Shill Bidding Empirical Evidence of Its

13
Shill bidding: Empirical evidence of its effectiveness and likelihood of detection in online auction systems Alexey Nikitkov , Darlene Bay 1 Brock University, 500 Glenridge Ave., St. Catharines, ON L2S 3A1, Canada article info abstract Article history: Received 15 March 2013 Received in revised form 4 February 2015 Accepted 5 February 2015 Available online 6 March 2015 Auction participants, academic researchers and the popular press con- tinue to express concerns about shill bidding in online auctions. Howev- er, the market makers (auction websites) do not behave as if they view shill bidding as a signicant risk. Further, important questions about shill bidding remain unanswered: how easily sellers are able to shill bid, how readily such actions can be detected by the market maker or bidding community, whether shill bidding results in signicant eco- nomic gains, and which shill strategies are most effective. We report the results of nine weeks of online auction trading. We nd that a price premium of between 16% and 44% can be achieved by shill bidding. Importantly, shill bidding is quite easy to implement and neither the market maker nor bidders showed any indication that they noticed. We conclude that market makers should make a careful re-evaluation of the risks of shill bidding, since only they are in a position to take meaningful action to prevent it from occurring. Crown Copyright © 2015 Published by Elsevier Inc. All rights reserved. Keywords: Online information systems IT audit Fraud Shill bidding 1. Introduction Anecdotal evidence from current periodicals and blogs suggests that shill bidding is a signicant issue for users of online auction sites. For example, a high-prole eBay drop-off store was accused of shill bidding in a series of articles on the PurseBlog.com discussion boards (http://www.ecommercebytes.com/C/blog/blog. pl?-/pl/2012/5/1337308248.html, May 2012). eDrop-off sued both the person posting the accusation and PurseBlog.com in California and Illinois. The defendants responded with counter lings in both states. International Journal of Accounting Information Systems 16 (2015) 4254 Alexey Nikitkov thanks Brock University for the grant that helped facilitate this study. Corresponding author. Tel.: +1 905 688 5550x3272; fax: +1 905 688 9779. E-mail addresses: [email protected] (A. Nikitkov), [email protected] (D. Bay). 1 Tel.: +1 905 688 5550x4524. http://dx.doi.org/10.1016/j.accinf.2015.02.001 1467-0895/Crown Copyright © 2015 Published by Elsevier Inc. All rights reserved. Contents lists available at ScienceDirect International Journal of Accounting Information Systems

description

jkkigu

Transcript of Shill Bidding Empirical Evidence of Its

  • Alexey Nikitkov, Darlene Bay 1

    Brock University, 500 Glenridge Ave., St. Catharines, ON L

    the results of nine weeks of online auction trading. We nd that a

    s a signicant issue fored of shill bidding in a

    International Journal of Accounting Information Systems 16 (2015) 4254

    Contents lists available at ScienceDirect

    International Journal of AccountingInformation Systemsseries of articles on the PurseBlog.com discussion boards (http://www.ecommercebytes.com/C/blog/blog.pl?-/pl/2012/5/1337308248.html, May 2012). eDrop-off sued both the person posting the accusation andPurseBlog.com in California and Illinois. The defendants responded with counter lings in both states.1. Introduction

    Anecdotal evidence from current periodicals and blogs suggests that shill bidding iusers of online auction sites. For example, a high-prole eBay drop-off store was accusmarket maker nor bidders showed any indication that they noticed.We conclude that market makers should make a careful re-evaluationof the risks of shill bidding, since only they are in a position to takemeaningful action to prevent it from occurring.Crown Copyright 2015 Published by Elsevier Inc. All rights reserved.FraudShill bidding Alexey Nikitkov thanks Brock University for the gr Corresponding author. Tel.: +1 905 688 5550x327

    E-mail addresses: [email protected] (A. Nikitkov1 Tel.: +1 905 688 5550x4524.

    http://dx.doi.org/10.1016/j.accinf.2015.02.0011467-0895/Crown Copyright 2015 Published by Elseprice premiumof between 16% and 44% can be achievedby shill bidding.Importantly, shill bidding is quite easy to implement and neither theAccepted 5 February 2015Available online 6 March 2015

    Keywords:Online information systemsIT audit2S 3A1, Canada

    a b s t r a c t

    Auction participants, academic researchers and the popular press con-tinue to express concerns about shill bidding in online auctions. Howev-er, the market makers (auction websites) do not behave as if they viewshill bidding as a signicant risk. Further, important questions aboutshill bidding remain unanswered: how easily sellers are able to shillbid, how readily such actions can be detected by the market maker orbidding community, whether shill bidding results in signicant eco-nomic gains, and which shill strategies are most effective. We reporta r t i c l e i n f o

    Article history:Received 15 March 2013Received in revised form 4 February 2015Shill bidding: Empirical evidence of itseffectiveness and likelihood of detectionin online auction systemsant that helped facilitate this study.2; fax: +1 905 688 9779.), [email protected] (D. Bay).

    vier Inc. All rights reserved.

  • A vigorous discussion about the reality of shill bidding ensued as hundreds of responses were posted onecommercebytes blog pages. Some of the comments, left by traders who clearly had a great deal of experiencewith online auctions, described their encounters with shill bidding and expressed frustration with the spreadof shill bidding and the lack of action taken by the web site's management (see Illustration 1). The incidentwas clearly not seen as an isolated example or as unimportant by those posting comments, indicating thatshill bidding is viewed as an important problem by auction website users.

    The academic literature has also identied shill bidding in online auctions as an important issue. However,empirical research has been limited by the inability of researchers to unequivocally identify transactions thatinclude shill bidding. While some methods to identify shill bidders have been proposed, each requires gath-ering many different types of data and involves making assumptions that not all nd convincing (Kauffmanand Wood, 2003; Trevathan and Read, 2006; Nikitkov and Bay, 2010; Dong et al., 2009; Engelberg andWilliams, 2009; Ford et al., 2010; Xu, et al., 2010). The crux of the problem is that publically available

    Illustration 1Response to ecommercebytes.com blog posting (http://www.ecommercebytes.com/C/blog/blog.pl?/pl/2012/5/1337308248.html).

    I will preface this by saying ***THIS IS ALL MY OPINION*** BASED ON YEARS OFWORKINGWITH T&SAND EXPERIENCE LOOKING AT SHILL BIDDING.

    let's just take an objective look at some recent listings (SIC):item 2210159****** (completed)

    bidder a***a - N almost 400 bids on 300 items with 92% only on this sellerbidder 8***g - N268 bids on 133 items, 87% WITH 51 bid retractions over the last 6 monthsbidder a***n - N310 bids on 137 items, 97%and the best: l***l - N350 bids on 105 items, 100% This bidder's name is also known, has 7

    feedback ALL from the above-mentioned seller.

    There are a few other bidders that are iffy, but I'm giving those the benet of the doubt (even thoughthey were 80-ish%).

    Next item up for consideration:item 27097555****bidder k***c - N121 bids on 64 items 100%another appearance by a***a - N393 bids, 301 items 92%

    NOW, let's take a look at some random bidders: one of the unluckiest bidders I've seen:r***n (4), 30-Day SummaryTotal bids: 777Items bid on: 144Bid activity (%) with this seller: 100%

    43A. Nikitkov, D. Bay / International Journal of Accounting Information Systems 16 (2015) 4254i***r, 30-Day SummaryTotal bids: 83Items bid on: 75Bid activity (%) with this seller: 98%d***a, 30-Day SummaryTotal bids: 196Items bid on: 20Bid activity (%) with this seller: 100%

    Andunfortunately, this list goes on ad innitum. You can click onANYauction andnd this kind of biddingpattern.

  • information is not adequate, making it necessary to rely on circumstantial data and probabilistic assessmentsof whether shill bidding has occurred. Transactions that seem to represent clear instances of shill bidding can-not be conrmed as such since only the auction house has access to the data that could be used to do so.

    Market makers (online auctioneers) have apparently not responded proactively to shill bidding and the threatit poses to theuser community and the integrity of themarket.While theycontrol thedata thatwouldbenecessarytomore clearly identify problematic transactions, they almost universally prevent the public and even auction par-ticipants from accessing that data. For example, while eBay prohibits shill bidding and posts sanctions that will beimposed against violators, it denies access to information thatwouldenable bidders to forma reasonable judgmentabout the presence or absence of shill bidding. Since September of 2007, when eBay redesigned the Bid Historypage, actual bidding identities are known only to the seller. Further, penny auctions (such as the recently popularpennygrab.com)maygenerate substantially increasedprots if shill bidding is employed,whichnot only raises thenal selling price, but extends the bidding process and requires bidders to purchasemore bids,whichmust be paidfor in advance. Thus, the practice of requiring bids to be purchased in advance of placing themon a specic auctionseems to directly benet sellers who shill bid at the expense of buyers.

    From an enterprise riskmanagement (ERM) perspective, this failure to address what is seen by both usersand academics as a signicant risk is interesting. First, marketmakersmay simply be unaware of the degree ofshill bidding and the risk that it poses. In view of the size and level of success of several online auction houses,this is unlikely. Second, it may be that shill bidding occurs less often than has been thought. This, too, seemsunlikely in view of the outcry seen on the blogs. Alternatively, market makers could believe that no real eco-nomic benet is obtained by shill bidders and thus no real harm is done to the buyers. Thiswould indicate thatno long term negative impactwill accrue to the auction sites as a result of shill bidding. It is also possible that acostbenet analysis convinced the market makers that the costs of controlling shill bidding outweigh anybenets associated with such actions. Without access to the decision making process of the market makers,it is impossible to knowwhy shill bidding seems to be ignored. However, it seemsmost likely that it is viewedas a relatively low level risk that would cost more to control than is warranted.

    Academic research to date has not explained the reasons for the observed lack of response ofmarketmakersto shill bidding. Since all major websites prevent access to the data, research about shill bidding is sparse andresults are often conicting. For example, some analytical studiesnd that the potential for shill bidding reducesprices, thereby decreasing prots to sellers (for example, Chakraborty and Kosmopoulou, 2004) while othersnd that sellers can increase their prots by shill bidding (for example, Graham et al., 1990; Lamy, 2009). Fur-ther, some researchers believe that bidders who bid quickly after being outbidmight be shill bidders (for exam-ple, Ford et al., 2010) while others indicate that an equally plausible explanation may be the so-called hot-headed bidder (Hlasny, 2006). Thus, a great deal remains to be studied with respect to online shill bidding:how easily it can be implemented, how readily it can be detected by the market maker, whether it results inan economic gain, how signicant the economic gain is, what strategies shill bidding sellers employ, and the im-pact of shill bidding on other members of the online auction community. Answers to all of these questions areimportant in evaluating the apparent risk management postures of the market makers.

    This study attempts to address these essential questions about shill bidding. A natural experiment wasconducted by offering items for sale in online auctions and observing the responses of bidders and themarketmaker to attempts to bid up the prices.Weworked closely with the Research Ethics Board of our university inconducting this study and never allowed any independent bidder to purchase an item onwhich a shill bid hadbeen placed. The results show that shill bidding does allow the seller to increase the nal price. In addition,there was no response to the shill bidding from the market maker. Finally, shill bidding in a sequence of auc-tions affects price more than a random series of shill bids.

    The rest of this paper is organized as follows. The next section reviews the literature about the denition,identication, and impact of shill bidding. Next, we analyze the responses of the market makers to the risksposed by shill bidding from an ERM perspective. Following that is a description of the research questionsfor the current study and the methodology. We then present the results and provide a discussion of the im-plications of the ndings.

    2. Literature review

    Shill bidding is dened as bids placed by the seller or a confederate on her own items for the purpose of

    44 A. Nikitkov, D. Bay / International Journal of Accounting Information Systems 16 (2015) 4254increasing the nal selling price. While shill bidding has been a potential problem long before online auctions

  • were developed, it has becomemore difcult to detect and prevent in the virtual version of the auction. This is

    45A. Nikitkov, D. Bay / International Journal of Accounting Information Systems 16 (2015) 4254a function of the ease with which information systems permit the creation of several identities by the sameuser, the ability of a confederate to act without being physically present or even geographically close, andthe inability of the system to detect prohibited activity among hundreds of thousands of transactions.

    2.1. Types of shill bidding

    Several different types of shill bidding have been reviewed in the literature. These types can be organizedinto three main categories: reserve price shilling, signaling effect shilling, and price-squeezing shilling. Re-serve price shilling is used to eliminate the need for open disclosure by the seller of her true reservationvalue while still preventing the item from being sold at a lower price. Some researchers believe that openlystating a high reserve pricemay reduce the number of interested bidders and therefore decrease the eventualselling price (Gerding et al., 2007). A shill bid placed near the beginning of the auction has the same effect as areserve pricewithout being public and, therefore, discouraging to potential bidders. In addition, a lower statedreserve price may result in lower fees to the auction house, depending on the specic fee schedule beingemployed (Kauffman and Wood, 2003).

    Signaling effect shilling may also be used to convince potential or current bidders to increase their valua-tion of the item. If bidders' valuation functions are not independent, then a bidmay be viewed as a signal thatanother bidder, perhaps with superior information, is willing to pay a higher price (Lamy, 2009). In addition,the number of bidders in any auction provides some indication of the valuation others place on an item. Thus,shill bids can be used tomake an auction appearmore important due to the (articially) increased number ofbidders (Chakraborty and Kosmopoulou, 2004). Other (honest) bidders may respond by adjusting their ownvaluation upward and either joining the auction or entering higher bids.

    Price squeezing, or competitive bidding, is used to extract the highest possible bid from a bidder who hadalready indicated a strong interest in the item (Engelberg andWilliams, 2009). The shill bidder enters a com-petitive bid but at a very small increment above the current bid, which is intended to encourage the honestbidder to increase her bid. This type of shilling is particularly effective if a proxy bidder2 is being used (for ex-ample, eBay or Manheim car auction). The shill can enter bids at the least possible increment, causing theproxy bidder to increase the bid of the honest bidder whose current bid may be below her maximum bid.In this way, the true valuation of the bidder can eventually be discovered and extracted by the seller(Barbaro and Bracht, 2006; Engelberg and Williams, 2009).

    2.2. Impact of shill bidding

    While the major online auction sites may publically seem to believe that shill bidding is not a signicantproblem, the actual level of such activity and its impact on the price, the sellers, the buyers and the marketremains unclear. Because shill bidding can be almost impossible to positively identify, empirical studies ofthe frequency and impact of shill bidding are few. However, several analytical studies, experiments, and sim-ulations have been conducted that attempt to determine how sellers, honest bidders and the auctioneer (ormarket maker) are impacted by shill bidding.

    The results of the extant studieswith respect to the incentive of the seller to shill bid and the short and longterm impacts of a decision to shill bid have been inconsistent. On the one hand, some studies show that, inequilibrium, the potential that the seller might shill bid causes bidders to protect themselves by biddingless than the optimal bid (Jenamani et al., 2007) which in most models has been shown to be the private val-uation of the item for each bidder. Thus,nal prices and protsmay be decreased (Kosmopoulou andDe Silva,2007) and market efciency is eroded (Gerding et al., 2007). Other studies, however, show that sellers doprot by shill bidding (Graham et al., 1990; Kauffman and Wood, 2003; Engelberg and Williams, 2009) andthat, under some assumptions, no equilibrium exists without shill bidding (Wang et al., 2002).

    Results with respect to the impact on bidders are similarly inconsistent. Experimental evidence exists thatthe potential for shill bidding does not cause bidders to avoid the market (Kosmopoulou and De Silva, 2007)and that when shill bidding is allowed and openly conducted, the optimal bid is not reduced (Graham et al.,

    2 A proxy bidder allows the potential buyer to enter hermaximumbid. The information system thenmanages the bidding process such

    that her bid is automatically increased each time a competing bid is entered until her reservation price has been exceeded.

  • 1990). However, if sellers successfully implement a shill bidding strategy, bidders may pay higher prices thanthey would otherwise (Wang et al., 2002). Alternatively, it has been suggested that shill bidding may actuallybenet bidders by reducing the occurrence of thewinner's curse, or the chance that a biddermay end upwin-ning the auction but at a higher price than originally planned (Kosmopoulou andDe Silva, 2007). To the extentthat bidders decide to place a bid that is lower than the optimal bid due to the potential existence of shill bid-ding, the risk exists of losing an auction that closes at a price less than the bidder's actual reservation value(Bhargava et al., 2005).

    2.3. Identifying shill bidding

    A consensus seems to be developing that honest bidders cannot adequately identify shill bidders due tolack of sufcient, readily available information (for example, Hlasny, 2006; Ford et al., 2010; Dong et al.,2012). Analytical research suggests that, given the potential for shill bidding, bidders should reduce theirmaximum bid below the amount at which they actually value the item by an amount that depends on thenumber of bidders and the probability that the seller will shill bid (Chakraborty and Kosmopoulou, 2004). An-other proposed adjustment for bidders is to drop out of the auctionwhen the bid amount exceeds the expect-ed price (Dong et al., 2012). However, both suggestions depend on knowledge that a casual bidder may notpossess (probability of shill bidding or expected price). Further, while both will work in the aggregate, indi-vidual buyers or sellers may achieve less than optimal outcomes. For example, reducing the maximumprice may result in some bidders failing to win an auction even when they would have purchased the itemat the nal price had no fears of shill bidding been present.

    Like individual bidders, individual sellers are powerless to prevent the problem. As noted by several re-searchers, no mechanism exists by which a seller can credibly commit to refrain from shill bidding (Wanget al., 2002; Chakraborty and Kosmopoulou, 2004; Kosmopoulou and De Silva, 2007). In an experimental set-ting, there is evidence that shill bidders,when publicly announced, are punished by bidderswith a lower pricein subsequent auctions (Kosmopoulou andDe Silva, 2007). However, no clear public information aboutwhichsellers have engaged in shill bidding exists on the major online auction houses. Thus, a reputation for honestdealing is not possible to attain and sellers who do not wish to engage in shill bidding cannot differentiatethemselves from those who do.

    Researchers have attempted to empirically identify shill bidding in online auctions by rst carefully exam-ining the strategy and then attempting to isolate characteristics that might, taken together, allow suspiciousbidders to be detected (Kauffman and Wood, 2003; Trevathan and Read, 2006; Nikitkov and Bay, 2010;Dong et al., 2009; Engelberg andWilliams, 2009; Ford et al., 2010; Xu et al., 2010).While eachmodel is differ-ent and is based on a slightly different understanding of how shill bidding is being implemented, results arequite similar. All of the studies cited above nd many cases that are quite suspicious, leading the researchersto conclude that shill bidding in online auctions is not an infrequent event. Unfortunately, none of these stud-ies can denitively prove that shill bidding has been successfully identied, since access to data that wouldverify a connection between the suspicious bidder and the seller is not possible.

    Other limitations with respect to shill bidding identication have also been identied. Perhaps the mostimportant is that these studies, as noted above, depend on assumptions about the strategy of the shill bidder.The strategy of a shill bidder may change from auction to auction or even during an auction (Graham et al.,1990;Wang et al., 2002). Thus, amodel intended to identify one type of shill biddermay not successfully iden-tify others andmay fail to nd shill bidders who change strategy in response to the bids that are being placed.Another limitation of shill identication models is that they often require information from many auctionsand are thus not helpful for individual bidders to employ in determining if they are being victimized by ashill bidder. Finally, the characteristics that are used to determine if a bid is suspicious are nearly always sub-ject to alternative explanations (Hlasny, 2006).

    2.4. Responses to shill bidding by the market makers

    Thus, it seems clear that the potentially negative impact of shill bidding on buyers, honest sellers and themarket in general poses a risk to the continued effectiveness of online auctions. Further, the only stakeholders

    46 A. Nikitkov, D. Bay / International Journal of Accounting Information Systems 16 (2015) 4254with sufcient access to data to both identify problematic transactions and to impose sanctions are themarket

  • makers. The principles of Enterprise Risk Management (ERM) would suggest that the large internet auctionsites should identify this risk and take action.

    ERM is a process designed to identify potential events (risks) that may affect the entity and tomanage theassociated positive or negative impacts in order to provide reasonable assurance that therm's objectiveswillbe achieved (COSO, 2004). ERM is intended to be implemented on a rm-wide basis and to be integratedwithstrategy discussions and initiatives. The traditional risk management approach, in contrast, leaves risk identi-cation and control to the separate units within the company, a.k.a. the silo approach. The ERM approachtreats all risks as part of the company's overall risk portfolio that is managed holistically (Liebenberg andHoyt, 2003), which is expected to be more successful at limiting risk than the silo approach.

    Risks under ERM are dened as the events that have a negative impact on the ability of the company toachieve its objectives,which are divided into four categories: strategic, operational, reporting, and compliance(COSO, 2004). The potential proliferation of shill bidding meets this denition and could impact the marketmaker at several of the levels of its objectives. If shill bidding is not sufciently controlled for, cases of shill bid-ding inevitably will increase and may also increase in severity as some sellers explore the boundaries of thepotential benets. Over time, if shill bidding continues to increase in frequency and severity, it will be increas-ingly noticed by the trading community and the press and may provoke a response from law enforcement,and regulatory bodies. At the strategic level, an online auction site's entire business strategy depends on thewillingness of buyers and sellers to interact via its platform,whichmay be threatened if shill bidding becomeswidely recognized as a major issue. At the level of operations, egregious shill bidding incidents may requireimplementation of emergency measures which tend to be costly. Public exposure may bring into questionthe reliability of internal controls and nancial reporting, and the effectiveness of compliance with anti-fraud-related laws and regulations.

    While many companies have implemented the ERM framework to increase the effectiveness of their riskmanagement activities in the last decade, they do not oftenmake it explicitly known. Internet auction sites areno exception. However, recent research suggests that important determinants of ERM adoption include: ap-pointment of a Chief Risk Ofcer, high independence of the board of directors, explicit calls by the CEO or CFOfor internal audit involvement in ERM, a Big Four auditor, size of the company, country of domicile, and indus-try (Kleffner et al., 2003; Liebenberg andHoyt, 2003; Beasley et al., 2005). To investigate if the internet auctionmarketmakers seem to be implementing ERMprinciples, we gather information from thewebsites of the top6 international online auctions (http://online-auction-sites.toptenreviews.com/index2.html) on these char-acteristics (see Table 1).We nd only one company, Taobao, in our samplewhich explicitly discloses adoptionof ERM. This company also closely meets the previously established determinants: TaoBao is a subsidiary ofAlibaba, one of the largest companies in the world, it is audited by PWC, 50% of the Board of Directors are in-dependent, and it has established a Chief Risk Ofcer position. There is little evidence that the other ve lead-ing online auctions have adopted ERM.

    An alternate source for evidence that the auction houses are actively managing the risks associated withshill bidding would be efforts to communicate that risk both within the organization and externally (COSO,2004). Only one company from our sample, eBay, clearly states that it does not allow shill bidding, andwarns sellers to follow the established guidelines (http://pages.ebay.com/help/policies/seller-shill-bidding.html). It cautions that IDs of users participating in shill bidding may be temporarily or permanentlysuspended. The company encourages buyers participating in online auctions to report suspicious activities,promising to thoroughly investigate every report. However, neither those ling a report nor the trading com-munity in general receive any information on such investigations. The company cites its privacy policy whichprevents it fromdisclosing the details of the investigation to othermembers. The other companies in our sam-ple of the top 6 online auction houses post no disclosure on their websites to indicate that they see shill bid-ding as a risk and are attempting to address the issue.

    In summary, we nd little evidence that themajor auction web sites are taking signicant action to detectand prevent shill bidding. One must consider that the market makers are either unaware of its existence orthey are aware and have chosen a limited response. ERM suggests four potential responses to an identiedrisk: avoiding, accepting, reducing and sharing (COSO, 2004). Absent any evidence of efforts to avoid, reduceor share the negative impact of shill bidding, it appears that the market makers have decided to accept thisrisk.

    A risk is considered acceptable if it does not exceed the rm's risk appetite (COSO, 2004). Risksmay be an-

    47A. Nikitkov, D. Bay / International Journal of Accounting Information Systems 16 (2015) 4254alyzed on the basis of the level of impact and the likelihood of occurrence. Those that are low on both

  • characteristics may be ignored, while those that are high on both clearly fall outside the area of the rm's riskappetite and require a proactive response. As long as shill bidding continues to be difcult to identify, it will bea high probability risk. It therefore appears that market makers consider it to have a low impact. The vast andspontaneous response that cases like eDropOff vs. PurseBlog invoke call this assumption into question. Shouldproliferation of shill bidding occur, the snowballing effect in the trading community and increased awarenessof the shill bidding may change it from the low to high impact category.

    There are some reasons that market makers may accept the risk associated with shill bidding. Several re-searchers have suggested that themarketmaker hasmore to gain than other parties from shill bidding. If shillbidding results in higher prices overall and the market maker charges a commission, it will benet from thehigher prices (Chakraborty and Kosmopoulou, 2004). Aggressive attempts by themarketmaker to counteractshill bidding may be met by resistance from sellers, who may migrate to another marketplace, reducing theoverall effectiveness and related revenue stream to the market maker. In addition, it might be data andtime intensive to gather the necessary information, check it for accuracy and impose sanctions. Therefore, itappears that the market makers have little incentive to control shill bidding.

    Currently, market makers have apparentlymade a decision to accept shill bidding as a low impact risk andmay even benet from shill bidding in the form of increased payments from sellers. However, if buyers werebeing harmed enough, they would become aware of the magnitude of the issue and might respond actively,

    Table 1Meeting determinants of ERM adoption for largest online auctions.

    Name Ownership Chief riskofcer

    % independentboard of directors

    ERM adoptionpublication

    Big Fourauditor

    Monthly visitors(proxy for size)

    Countrydomicile

    eBay1 Inc. No 91% No PWC 120,000,000 USA/Int.eBid.net3 Private No N/A No N/A 36,000 USA/Int.Onlineauctions2 Private No N/A No N/A 39,000 USATaoBao6 Inc. Yes 50% Yes PWC 316,605,000 China/Int.uBid5 Private No 50% No N/A 3,200,000 USAWeBidz4 Private No N/A No N/A 124,000 USA

    This table contains data gathered from the websites of each online auction. Data was current as of February of 2014 and is presented inalphabetical order by rm.

    1 www.ebay.com2 www.onlineauction.com3 www.ebid.net4 www.webidz.com5 www.ubid.com6 www.taobao.com

    48 A. Nikitkov, D. Bay / International Journal of Accounting Information Systems 16 (2015) 4254forcing market makers to re-evaluate their position. It may be that buyers notice shill bidding and simplychoose to accept it as a cost of doing business in a forumwhere prices are generally lower than in other mar-kets. Alternatively, it may be that buyers understand that they have insufcient evidence to prove shill bid-ding. Some may have chosen to simply not use the website, while others do the best they can to protectthemselves.

    3. Research questions

    While there is clear evidence that researchers and some members of the user community are concernedabout shill bidding, the behavior of the market makers would seem to indicate that the problem is eithernot pervasive or of little economic impact. Importantly, the actual ability of sellers to use shill bidding to in-crease their prots remains an open question. Even if sellers can increase the price atwhich the auction closes,the amount of the increase is unknown. Several of the studies that propose models to identify shill biddingprovide evidence that it does occur. Therefore, it seemsunlikely that sellerswould take the risk of shill biddingif it were not protable. However, we have no direct evidence of the effectiveness of shill bidding. Thus wepropose the following two research questions:

    RQ1: Can shill bidding be used to increase the payoff in an online auction?RQ2: How large is the premium generated by shill bidding?

  • In addition to evidence about the actual impact of shill bidding on the nal price, the economic impact ofthis strategy must be understood in the context of the associated risks to both the individual seller and to themarketmaker. One risk to the shill bidder (known as the shiller's curse) is that the shill bidwill be the nal bidand the seller will be forced to re-list the item. Further, if buyerswere able to identify sellers that are using this

    49A. Nikitkov, D. Bay / International Journal of Accounting Information Systems 16 (2015) 4254technique, they might make such information public, causing the risk of losing potential future buyers and ofaction by themarketmaker.Whilewehave found little evidence that themarketmakers are acting to discour-age shill bidding, the degree towhich they respond to problematic transactions is still an open question. Thus,it is important to understand the likelihood of detection for sellers who choose to shill bid:

    RQ3: Are buyers likely to be able to detect a shill bidder?RQ4: Does the market maker take action against a shill bidder?

    4. Method

    In order to address our research questions, we implemented an experiment on a large online auction. Weconducted 101 identical auctions for a computer memory card. We chose this item since it is easy to acquireand is in relatively high demand, thus ensuring our ability to secure product and that our auctionswould gen-erate some response. Also, our intentionwas to generate results for business as usual or typical transactionsso we selected a fairly popular, but inconspicuous product. We did not wish to inate our results by usingmust have products, like Apple iPhone, beats headphones, or designer clothes. Finally, we chose a producewe could acquire and sell with a limited research budget.

    For nineweeks, we listed from six tofteen auctions perweek. In about half of the auctions, we placed sev-eral shill bids in an effort to increase the nal price being offered.Weworked closely with our Research EthicsBoard to ensure that due attention was paid to the interests of the market maker, other sellers, and buyersduring the experiment.We paid all the fees due to themarketmaker and responded quickly and as requestedwhen we received communications from the market maker during the experiment. We listed only a limitednumber of cards for sale (less than 15) in a givenweek, so as not to substantially interfere with the offers andtrade of the other sellers. During the time of the experiment therewere 26 other sellers offering from 1 to 610similar cards each. Thus, all buyers had ample opportunity to purchase the same card from other sellers, andour auctions represented only a small proportion of the cards being offered. Based on this evidence, we feelcondent that our auctions had no impact onmarket price. Perhapsmost importantly, we ended each auctioninwhich a shill bid had been placed by placing a very large bid at the end of the auction, ensuring that no hon-est bidder would pay the articially raised price. Cards in the auctions in which no shill bid had been placedwere sold to the highest bidder. We took the necessary precautions to ensure that this real-life experimenthad a negligible impact on the market maker and the community.

    Prior to the experiment, we attempted to create fteen new user IDs, such that registration would involveonly ctitious information and newly created generic (e.g. gmail) accounts. Five of these accounts wereagged by the market maker's system and personal identifying information was required. We opted not toprovide this information and retained only ten accounts. For these ten user identitieswe completed a numberof transactions unrelated to the project and received ve-star positive feedback. This step was required to in-sure that these identities appeared sufciently innocent when used in the subsequent experiment; identitieswith no feedback are easily spotted by other auction participants and usually viewed with suspicion.

    After the new identities were acquired, we used an old and trustworthy-seeming user identity as theseller to list the auctions. This ID has been used for trading with the market maker for several years andhad received 100+ positive feedbacks. In order to rule out the impact of product presentation on the auctionending price, we studied and closely matched our presentation to that of the competitors' on 15 characteris-tics.3 Such close matching serves to establish a valid comparison between the results of our experiment andprices achieved by the other sellers of the same product.

    3 These were: content of the item description; stock picture; return policies; shipping method, price, and policies; seller reputation;product's brand, capacity, content, and color; duration of the auctions; location of the seller and product; payment method; timing of

    the auctions (same day as the competitors' for the same weeks).

  • We began our experiment assuming that the bidding process for each auction would be independent ofprevious and subsequent auctions. Thus, our initial shill bidding strategy was F-S-F-S-F-S, where F is an auc-tion free from shill bids and S is an auction manipulated by shill bids. As our experiment unfolded, we ob-served that the dynamic of successful shill bid placement was contrary to this initial expectation: placing

    RQ3 andRQ4 relate to the possibilities that shill bidding activitywould bedetected by thebuyers or by the

    50 A. Nikitkov, D. Bay / International Journal of Accounting Information Systems 16 (2015) 4254website. We conducted 101 auctions, nearly of which involved shill bidding. Had some evidence surfacedthat either stakeholder questioned the transactions, we could report their responses and discuss how theirsuspicions were aroused. However, during the entire experiment, we did not have any indication that our ac-tivity was monitored or noticed as suspicious by themarket maker or the trading community. There was oneexception. However, it was not a result of our shill bidding activity. Themarketmakermaintains a selling vol-ume restriction: on attempting to post auction #70wewere informed that we cannot post any more listings.This restriction was lifted in a week when 15 of our previous auctions were completed. We encountered theselling volume restriction twice before we posted all 101 auctions. Thus, while there is evidence that themar-ket maker monitors transactions for some purposes, there is no evidence that either themarket maker or thebuyers became suspicious that shill bidding was an issue.

    4 Due to concerns regarding the distribution of such a small sample size, we repeat the analysis using nonparametric tests. TheMannWhitney U test of means across effects (shill bidding or no shill bidding) results in a value of 12.50, p = .662. A KruskalWallis test fortrend across the 10 transactions results in a Chi-square of 10.0, p = .440.5 We also rerun all analyses with the outliers winsorized. The inferences based on these alternative analyses are qualitatively the sameshill bids in one separate auction appeared to inuence the ending price least, especially if the next auctionwas free from any shill bidding activity. Before proceeding, we tested this observation by analyzing the datafrom week one using a regression model with price as the dependent variable and type of auction (shill bidor free) as the independent variable. The results are contained in Table 2. The coefcient of transaction typein the tted model has a p-value of 0.946 indicating that it is not signicant. Hence, we concluded thatthere is no effect of shill bidding in the week with intermittent application of shill bidding.4

    Having determined that an alternating shill and free strategy was not effective, in subsequent weeks, shillbids were placed in a series of auctions ranging from 2 to 9 in a row. We varied the order and duration of se-quences of shill bids. In some weeks, the free auctions were the rst few auctions, followed by the auctionswith shill bidding. In others, the order was reversed, with shill bidding auctions coming rst.

    5. Results

    To analyze the results of our attempts to increase the pricewith shill bidding, we begin by looking at pricesfor the auctions. Table 3 contains information about mean prices in auctions with and without shill biddingover the course of our experiment. Information is provided separately for therstweek, inwhich the shill bid-ding strategy was different from the remaining eight weeks. Before proceeding with the analysis of our re-sults, we analyzed the data for outliers using Z-scores. Two observations were found with Z-scores thatexceeded 3. These observations have been eliminated in what follows.5 As can be seen in Table 3, in weeks2 to 9, the average pricewith shill bidding (mean of $20.30) is greater than the average pricewithout shill bid-ding (mean of $17.26). This difference is statistically signicant (independent samples t-test, t = 6.976; p=0.000). Thus, RQ 1 can be answered in the afrmative: it is possible to increase the average price using shillbidding.

    Wenext turn to RQ2 regarding the economic signicance of the price premiumachieved through shill bid-ding. According to the results of our experiment, the mean price difference between shill-bid-affected auc-tions and auctions free from shill bidding is $3.04 or 17.6% (if one considers including the results fromweek one, the mean difference is $2.70 or 15.7%). As we discuss below, repeated application of shill biddingmay lead to substantially higher price premiums. For example, sequences of shill-bid-controlled auctions re-sulted in ending prices in excess of $25 for weeks 2, 6, 7, and 9; that is $7.64 or 44% higher than the price inauctionswithout shill bidding. The ability to increase the selling price by asmuch as 44% over the price of hon-est competitors through shill bidding may represent an important economic benet to those who engage inthis practice. We note that the premium we were able to achieve in our auctions might not be available ifall sellers were shill bidding.as those reported in the body of the paper.

  • 5.1. Sh

    Table 2Regression results for intermittent strategy.

    ANOVA

    Model Sum of squares Df Mean square F Sig.

    1 Regression .002 1 .002 .005 .946(a)Residual 3.757 9 .417Total 3.759 10

    Coefcients

    Model Unstandardized coefcients Standardized coefcients T P value

    B Std. error Beta

    1 (Constant) 18.518 .418 44.322 .000

    bid plac

    51A. Nikitkov, D. Bay / International Journal of Accounting Information Systems 16 (2015) 4254Graphs of prices from two typicalweeks are shown in Fig. 1. The patterns displayed in these graphs suggestthat repeated shill bidding results in an increasing price from auction to auction and that repeated lack of shillbidding results in a decreasing price from auction to auction.

    This interpretation of the visual evidence suggests that an interaction exists between repeated transac-tions and the application or absence of shill bidding. To test this idea, data from weeks two through ninewere used. A regression model was estimated with price as the dependent variable and three independentvariables: transaction number (for the trading week), a dummy variable (effect) taking the value of 1 if theprice was obtained by repeated application of shill bidding and 0 if the price was the result of repeated ab-sence of shill bidding and the interaction of transaction number and effect.

    Results of the regression analysis are given in Table 4. Unlike the analysis for week one, themodel ts well(F = 18.65; p = 0.000; R2 adjusted = .378). Importantly, the interaction term is signicant with p value of0.024 (signicant at 5%). This indicates that sequential shill bidding produces a signicantly different impacton price than sequential absence of shill bidding. Following procedures from Aiken andWest (1991), we usethe data in Table 4 to construct two different regression equations by setting the dummy variable for transac-tion equal to 1 and then 0:

    For shill bidding : price 19:332 0:23 transactionThtions a

    Table 3Sample

    WeekFS

    WeekFS

    All traFSill bidding strategies

    ed) as the independent variable. R2 adjusted for the model was.111. Data is from the rst week of the experiment only.Shill bidding .004 .062 .023 .069 .946

    This table contains the results of an ANOVA analysis with price as the dependent variable and shill bidding (0=no shill bidding and 1=shillFor absence of shill bidding : price 18:239 0:116 transaction

    e signicant interaction in the regression analysis indicates that the slopes in these two regression equa-re different from each other. This means that in repeated transactions with and without shill bidding,

    descriptive statistics.

    N Mean Std. dev

    1ree of shill bidding 5 $18.56 0.371hill bidding applied 6 $18.53 0.715

    s 29ree of shill bidding 46 $17.26 1.308hill bidding applied 42 $20.30 2.627

    nsactionsree of shill bidding 51 $17.38 1.306hill bidding applied 48 $20.08 2.537

  • 52 A. Nikitkov, D. Bay / International Journal of Accounting Information Systems 16 (2015) 4254Fig. 1. Prices in representative weeks.

    Table 4Regression results for sequential strategy.

    ANOVA

    Model Sum of squares df Mean square F Sig.

    Regression 225.344 3 75.115 18.647 .000Residual 338.375 84 4.028Total 563.719 87

    Coefcients

    Model Unstandardized coefcients Standardized coefcients T Sig.

    B Std. error Beta

    (Constant) 18.239 .742 24.582 .000Effect 1.093 .966 .216 1.131 .261Transaction number .116 .080 .174 1.445 .152Interaction .346 .150 .369 2.302 .024

    This table contains the result of a regressionwith price as the dependent variable and effect (a dummy variable taking the value of 1 if shillbiddingwas present and 0 otherwise), transaction number and the interaction of transaction number and effect as independent variables.R2 adjusted was .378. Transaction number for each trading week begins with one and ends at the last transaction for the week. Dataincludes only transactions from weeks 2 through 9.

  • the effect on price is different. In order to test if each slope is different from zero, we conduct t-tests (see Aiken

    53A. Nikitkov, D. Bay / International Journal of Accounting Information Systems 16 (2015) 4254and West (1991)) by dividing each simple slope by the corresponding standard error. For the equation withshill bidding, the t value is 1.399 (p = .170), indicating that the slope is not signicantly different from zero.For the equationwithout shill bidding, the t value is2.324 (p= .025), indicating that the slope is signicant.

    The negative coefcient in the model depicting a series of transactions free from shill bidding suggests aninteresting implication: if the seller does nothing to impact the price in his auctions (no shill bidding), the sell-er should expect the ending auction price to drop with every subsequent listing. Realization of this fact pro-vides a strong motivation for sellers to shill bid. Thus experienced sellers, who may have understood thisaspect of trade dynamics, may resort to shill biddingmore often than novice sellers. Secondly, this result sug-gests that listingmany auctions in the short period of time (with anhour, or a day)may be counter-productivefor the seller: the average ending price across a series of auctions would decrease.

    The failure to nd a signicant, positive slope for the regression equation with shill bidding in spite of theevidence from the graphs and the clear price premium as presented above may be a result of the larger stan-dard error of the estimate compared to the standard error of the estimate without shill bidding. This indicatesthat there is more variability in price under shill bidding than without. However, it is not really necessary forboth slopes to be signicantly different from zero in the expected direction, since the important question inthis study is whether they differ from each other. The signicant interaction term in the regression analysissupports a statement that the simple slopes of the two regression equations are statistically different. Sellersreally have only two choices: refrain from shill bidding or engage in shill bidding. If they refrain from shill bid-ding, the selling price may fall. If they engage in shill bidding, the selling price will likely increase over time.Thus, our evidence indicates that sellers who shill bid will achieve higher prices than those who do not.

    6. Summary and conclusions

    Our study provides evidence that, contrary to ndings in some analytical studies, shill bidding can be usedto increase the nal price in an online auction. Further, our results suggest that the economic premium thatcan be achieved by shill bidding, which was as much as 44% in someweeks, is substantial. We did not receiveany indication that even one of the many bidders in our nearly 100 auctions developed any suspicions aboutour auctions.6 In addition, therewere no communications from themarketmaker that indicated it had noticedany problems with our auctions. Thus, at the current time and in the market for this one product at a mini-mum, it is still quite possible to protably shill bid.

    The nding that price continues to increase as a result of higher prices in prior auctions of the same seller isan important contribution of this study. To date, there has been no suggestion in the literature that the effectof shill bidding could be cumulative across auctions. Based on our data from the rst week of trading, it ap-pears that placing a shill bid on a random basis is not as likely to result in clear economic gain as is an ongoingprogram of shill bidding in all auctions.

    Wehave speculated that themarketmaker's response to the risk posed by shill bidding is to ignore it basedon the assumption that the impact on the rm's strategic and operational initiatives is small. Based on the re-sults of this study, it would seem prudent and in line with best risk management practices to re-evaluate thisstance. Given the ease with which shill bidding can apparently be conducted and the prots that can appar-ently be garnered, one might speculate that many sellers are currently engaged in this practice. Events suchas those described above with respect to eDrop-off may seem, at rst glance, to support a policy of no re-sponse. The market maker was not a party to the lawsuits that resulted. However, the public discussion ofshill bidding and especially the criticism from buyers about the inaction of the market maker may indicatethat buyers are notwilling to accept the presence of shill bidding andmay foreshadowa timewhen either reg-ulators or market makers will have to take action.

    As long as buyers, even those who demonstrate some degree of suspicion, remain unable to conclusivelyidentify problematic transactions, the risk that shill bidding represents tomarketmakersmay remain limited.However, if these conditions change, shill bidding becomes a signicant risk. At aminimum, it is necessary forthemarket maker to remain alert to changes in the risks posed by shill bidding by continuously paying atten-tion to communications from users as well as monitoring news stories about shill bidding on other websites

    6 It is clearly not possible to prove that none of the buyerswere suspicious. However, we received no communication from any of them

    indicating any problems. Further, no actions were led against us with the market maker.

  • and the related responses on blogs and other public fora. In addition, communications frombuyers to themar-

    54 A. Nikitkov, D. Bay / International Journal of Accounting Information Systems 16 (2015) 4254ket maker should be taken seriously. It would be particularly important to maintain historical informationabout the number and economic magnitude of such reports.

    As with all research, this study has some limitations that should bementioned. We did only trade for nineweeks and conducted very few transactions, relative to total transactions on thewebsite. In addition, the pricepremiumwe achievedwould not necessarily be the same across all products and for all sellerswho shill bid. Itis also possible that a larger program of shill bidding across manymore auctions might have attracted the at-tention of the market maker.

    Further, the current study had to use relatively low ticket items due to research budgetary constraints.Whether shill bidding can result in such high price premiums for high (or higher) ticket items remains anopen question. However, sellers of high ticket items have a strong incentive to engage in shill bidding sincethe absolute value of the price increase (in terms of dollars) may be signicantly higher. On the other hand,shill bidding on these itemsmight attract more attention if buyers are more sensitive to the risk and themar-ket maker is more vigilant and would respond to evidence of shill bidding more actively. How much publicexposure is required before action must be taken and whether high ticket items should be monitored morediligently than others are among the questions that remain to be answered.

    References

    Aiken L, West S. Multiple regression: testing and interpreting interactions. Newbury Park, Ca: Sage Publications; 1991.Barbaro S, Bracht B. Shilling, squeezing, sniping: explaining late bidding in online second-price auctions. Working Paper; 2006 [accessed

    at http://www.staff.uni-mainz.de/barbaro/BarbaroBracht.pdf].Beasley MS, Clune R, Hermanson DR. Enterprise risk management: an empirical analysis of factors associated with the extent of imple-

    mentation. J. Account. Public Policy 2005;24(6):52131.Bhargava B, Jenamani M, Zhong Y. Counteracting shill bidding in online English auction. Int J Coop Inf Syst 2005;14(2 & 3):24563.Chakraborty I, Kosmopoulou G. Auctions with shill bidding. Economic Theory 2004;24(2):27187.COSO. Enterprise risk management integrated framework. Jersey City, NJ: Committee of Sponsoring Organizations of the Treadway

    Commission; 2004.Dong F, Shatz SM, Xu H. Inference of online auction shills using DempsterShafer Theory. Proceedings of the 6th international conference

    on information technology: new generations; 2009. [accessed at http://www.cs.uic.edu/~shatz/papers/itng09b.pdf].Dong F, Shatz SM, Xu H, Majumdar D. Price comparison: a reliable approach to identifying shill bidding in online auctions? Electron

    Commer Res Appl 2012;11(2):1719.Engelberg J, Williams J. eBay's proxy bidding: a license to shill. J Econ Behav Organ 2009;72(1):50926.Ford BJ, Xu H, Valova I. Identifying suspicious bidders utilizing hierarchical clustering and decision trees. International conference of ar-

    ticial intelligence; 2010. p. 195201. [accessed at http://www.cis.umassd.edu/~hxu/Papers/UMD/ICAI-2010-XU.pdf].Gerding EH, Rogers A, Dash RK, Jennings NR. Competing sellers in online markets: reserve prices, shill bidding and auctions fees. Pro-

    ceedings of the 20th international joint conference on articial intelligence; 2007. [accessed at http://www.aaai.org/Papers/IJCAI/2007/IJCAI07-208.pdf].

    Graham DA, Marshall RC, Richard JF. Phantom bidding against heterogeneous bidders. Econ Lett 1990;32(1):137.Hlasny V. Testing for the occurrence of shill-bidding (in internet auctions). Grad J Soc Sci 2006;3(1):6181.Jenamani M, Zhong Y, Bhargava B. Cheating in online auctiontowards explaining the popularity of English auction. Electron Commer

    Res Appl 2007;6(1):5362.Kauffman R,Wood CA. Running up the bid: detecting, predicting, and preventing reserve price shilling in online auctions. Proceedings of

    the 5th international conference on electronic commerce (ICEC); 2003. p. 25965. [accessed at http://www.aaai.org/Papers/IJCAI/2007/IJCAI07-208.pdf].

    Kleffner AE, Lee RB, McGannon B. The effect of corporate governance on the use of enterprise risk management. Manage. Insur. Rev.2003;6(1):5573.

    Kosmopoulou G, De Silva DG. The effect of shill bidding upon prices: experimental evidence. Int J Ind Organ 2007;25(2):291313.Lamy L. The shill bidding effect versus the linkage principle. J Econ Theory 2009;144(1):390413.Liebenberg AP, Hoyt RE. The determinants of enterprise risk management: evidence from the appointment of chief risk ofcers. Risk

    Manage. Insur. Rev. 2003;6(1):3752.Nikitkov A, Bay D. Online auction fraud: an empirical analysis of shill bidding practice. J. Forensic Invest. Account. 2010;2(3).Trevathan J, Read W. Detecting shill bidding in online English auctions. ACM transactions on computational logic; 2006. [accessed at

    http://dunk2.jcu.edu.au/~jc194392/RASNew/research/shill.pdf].WangW, Hidvgi Z, Whinston AB. Shill bidding in multi-round online auctions. Proceedings of the 35th Hawaii international conference

    on system sciences; 2002. [accessed at http://scholar.google.ca/scholar?as_q=shill+bidding+multi+round&as_epq=&as_oq=&as_eq=&as_occt=any&as_sauthors=Wang&as_publication=&as_ylo=&as_yhi=&btnG=&hl=en&as_sdt=0%2C5].

    Xu H, Bates CK, Shatz SM. Real-timemodel checking for shill detection in live online auctions. Proceeding of the international conferenceof software engineering; 2010. [accessed at http://www.cs.uic.edu/~shatz/papers/serp09.pdf].

    Shill bidding: Empirical evidence of its effectiveness and likelihood of detection in online auction systems1. Introduction2. Literature review2.1. Types of shill bidding2.2. Impact of shill bidding2.3. Identifying shill bidding2.4. Responses to shill bidding by the market makers

    3. Research questions4. Method5. Results5.1. Shill bidding strategies

    6. Summary and conclusionsReferences