Correlations
1 2 3 4 5 6 1. Final Price 202.0255 29.23717 936 1 2. Bids 20.43 10.138 936 0.092** 1 3. Bidders 9.69 4.376 936 0.075* 0.750** 1 4. Feedback 98.649 3.9638 936 0.091** 0.002 0.016 1 5. Snipe 4.05 3.938 936 -0.175** 0.299** 0.279** 0.045 1 6. Dummy Reserve .49 .501 371 0.063 -0.549** -0.475** -0.042 -0.042 1
*Correlation is significant at the 0.05 level (2-tailed) **Correlation is significant at the 0.01 level (2-tailed)
TABLE 5
Descriptive Statistics and Bivariate Correlations France
Variables Mean SD Cases Correlations
1 2 3 4 5 6 1. Final Price 210.3203 19.46334 439 1 2. Bids 28.63 19.417 439 0.172** 1 3. Bidders 10.23 5.978 439 0.119* 0.844** 1 4. Feedback 97.502 7.5205 439 0.137** -0.014 0.000 1 5. Snipe 2.68 3.529 439 0.122** 0.386** 0.356** 0.005 1 6. Dummy Reserve .36 .482 117 0.217* -0.541** -0.694** -0.270** -0.178 1
The correlation matrix for France shows significant correlation between the number of bidders and the
dummy reserve (-0.694) and between the number of bids and the dummy reserve (-0.541). The test for
multicollinearity will show whether the correlations between the variables will affect the model. Again the number of bids and the number of bidders show a high correlation (0.844) and are not used for the same analyses.
Hypothesis 1a states that the relationship between the number of bidders and the final price is positive
in less uncertainty avoidant cultures, while insignificant in uncertainty avoidant cultures. France should therefore show insignificant results, whereas the US and UK should show positive results. The data were analysed by using multiple linear regression, with the number of bidders, feedback rating and the dummy of reserve price as regressors. The results for the UK are given in table 6. The regression for the number of bidders in the UK, controlling for feedback, is a very poor fit (adjusted R2
= 1.3%), but the overall relationship is significant (F2,775 = 6.054, p < 0.01). The final price is
positively related with the number of bidders, increasing by approximately £ 11.90 for every bidder entering, with a significant effect (t775 = 3.271, p = 0.001).
When including the dummy variable of reserve price as the control variable, the results change. The adjusted R2 is still a poor fit (0.3%) and the overall relationship is insignificant (F
3,406 = 1.638, p >
0.05). 1 decreases to 0.075 and becomes insignificant (t406 = 1.174, p = 0.241). The dummy variable
does take a part of the influence of the independent variable, now that the influence of the number of
bidders has become insignificant. Note that in the table the sample shrinks to 407 iPods when
including the dummy of the reserve price as a control variable, this owing to the delayed start of collecting data on the reserve price. Of the total of 495 iPods that has data on the reserve price, 407 where newly sold iPods and 88 used ones. The 407 is included in the 776 iPods.
The results showed no violations of the assumption of a regression model. The model is according to the normal probability plot linear (see figure 1 in appendix). Furthermore, when looking at the scatterplot, with the standardized residual values against the standardized predicted values, the model is homoscedastic (see figure 2 in appendix). Also, no multicollinearity is found between the number of
bidders and the reserve price, with a Variance Inflation Factor (VIF) of 1.661. This result does weaken
the results from the correlation matrix in table 6. The relationship between the dummy reserve and the
number of bidders is not strong enough to cause noise in the model.
All the results for the regression with used iPods are a very poor fit, with an adjusted R2 between 0.3%
TABLE 6
Regression Results Hypothesis 1a: UK
(1) (2) (3) (4) New Bidders 0.119** (3.330) 0.075 (1.174) Feedback 0.036 (1.003) 0.042 (1.184) 0.071 (1.424) 0.074 (1.499) Dummy Reserve -0.058 (-1.169) -0.010 (-0.163) Constant 124.917** (5.970) 117.584** (5.625) 113.954** (4.591) 109.165** (4.342) F-value 1.006 6.054** 1.766 1.638 Degrees of Freedom 775 776 406 406 R2 0.001 0.015 0.009 0.012 Adjusted R2 0.000 0.013 0.004 0.005 Cases 776 776 407 407 Used Bidders 0.065 (0.970) 0.094 (0.612) Feedback 0.114 (1.711) 0.109 (1.632) 0.173 (1.617) 0.167 (1.547) Dummy Reserve 0.045 (0.421) 0.112 (0.731) Constant 93.421** (5.195) 93.210** (5.182) 88.737** (3.974) 86.699** (3.657) F-value 2.926 1.933 1.446 1.082 Degrees of Freedom 223 223 87 87 R2 0.013 0.017 0.033 0.037 Adjusted R2 0.009 0.008 0.010 0.003 Cases 224 224 88 88
Power Seller Bidders 0.231**
(4.310) -0.056 (-0.653) Feedback 0.016* (2.117) 0.117* (2.118) 0.267** (4.176) 0.264** (4.121) Dummy Reserve -0.197** (-3.085) -0.235** (-2.724) Constant -16.410 (-0.214) -25.357 (-0.339) -603.558** (-3.352) -593.942** (-3.283) F-value 4.484* 11.651** 14.043** 9.479** Degrees of Freedom 328 328 219 219 R2 0.014 0.067 0.115 0.116 Adjusted R2 0.011 0.061 0.106 0.104 Cases 329 329 220 220
Dependent Variable: Final price
