The Production of Trust in Online Markets [1]

At the heart of any unsecured transaction is a social dilemma. By social dilemma, I mean a situation in which behavior that is rational for the individual leads to an irrational collective outcome.[2]  In the case of bilateral exchange, there is the temptation to receive a good or service without reciprocation, but if both parties hold back on their side of the exchange, the trade is never consummated and both are worse off. Thus, an unsecured bilateral exchange typically has the structure of a Prisoner’s Dilemma.

The temptation to defect in the exchange has led to a wide range of formal and informal mechanisms for managing this risk. The simple act of meeting face-to-face for the transaction helps reduce the likelihood that one party will end up empty handed. Separating the two sides of the transaction by time or space (such as purchasing something by mail or on credit) introduces greater risks: the party who moves second must be considered trustworthy or have some other form of guarantee. The formal infrastructure that exists to manage these risks is vast and includes such elements as credit card companies, credit rating services, public accounting firms, and – if the exchange goes bad – such service as collection agencies or the court system.

Such third-party enforcement mechanisms are inevitably imperfect and impose their own set of costs. Pointing to them as an explanation for how the risks of trade are overcome also begs the question of emergence: when and how are groups able to manage the risks of trade when they do not have access to external enforcement mechanisms? It is the emergence of endogenous solutions to the problems of risky trade that I wish to explore here – to investigate informal markets which are still very much in the process of solving the political problems of exchange. While I am mindful of the distinctions that have been made between markets, hierarchies, and networks, in this paper I will use the term market in a very broad sense, to refer to any exchange setting with multiple buyers and sellers, regardless of whether it approximates a formal market.

Online Markets

In order to study endogenous solutions to the problem of risky trade, the ideal research site would be a collection of large, informal markets in which one was able to observe thousands of transactions, record participants’ reactions, and track the development of risk management strategies. Remarkably, such research sites exist in the informal markets that have emerged on the Internet. These markets are striking in terms of their size and the value of goods exchanged.

One set of informal online markets is the collection of Usenet newsgroups in which individuals buy and sell everything from computer equipment to bootleg music tapes. In each newsgroup, participants simply list an item for sale along with a description. Potential buyers then contact the seller via email if they are interested. These markets are very loosely structured. Nevertheless, a detailed Frequently Asked Questions (FAQ) file has been produced which has gathered the accumulated wisdom of participants concerning such matters as the best way to structure an advertisement or complete the transaction (King 1996). This is also a very large market – it was estimated in 1996 (King 1996) that over 300,000 individuals from around the world participated in the Usenet forsale newsgroups, and the number of participants has grown very significantly since then.

The emergence of the World Wide Web has also made it possible to easily establish informal marketplaces for different goods. In particular, a thriving set of barter markets exist which focus on collectible items, such as game cards. Again, these sites attract a large and global audience.

There have also emerged more structured online auction houses in which individuals can post items that are then bid on by other market participants. The largest of these auction sites boasts a market with over one million participants, and in terms of gross revenues, the most successful auction houses are starting to rival some financial exchanges.

In fact, it is useful to contrast these informal online markets to financial exchanges – the most formalized marketplaces that exist. As Jaycobs (1998) describes it, a financial exchange provides four key services: matching, liquidity, information, and clearing. The financial exchange promotes matching by offering a centralized location where buyers and sellers can meet. The presence of market makers at the exchange mean that buyers and sellers do not need to coordinate when they will arrive at the market – the market makers are there to take the other side of the transaction in order to assure a liquid market. The financial exchange also serves as a place to aggregate, store, and distribute information, which contributes to its function as a price discovery mechanism. Finally, financial exchanges provide a clearing service to guarantee the transactions and eliminate counter-party risk.

The economies of communication and coordination that exist in online interaction (Kollock 1998b) mean that even informal online markets can excel at the matching and information function. However, I have not yet seen evidence of market makers in these settings, although there are clearly opportunities for speculation within a market and arbitrage across markets, especially in the more structured sites such as the auction houses. The most important difference between these online markets and formal exchanges is the absence of a clearing mechanism. I use this distinction as the defining characteristic of an informal market: the absence of a formal clearing system or other third-party enforcement mechanisms.

Risk and Reputation

While the monetary value of trades in these online markets is small compared to transactions at a financial exchange (at least for now), the risks are still very significant for the individuals involved in the trades.[3]  First of all, there is the fact that one is dealing with a party who may be identified only by an email address and who in all likelihood lives far away, perhaps even in a different country. This is a Prisoner’s Dilemma transaction that requires a great amount of trust given that it may not be possible to track down or even identify the other party and that the two sides of the transaction are likely to be separated in both time and distance. And indeed, there is no shortage of trades that go bad. The National Consumers League [4] has gathered information on Internet fraud since February of 1996, and has found that online auctions were the third most common source of complaints in 1997 and the number one source of complaints in 1998. There are also some infamous examples of large-scale fraud within these informal markets. For example, in 1994, someone with the username "T. Le" posted several ads on one of the Usenet forsale newsgroups offering to sell a variety of computer components (Barrett 1996). The postings contained a great deal of information and offered the goods at attractive prices. Approximately 50 interested buyers sent this person money in response to the ads. In returned they received empty boxes, and the person in question disappeared with over $13,000.

Yet despite such incidents and the absence of formal enforcement schemes, both observation of these markets and reports from the participants indicate that only a small fraction of traders default on their transactions. This is remarkable given the inherent risks in online trades, and invites an investigation into these risky exchanges. These markets are, in other words, huge naturally occurring pools of prisoner's dilemma situations, and my goal is to explore how cooperation is created and maintained in such a seemingly unpromising setting. In particular, I wish to focus on the use of reputation reporting systems to encourage trustworthiness in transactions.

The study of reputation as a risk management technique is directly relevant to both theoretical work on cooperation and trust, as well as to the practical issues of designing and running online markets. A reputation, as defined by Wilson (1985, pp. 27-8), is a "characteristic or attribute ascribed to one person ... by another (e.g., ‘A has a reputation for courtesy’). Operationally, this is usually represented as a prediction about likely future behavior (e.g., ‘A is likely to be courteous’). It is, however, primarily an empirical statement (e.g., ‘A has been observed in the past to be courteous’). Its predictive power depends on the supposition that past behavior is indicative of future behavior."

Having interacted with someone before is, of course, a key source of information and a powerful determinant of that person’s reputation. But relying only on direct personal experience is both inefficient and perilous: inefficient, because any one individual will be limited in the number of exchange partners she or he has, and perilous, because one will discover untrustworthy partners only through hard experience. Great gains are possible if information about past interactions is shared and aggregated within a group. This sharing of past interaction histories can take many forms: informal gossip networks, institutionalized review systems, and even specialists whose sole job is to consume and evaluate a good or service (e.g., a restaurant critic). Theoretical work (e.g. Raub and Weesie 1990), has demonstrated the beneficial effects of shared reputations, and there is even some experimental work (Rapoport, Diekmann, and Franzen 1995) which demonstrates greater levels of cooperation when reputations are shared.

It is important to note that reputations serve as both a source of information and as a potential source of sanctions (Yamagishi and Yamagishi 1994). For the person deciding whether to enter into a transaction, the partner’s reputation is a source of information that can reduce uncertainty and guide the decision of whether to trust the partner. Because of this same dynamic, the existence of shared reputations serves as an incentive for the partner to be trustworthy because of the damaging effects of acquiring a bad reputation. However, the threatened or actual sanction of acquiring a bad reputation will only be effective to the extent that accurate information is collected and disseminated among likely exchange partners. If people do not talk among themselves, if the information exchanged is inaccurate, or if a person can hide their identity, then reputation systems will not be an effective means of managing risk.

These concerns highlight the importance of informal online markets as a research site. On the one hand, there are new challenges to managing risk because of the ease of changing identities in online environments. On the other hand, the economies of online interaction mean that the costs of collecting and distributing information can be extremely low, making possible reputation and evaluation systems that could otherwise not be supported (cf. Avery, Resnick, and Zeckhauser forthcoming, Friedman and Resnick 1998).

In what follows, I will first discuss general approaches to managing risk in online markets and then examine two classes of systems for using reputations as a risk management tool. These sections focus on the Usenet markets and web barter sites. I then detail the attempts to institutionalize reputation systems by online auction houses. I close by discussing the prerequisites for an online reputation system and design issues for such systems. It is important to stress that this is an investigation of a rapidly changing phenomena – while the general issues and principles discussed will remain relevant, the details about specific markets are inevitably out of date. I have provided web addresses below for the markets under investigation, so that readers can examine the sites as they currently exist.

MANAGING RISK

The numerous hazards faced by individuals in these markets have encouraged participants to sometimes seek out third party services in an attempt to manage these risks. While third party services are an interesting part of online markets, these services do not guard against all risks, impose costs of their own, and are in fact usually not used in online trades. Nevertheless, it is useful to briefly survey these risk management methods before turning to endogenous solutions to the problem of risky trade.

Third Party Services

Suggestions for the use of third party services for online transactions have been compiled in a number of documents (e.g., King 1996, Barrett 1996).

The first issue is that parties wish to guard against goods or funds that never arrive. At times individuals may also attempt to claim that goods or funds never arrived when in fact they did, as a way of backing out of their part of the exchange. Thus, it is recommended that individuals use certified mail with a return receipt or an express mail service that provides proof of delivery. The fact that a check has been received is small comfort if the check later bounces. The alternative here is to ask the buyer to send funds in the form of a money order. Interestingly, the buyer may resist doing so in order to preserve the option of stopping payment on the check if the goods are judged inadequate.[5]

There is also the problem that each party would prefer not to be the first to send their half of the exchange, and so unilaterally expose themselves to risk. Ideally, the goods and funds would be exchanged simultaneously by a neutral party. This can be accomplished through a COD (collect on delivery) service. However, there are limitations here as well. One cannot open the package in order to inspect the goods prior to payment, which means there is the threat that the seller has simply sent a "box of rocks." On the seller’s side, there is the risk that the buyer will refuse delivery, and the COD service can sometimes take a substantial length of time before delivering the funds. The is also the problem that COD is not available in some countries (many online markets are global markets).

These limitations and the fact that participants are increasingly dealing in high value items has encouraged the emergence of escrow services who specialize in online markets. Typically, the buyer transfers funds to the escrow service. Once the funds have cleared, the service instructs the seller to ship the goods. The buyer than has some period of time to inspect the goods and approve payment to the seller. These services charge anywhere from 2% - 15% of the purchase price.

While escrow services eliminate many of the risks associated with these transactions, they are used only a very small percentage of the time. These services are relatively new and the significant fees they charge may only make sense for high value items. Further, there are large markets for which escrow services are not an option. For example, there is a very healthy trade that exists for bootleg audio tapes. The ambiguous legal status of these tapes precludes the use of these services. There are also many markets in which goods are exchanged for goods in barter transactions. The value of these goods may be considerable, but participants are reluctant to spend money on escrow fees, and there is the issue of how escrow agencies would set fees on a non-cash transaction or administer the transfer of goods.

The fact remains that the vast majority of online transactions in these markets do not involve escrow services. Thus, the issue is to investigate the manner by which participants have endogenously managed the risks of trade.

Identity and Accountability

A collection of advice and trading tips has been assembled in a variety of documents created by experienced online traders (Kuhn 1997; Mak 1997; Barrett 1996; King 1996). A number of observers have ruefully noted that if a deal seems too good to be true, it probably is being offered as bait in hopes of taking advantage of the other party. Nevertheless, the working assumption is that most individuals can be trusted given reasonable precautions. Miscommunication and misunderstanding, however, often create difficulties among otherwise reputable parties. Hence, several observers stress the importance of frequent communication via email between the buyer and seller in order to thoroughly describe the items, the shipping and payment terms, when funds have cleared, when a shipment has been sent, when it has arrived, etc. Another piece of advice given in several markets is that one should start with a small trade when dealing with an unknown participant, and scale up the value of the trades if the initial exchanges go well.[6]

Most of the advice that is given centers on the issues of identifiability and accountability. Participants have a deep understanding of a lesson that is at the heart of work on social dilemmas: cooperation is much less likely when dealing with an anonymous actor. Hence, a variety of trading tips are offered that are directed toward evaluating someone’s physical and electronic identity.

As a matter of course, one is advised to ask for a person’s full name, address, and phone number, and to not trade with anyone who withholds this information. Along these lines, it is very common to see cautions against sending goods to someone using a PO box, either recommending that one should never trade in such circumstances or that one should insist on receiving payment before shipping the goods. Several observers recommend calling the other party, both to confirm the phone number and perhaps to also get a better sense of the person with whom one will trade. A related suggestion is to call back at some random time in order to guard against the possibility that the person originally gave the number of a public phone and simply waited around for one’s return call. Finally, some individuals have attempted to eliminate certain risks by only trading with those who are in the same geographic area. There are, for example, Usenet newsgroups that are specific to certain cities (e.g., dc.forsale) as well as web sites that target particular areas (e.g., CityAuction). In such cases, buyer and seller can meet in person in order to carry out the trade. While this certainly reduces the risks of the transaction, it also imposes costs (at least in terms of the time and effort needed to get together) and eliminates some of the most powerful advantages of online markets: access to a vast variety of goods and a global market of potential trading partners.

Participants are, of course, also very concerned about one’s online identity. If the interactions have taken place via a Usenet newsgroup or a web site, one is cautioned to always test the other person’s email address to insure it is a valid address and that the potential trading partner responds. And it is widely understood that there are reputable as well as questionable neighborhoods online, as indicated by domain names and email addresses. An offer to sell something from a trader with the address kollock@ucla.edu signals something very different than the same offer from pete@crunchy.net. A known university (ucla.edu), company (wsj.com), or government domain (jpl.nasa.gov) implies not just a certain respectability, but also a greater probability that one would be able to track down and hold a person accountable in the case of fraud.[7]

There is also broad suspicion, or at least caution, of individuals using America Online email addresses. The reason is not so much a general prejudice against users of America Online (although this exists) as it is an understanding that identity is not fixed in this system. When a person signs up for an AOL email account, he or she establishes a parent account that cannot be changed and is tied to the user’s personal information and credit card number. However, every user is also given a number of secondary accounts. The idea is that an entire family can operate through one account, with the parent account having certain controls over the secondary accounts. The problem is that the user name, and therefore the email address, of any of these secondary accounts can be changed at will and without cost. This makes it very easy to manufacture an email address that can be tossed aside after taking advantage of someone. While one can contact AOL in an attempt to hold the person accountable, this may not be worth the trouble except for high value items, and the common belief is that one is likely to get a very slow response or even no response from AOL.[8]

An even more difficult situation has been created by the emergence of free email services, which give anyone a free email address on demand (such sites support themselves through advertising revenue). Given that there is no financial relationship between free email services and their users, there is even less hope of actually tracking a person down or holding that person accountable. This is such a serious issue that some observers recommend using extreme care whenever dealing with a party using a free email service. One person reports that 80% of the bad traders at his site used these free email services (Mak 1997). Some market participants have also collected the names of these services so that one can look out for their domain names in email addresses.[9]

The advice discussed above is directed toward managing risks prior to the execution of the trade. If the trade occurs and something goes wrong, there are also a variety of suggestions that are offered. Given the everyday challenges of one’s work and personal schedule, the first piece of advice some people give is to show a little patience and not expect the goods or funds to arrive immediately. Increasingly frequent email and phone queries are recommended as time goes on. If the contact information for one’s trading partner was never gathered or is fraudulent, a number of observers recommend trying to track down the phone number and address of the person using a variety of search services available on the Internet.[10]  If it is clear the other party does not intend to uphold their side of the exchange, one still has the options of posting a report of the fraudulent behavior to other members of the market (discussed below) or pursuing legal action by making a claim of mail fraud to postal authorities or bringing the person to small claims court. However, such legal responses seem to be very rare.

NEGATIVE REPUTATION SYSTEMS

There is little satisfaction in suffering silently. Having experienced a bad trade, the common reaction is to complain publicly, perhaps in the hope that it might motivate one’s trading partner to make restitution or at least in order to warn others to watch out for this person in future trades. In our "offline" world, we might complain to our friends or business colleagues, or to an agency such as the Better Business Bureau in the case of a commercial enterprise. But on the Internet, we can complain to the world, and at almost no cost.

The Internet has radically reduced the costs of distributing information to a global audience. Thus, it is not surprising that individuals take advantage of these extraordinarily low distribution costs to air their grievances.[11]   One possibility is to simply set up a web page with a report of the fraudulent trade. In one case, I came across a web site in which the person had uploaded scanned images of bad checks, complete with the accused person’s name, address, phone number, and email address. However, such untargeted and unorganized displays are likely to be of limited value – one has to seek out this site and it contains the experiences of only a single trader.

In the case of some Usenet trading newsgroups, the common response after a bad trade is to post a report of the trade to the group itself and ask what experience others have had with the questionable trader. This has the advantage that traders in this market all have a chance to view the complaints.[12]  An interesting case study in this regard is the newsgroup alt.music.bootlegs. This group is devoted to trading bootleg audio and video tapes. Drawing from the group’s Frequently Asked Questions file (Kuhn 1997), bootlegs are defined as "recordings that have not been released by an artist’s main record label. They could be live recordings, studio outtakes, rehearsals, or just jams." The focus of the group is on the trading of tapes for tapes, although sales also occur. Members of the group make a strong distinction between bootlegging and pirating, which is "making copies of legitimate releases and selling them as if they were legitimate" (Kuhn 1997). This is a key point for members, who maintain that trading bootleg tapes for personal use is either legal (as it certainly is in the case of groups that permit personal taping of shows – most famously, The Grateful Dead) or at least does no harm. Indeed, attempts to buy or trade pirated tapes are usually met with strong disapproval and flaming within the newsgroup.

Nevertheless, the ambiguous legality of bootleg tapes forecloses the use of an escrow service, and using a COD service does not address the main concern of participants in this market: the quality of the tape. Thus, the group must rely on its own devices to manage the risks of trading.

Quality is such a serious issue because of the great variance in skill and equipment among those who tape at live concerts. There is also the question of a tape’s generation (e.g., a tape of a tape of the original recording is a second-generation tape), given that the quality of even an excellent recording deteriorates significantly with each generation. Thus, the participants face a lemons market (Akerlof 1970) in which goods are purchased before one can definitively ascertain the quality of the good.

One response has been to gather together the accumulated knowledge of the group in a very detailed Frequently Asked Questions file (Kuhn 1997). This document goes on for over 60 pages, providing extremely detailed specification about how to tape, what tapes to use, how to prepare the tapes for shipping, etc. There are even shared grading systems to rate the quality of tapes. In its detail the document rivals the contract and delivery specifications of commercial commodities.

As helpful as this information is, it certainly does not eliminate bad trades, whether the issue is tapes or funds that never arrived, or receiving a tape of poorer quality than was advertised. One is encouraged to first try to work things out with the other trader, but if such efforts are unsuccessful, it is customary to post a complaint to the newsgroup. This has the advantage of making the report public to participants of the market, and is the beginnings of a negative reputation system: a public system that distributes information on untrustworthy traders.

However, simply posting reports to the newsgroup has a number of shortcomings. If one is not constantly checking in with the newsgroup, it is possible to miss someone’s report on a trade gone bad. Because Usenet posting are automatically deleted after a period of time, the information can be lost even to someone who tries to go over all the messages posted since one’s last visit. It is also an inconvenient system in that the information is not aggregated.

Thus, it is not surprising that early in the group’s history there emerged efforts to create lists of suspect traders, known as blacklists. These list were either posted periodically to the group or posted on a web page. Blacklists create a kind of durable gossip that can be used as a reference by all group members in managing their risks in trading.[13]  While some lists simply noted the names of bad traders, others provided richer information, including the name and email address of both the accused and accuser as well as an explanation of what went wrong with the trade. And the extremely low information distribution costs meant that the lists could be made available to an unlimited number of individuals.

Almost immediately, however, participants realized there were problems with negative reputation systems such as blacklists. Both participants of these markets (e.g., Kuhn 1997) and scholars (e.g., Miller and Drexler 1988) have noted the same weaknesses. First, there is the risk that one can be wrongly accused. Most commonly, a person is added to a blacklist solely on the weight of a single person’s report, without any further fact-finding. In terms of signal detection models, the criterion threshold has been set in this case to err on the side of false alarms: avoiding trades with anyone for whom there is even the hint of untrustworthiness. Given the hazards of the transaction, it is perhaps understandable that any report of untrustworthiness should be enough to stain one’s reputation. But the serious danger here is that as a result of a misunderstanding, circumstances beyond an individual’s control, or the deliberate desire to damage someone’s name, a person will be falsely accused and suffer the consequences. As a result, some blacklists allow the accused to offer a response or even to be removed from the list if the accused reaches a satisfactory resolution with the accuser (the accuser then contacts the manager of the list to say the trade has been settled). Such procedures still leave blacklists open to mistakes and manipulation.

There is also a second weakness in negative reputation systems that condemns such systems to failure in most online environments. Blacklists are ineffective to the extent that it is easy for an individual to assume a new identity and so cast off the negative reputation. While this is difficult in the physical world (though not impossible, as witness relocation programs demonstrate), changing identities online is trivially easy, especially given the rise of free email services. Thus, it is probably the case that pure negative reputations systems are doomed to failure in current online environments.[14]

POSITIVE REPUTATION SYSTEMS

An alternative way to structure a reputation system is to base it on positive references. A positive reputation system attends to the other tail in the distribution of trustworthiness – rather than marking and avoiding untrustworthy traders, the point is to mark and seek out those market participants who have a history of successful trades. This can take the form of individual traders appending a list of positive references (with email addresses) to their post, or aggregated lists of good traders with references.

The requirements of a positive reputation system are less onerous than a negative system: the key requisite condition is that one cannot easily claim the identity of another (known as spoofing in many online groups), so that one cannot improperly damage or free-ride on that person’s reputation. Changing identities serves no purpose if an individual has a positive reputation as that individual cannot carry his or her reputation capital to the new identity. It would still be possible for an untrustworthy trader to continually change identities as a way of at least having a neutral reputation, but if others will only trade with those who have established successful track records, or impose additional conditions on those who have no track record (see below), this will not be a successful strategy.

An interesting case study in the use of positive reputation systems can be found in the web-based trading posts that have been established for game cards. The greatest amount of trading seems to revolve around the game Magic. Magic is a sophisticated strategy card game that inspires a near-obsessive devotion in many of its players. Drawing from a large population of cards, a player assembles a deck of cards that is used in a contest against another player’s deck. The skill comes in picking a general strategy among the many that are available as well as in the tactical decision of how to assemble the deck. The game is constructed such that no one strategy is likely to dominate play – each approach has its strengths and limitations and is particularly useful against certain other families of strategies and particularly vulnerable against others. In addition, the manufacturer of the game periodically releases new cards in order to keep the game fresh. Given that one purchases a pack of cards without knowing the contents of the pack, assembling a particular deck can be an expensive proposition and result in many unwanted cards.

For all these reasons, a huge secondary market quickly emerged in which players traded (or sold) cards to each other. For many individuals, trading Magic cards is a serious activity that has become more important than the original game. The is no shortage of strategic challenges in trading the cards and there is the possibility of taking on speculative positions as the value of certain cards ebb and flow. There are even individuals involved in this secondary market for whom selling cards is a significant or primary source of income. A valuable card can command hundreds of dollars and there is a global market of players.

The sites I focused on are concerned with the barter exchange of cards for cards. Hence Collect On Delivery services are irrelevant. Escrow services are also not useful because it is a barter exchange. Thus, the participants in this market must develop their own mechanisms to manage risky exchange. These risks are quite serious: while the dollar value of many transactions may not be that great, the subjective value of the cards to the players is very substantial. And occasionally, large-scale frauds are committed.

An elegant procedure has been developed based on acquiring positive references from previous trading partners (e.g., Burrows 1997; Mak 1997). The norm is that the person with the fewer references sends his or her cards first. Once the cards are received and inspected, the trader with the greater number of references then sends his or her cards to complete the trade. Traders with approximately equal number of references are expected to "simul-send" – sending their cards to each other at the same time. Note that this structure means that a neutral reputation (no references) puts one at a disadvantage and so provides a disincentive for switching identities.

Another implication of this structure is that a new entrant to the market will have to send their goods first until they establish a reputation. Because the newcomer bears a disproportionate amount of the risk in these trades, it is suggested that one begin by proposing trades with those who have established long lists of references as a way of mitigating one’s own risk. Further, every attempt should be made to be sure the person with whom one is trading is happy with the terms and execution of the trade so that one might ask to use the person as a reference. As one participant puts it: "Net refs are like golden eggs and should be earned by fairly trading with everyone you trade. Nurture them. Cultivate each trade as though all your future trades depended on each one. They do! They are you[r] ticket to not having always to ship first" (Burrows 1997).

Merely seeing a list of references is not enough. Cases are recounted of traders who simply copied someone else’s set of references in hopes of masquerading as a reputable trader. For this reason, participants are cautioned to always check at least a subset of the references to make sure the email address is valid and that the reference has in fact traded successfully with the person in question. References are discounted or rejected if they are from individuals with email addresses from the same domain as the trader in question. The reason is that it is assumed the references may just be alternate accounts the trader has set up in order to inflate his or her reputation. References using email address from America Online or one of the free email serves are particularly suspect, for reasons that were discussed above.

It is understood that such efforts can decrease the risks of trade but not eliminate them. A particularly disturbing strategy for fraud that has been seen in a number of online markets is the person who works hard at establishing a trustworthy reputation and then sets up a whole series of significant trades and defaults on all them, disappearing into a new identity after the fact.[15]

Aggregated lists of trustworthy traders have also been established. Sometimes anyone with positive references are listed, with the names of the references following their name. On other lists there is a criterion that must be met, such as a minimum of 20 references, all from different domains than one’s own, and all subject to verification prior to admission to the list (Burrows 1997).

INSTITUTIONALIZED GOSSIP: ONLINE AUCTION HOUSES

Many of these reputation-based methods of risk management are now being institutionalized by for-profit companies that have set up online auction houses. While some online auctions sell off new merchandise from manufactures, my focus is on person-to-person auction sites in which individuals are trading with other individuals. These person-to-person auction companies are still informal markets in the sense that there is no formal clearing mechanism and the companies do not guarantee the terms of the trade.[16]

In exchange for a commission on the sale, and sometimes a listing fee, anyone with Internet access can post an item for auction at these sites. The posting can include an elaborate description and pictures, and interested parties can place bids for the item. Almost any object imaginable has been offered for auction, although illegal goods are banned and adult material is sometimes restricted or banned as well.

The best designed auction sites address a number of limitations that exist in the Usenet forsale newsgroups. In the Usenet groups, it can take time for listings to propagate to all the servers, traders have little control about when a post to a Usenet group will disappear, it can be difficult to estimate the market value of an item, and there is no structured, institutionalize system for collecting and distributing information about participants’ reputations. Instead, the better online auction markets offer sites in which:

• Listings appear immediately after they’ve been posted.
• The seller has direct control over how long the offer is available by specifying the length of an auction in days.
• Buyers get immediate feedback about the current best offer for an item, and know exactly when the auction for that item will end.
• Buyers and sellers can get a good sense of fair market value by reviewing the history of auctions for similar items.
• Buyers and sellers are both subject to peer review. [17]
  

As of autumn 1998, there are thousands of auction sites on the web and some of these sites are huge and growing exponentially. I have focused my attention on those person-to-person auction sites which are the largest or have been highly rated for their design and offerings (Hughes 1997).

EBay Auctions

Currently the largest, and one of the oldest (launched September 1995), person-to-person auction houses is eBay.[18]  Figures from autumn 1998 show it has approximately 1 million items for sale on any given day, over 1.2 million registered users, and had over 195 million dollars in gross merchandise sales for the third quarter of 1998. Since its inception, more than 30 million auctions have been completed at the site, and eBay boasts a sell-through rate of approximately 60% (e.g., the percentage of offered items that result in a sale). It has grown at the rate of 20-30% per month and has been profitable since its inception. The site receives over 20 million hits per day and more time is spent at this site by Internet users than at any other commerce site. In fact, only Yahoo! and AOL rank above eBay among all web sites in terms of time spent at the site.[19]

Remarkably, eBay offers no warranties or guarantees for any of the goods that are auctioned off – buyers and sellers assume all risks for the transaction, with eBay serving as a listing agency. It would seem to be a market ripe with the possibility of large-scale fraud and deceit, and yet the default rate for trades conducted through eBay is remarkably small. Both eBay and the participants in its market credit an institutionalized reputation system at the site – known as the Feedback Forum – for the very high rate of successful trades.

After every seller’s or bidder’s name is a number in parenthesis. In the case of a seller, the information is displayed as follows:

Seller       name@company.com (265)

       (view seller’s  feedback) (view seller’s other auctions) (ask seller a question)

The number is a summary measure of a person’s reputation in the eBay market. Registered users are allowed to post positive, negative, and neutral comments about users with whom they have traded. Each positive comment is given a score of +1, each negative comment is given a score of –1, with neutral comments not affecting one’s score in either direction. Thus a score of 10 might mean 10 positive comments, or 110 positive comments, 100 negative comments, and any number of neutral comments. At certain levels, market participants are also awarded a color star which marks the number of net positive comments they have received (e.g., a rating of 10-99 receives a yellow star, a rating of 100-499 receives a turquoise star, a rating of 500-999 receives a purple star, a rating of 1,000-9,000 receives a red star, and a shooting star is reserved for those with a rating of 10,000 or higher). An interesting enforcement mechanism is that eBay users with a net negative score of –4 or lower are automatically barred from trading by the software system.

One is able to contact the person via email by clicking on the name, and clicking on the number following someone’s name leads to their full feedback profile. There one finds the full list of comments, with email links and ratings numbers for every evaluator as well (thus, one can explore the reputation of the evaluators just as one can for the evaluated). A typical positive comment might be "Well packaged, fast delivery. Highly recommended. A+." There is also a summary table at the head of the comments:

While a user may rate someone an indefinite number of times, he or she can affect that person’s rating by at most one point. This restriction prevents both attempts to completely undermine someone’s reputation as well as attempts to inflate someone’s rating (e.g., a circle of friends pushing each other’s ratings up in a never ending spiral).[20]  The summary table also notes if some of the comments came from users who are no longer registered participants of the market. Comments from someone who is no longer registered are automatically converted to a neutral rating.

A high feedback rating is an extremely valuable asset. Many participants report that they are more willing to trade with someone with a high rating, or even that they will only trade with individuals with high ratings. In that sense, some traders are able to create a brand identity that increases their volume of sales or even the price at which they are able to sell items. On a rotating basis, eBay has even featured some of its most highly rated traders on its opening web page. Such public displays of highly rated traders are likely to be a powerful motivator, both in terms of the status of being marked as a top trader and because it is likely to send more bidders to one’s auctions. One auction house – Auction Universe [21] – even maintains a constantly updated list of its top 50 traders based on reputation rating. For all these reasons, users in these auction markets often stress the importance of always adding a positive comment to someone’s feedback profile when a trade goes well.

Even a few negative ratings can seriously damage a reputation, and so frequent traders are very careful about nurturing their rating by providing swift execution of honest trades. The potential damage of a negative comment is a subject of great concern among frequent participants. Both eBay and informal web sites that have been set up by users caution that one should post a negative comment only after extensive efforts at trying to resolve the difficulty directly with the other party. The fact that any one user can only affect someone’s rating by one point serves as a limit to the amount of damage one can do. Nevertheless, it is inevitable that negative rating are posted (sometimes hastily) which the counter party regards as illegitimate.

Interestingly, eBay has a policy that once a comment is posted, neither eBay nor any user can delete that comment. One can post a rebuttal to a comment to give the other side of the story [22], but it is not possible to retract a comment if both parties have reached an understanding later in time. Nor is it possible to selectively edit or censor comments. One can choose to make one’s entire feedback profile private, but this is a huge disadvantage in a market which relies on these reputations.

Other Online Auction Sites

While eBay is by far the largest person-to-person auction market at this time, it is useful to compare other prominent auction sites to see how their systems are differently structured. Risk management at each of the auctions houses I will discuss is also centered around a reputation system.

One such auction site is Haggle Online.[23]  Unlike eBay, this auction system disaggregates the single ratings number into two totals: one for positive comments and one for negative comments. In the case of a seller, the following information is displayed:

Splitting the positive and negative comments eliminates some of the ambiguity that occurs in eBay’s one number rating. The length of time one has been a participant in this market is also noted. Haggle Online does not display the number of neutral comments in the summary ratings, though of course this could be done very simply.

In fact, another auction site – CityAuction [24] – display totals for positive, neutral, and negative comments after a users name, preceded by the total number of trades in which the user has been involved:

Seller:   name@company.com (365: 265+ / 4 / 2-)

  (view feedback on seller) (view seller's other auctions) (contact seller)

Listing the total number of trades is useful as a comparison to total number of comments, in order to get a sense of what proportion of trades have been rated. Also, having information about both length of time in the market and number of trades could be helpful in evaluating the experience of a trader and determining the rate at which the trader participates in auctions.

A different system of arriving at a summary measure of reputation was used by Onsale Exchange. Onsale is a new merchandise auction house that attempted to open up a separate section – known as Onsale Exchange – for person-to-person auctions.[25]   In this system, rather than being given a binary choice between coding one’s feedback comments as positive or negative (or perhaps neutral), participants were asked to rate their satisfaction with the trade on a scale of 0-5. What appeared after the trader’s name was his or her average rating, displayed in stars. Thus, in the case of a seller, the format was as follows:

JK of Los Angeles, CA, USA About the Seller Current rating:

Giving users something other than a dichotomous choice in categorizing their comments promises a more fine-grained rating system, although the fact that there are only 5 discrete summary ratings displayed (one cannot display a rating of, e.g., 4.32 stars) means that a lot of information is lost.

Just as in eBay, clicking on the summary rating at each of these auction sites takes one to the actual list of comments. Here again we see a variety of different formats. For Haggle Online, the name and email address for each evaluator is include, but not the evaluator’s rating. Nor is a summary table of comments for recent time periods given, as is the case on eBay. However, Haggle Online does provide a link for each comment that references the transaction in question, which means one can recover the value of the good, how active the bidding was, etc.[26]  Haggle Online also provides a very different way of processing comments. When a participant submits a comment about a trader, the comment is first sent to that trader, who has the option of publishing the comment or censoring it. If the comment is censored, this is noted on the trader’s list of comments. It is also possible for a comment to be retracted by the person who originally made it if, for example, an initially troublesome trade was eventually settled to the satisfaction of both parties. When a comment is retracted, the text disappears, but the fact that a comment has been retracted is explicitly noted on the trader’s feedback page.

This different format has a number of implications. Unlike eBay, on Haggle Online it is possible to exercise a "line-item veto" in which only certain comments are censored. It is also possible to repair (at least in part) a negative comment if the evaluator later agrees to retract the comment. That fact that such actions are noted on the trader’s feedback page mean that a history of troublesome trades cannot be completely hidden, even if they are not reflected in the summary rating numbers.

The options for handling comments on CityAuction are very similar to eBay: individuals can respond to the comments evaluators have made or block all their comments from being seen, but currently one cannot selectively censor or retract comments. On Onsale Exchange, such options were also not possible, and the site had the curious policy of only listing the most recent comments about an individual. The attempt may have been to avoid information overload or save on storage space, but the absence of a complete feedback history for a trader means a great deal of important information is lost. This is especially true given that the summary reputation measure is an average, and so one has no information about the total number of 5-star, 4-star, etc. comments or the total number of trades in which the person had been involved.

The Impact of Reputation

The efficacy of these reputation systems to manage the risks of unsecured trades seems to be impressive. Two years into its history (summer 1997), eBay released a report stating that of the 2 million auctions that occurred from May through August 1997, only 27 were considered to involve possible criminal fraud (these cases were referred to the US Postal Authority for prosecution as mail fraud). The report by eBay also stated that over 99.99% of auctions attracting bids were successfully complete (reported in Hughes 1997; the same low levels of fraud were also reported by eBay’s CEO in an October 1998 article – Chervitz 1998). An online publication (Hughes 1997) which reported on Internet auction sites also stated that its own research found that negative comments on trades were quite rare, accounting for less than 1% of feedback on eBay. Once a trader has established a positive reputation there is a very great incentive to maintain and improve on one’s rating. And traders with negative reputations are selected out by at least two mechanism: other market participants will be reluctant to trade with them, and at a certain point (a net negative rating of –4), the software prohibits further trading. Participants in these reputation-based auction markets also claim that the vast majority of trades are successful despite the inherent risk.[27]

Conceptually, we should expect reputations to affect not just rates of cooperation, but also the price of goods in these markets. If these reputation systems do in fact provide useful information and an incentive to behave in a trustworthy manner, buyers should be willing to pay more for a good if it comes from a highly rated seller, at least when the transaction involves significant risk. Preliminary evidence from a quantitative study of reputations on eBay suggests this is in fact the case (Kollock 1998c). At least for some high value goods, the seller’s reputation had a positive and statistically significant effect on the price buyers paid for identical goods of equivalent quality. This effect of reputation seems to diminish or disappear for low value goods.[28]

DISCUSSION

In what seems at first to be a very unpromising environment, participants in informal online markets have created an elegant and efficacious set of solutions for managing risk.

Some of these markets have also been elaborated in impressive ways by the participants. Users of eBay have created their own independent web sites in which advice, strategy, and answers to frequently asked questions are collected. Some experienced eBay traders have even set themselves up as for-fee consultants, offering new users advice on how to set up an auction to attract attention. But perhaps more noteworthy are the number of experienced traders who offer free help and advice to others on a continuing basis. While there are undoubtedly some individuals who visit eBay only once or sporadically to buy particular objects, there are also core groups of traders in each category area (e.g., trading cards, coins, antiques, Beanie Babies, etc.) who are there continually and come to know each other. EBay has encourage the formation of these communities of traders by providing message boards so that users can interact with each other. Both casual interaction and serious discussions occur here, including such issues as spotting counterfeit goods or warnings about rouge traders. Some traders have even created "neighborhood" watch groups, watching over particular categories of auctions for violations of trading rules or suspicious activity. Further encouraging a sense of identity and commitment, eBay has recently begun offering its registered users free personal web pages, where traders can post information about themselves and their auctions. The sense of fellowship felt in some of there trading communities is striking. There are even cases of traders contributing goods or money to another participant who could not otherwise afford a particular object. At least for the core users, this is not a market of atomized price-takers.

Requisite Features

As impressive as these risk management techniques seem to be, it is important to ask what the prerequisites are for an online reputation system. This speaks to the question of how generalizable these solutions are to other settings.

There are a number of elements that must be in place in order for an online reputation system to work effectively and provide an incentive for trustworthiness in trading. First, these systems require that the costs for submitting and distributing ratings are very low. This is, of course, one of the great advantages of online trading – distributions costs can approach zero, and the system software can also be used as an automatic monitoring agent, prohibiting, for example, individuals with low ratings from trading.

Second, in order to be effective, these reputation-based markets require that there be many alternative trading partners. It is the fact that one can easily go to another trader with a better reputation that serves as a powerful motivator to create and maintain a positive reputation. If a single trader or cartel has near monopoly control over particular desired goods, one may have to trade despite a questionable reputation.

Third, the costs involved in impersonating another trader must be very great, in order to prevent a trader from damaging or free-riding on another trader’s reputation.

Finally, the reputation server itself must be trustworthy. Whether it is an individual or a company that is collecting and distributing reputation ratings, they lose their value if it is believed there is negligence or deceit on the part of the owner of the system, or if it is believed that the software does not work properly.

Further Studies and Design Issues

The success of these markets reinforces the importance of some recent trends in theoretical and experimental work on social dilemmas. Early work used models in which actors could neither leave the interaction nor choose their partners. More recently, researchers investigating the dynamics of cooperation have relaxed these constraints, examining settings in which actors have the option of choosing new exchange partners (e.g., Hayashi 1993, Yamagishi et al. 1994) or even choosing the game structure (i.e., the value of how much is at risk; Kakiuchi & Yamagishi 1997). Such options are clearly important to the working of these markets, where participants can decide with whom to trade or the level of risk they want to take on. The intriguing success of these markets invites controlled experimental studies that tease out the structural features that encourage cooperation.

And as well-structured as some of these reputation systems are, they by no means exhaust the possibilities. Both in terms of designing better systems for actual use and in order to study these systems rigorously, it is important to lay out the elements of a distributed reputation system and ask about the different ways in which it might be constructed. Below I outline some of the key design issues and some different possible formats:

Evaluators: The first issue that needs to be addressed is, who is permitted to post an evaluation? "Anyone," is one possible answer. Somewhat more restrictive would be to let only registered participants in a market submit comments about a trader. Some of the auction houses I have examined further restrict the ability to comment on a transaction to only those who have actually placed a bid.[29]  It would be possible to be even more restrictive and only allow those who have completed a trade with someone to comment on that person’s trustworthiness. However, there are time when it seems natural to allow at least other bidders to offer feedback even if they have not consummated a trade – for example, to protest a seller ending an auction earlier than scheduled without offering a valid excuse. Other sorts of restrictions might be added: eBay allows an individual to affect another’s reputation rating by at most one point, and the rating of comments from a person who is no longer registered are automatically converted to neutral. CityAuction requires a 3 day wait after the close of an auction before submitting comments. Presumably, CityAuction hopes to diminish the number of comments that might be written in haste, and allow the parties a reasonable amount of time to complete the details of the transaction.

Evaluations: A very large issue is how the evaluations themselves should be structured. To date, the common format is to allow a free-form comment that is categorized as positive, negative, or perhaps neutral. Other possibilities include allowing ratings on a continuous scale, or perhaps using a structured set of survey questions as a substitute or addition to free-form comments. Given that many of the comments have started to follow a common pattern (e.g., commenting on the quality of goods, how well the goods were packed for mailing, how quickly the other trader answered questions and passed on information, and a grade such as "A+"), it would be reasonably easy to create a set of questionnaire items that the evaluator could fill out. Collecting comments in this way would also permit other sorts of data analyses and ratings, such as separate ratings for quality of packaging or speed of service.

History of Evaluations: The aggregated collection of comments for a trader might simply be put together as one long list. But even in this simple case there are a number of design questions. Should every comment be displayed or only the most recent comments (as was the case for Onsale Exchange)? What information for each evaluators should be displayed? Haggle Online, for example, provides the email address for all evaluators, as well as a link to the transaction that is being commented on, while eBay lists each evaluator’s reputation rating. Listing the evaluator’s rating can be an important way of deciding how much weight to place on a comment, and I have seen recipients of negative comments pointing out the evaluator’s low rating, as a way of discounting the feedback. Some auction houses also try to digest the list of comments in order to make it easier to evaluate. For example, eBay provides a table showing the breakdown of positive, negative, and neutral comments for the past 7 days, past month, and past 6 months. Tables and other ways of distilling the history of evaluations are likely to become more important over time, as the list of comments for active traders extends into the thousands.

Summary Measures: Each of the auction houses examined also recognized the importance of providing some sort of short, summary measures that could be appended to a person’s name. Possibilities here include: totals for positive and negative comments, net positive comments (positive – negative), an average evaluation score (as in Onsale Exchange’s five star system), total number of trades, and starting date as a participant in the market. Other criteria and other measures of central tendency could also be explore – for example, might it be useful to report median scores as well as the mean, or to weight the rating one receives by the value of the good traded? And apart from the issue of what summary measures should be appended to a trader’s name, there is the question of when this information should be displayed: When one is a seller? A buyer? An evaluator? At all times? Finally, highlighting traders with exceptionally impressive reputations is likely to have powerful effects. Auction Universe, as an example, provided a constantly updated list of the 50 traders with the highest ratings.

Modifying Evaluations: Another design issue is the extend to which posted evaluations can be censored, retracted, or commented on. The strictest policy, such as the one eBay follows, is to not allow individuals to censor specific comments or evaluators to retract comments – the only option is to post a reply to a comment. Haggle online, as a contrast, permits both censoring individual comments and the retraction of comments, although in both cases these actions are noted in the trader’s permanent history. Another option would be to allow censoring or retracting comments without the "residue" of a note indicating the action.

Evaluating the relative strengths of different designs suggests a whole set of experimental studies. In general, online markets represent an extraordinary opportunity to study the dynamics of exchange and cooperation. In-depth field studies, quantitative analyses of market data, and experimental studies of the efficacy of different reputation systems should all be pursued.

REFERENCES

Akerlof, George A. 1970. "The Market for ‘Lemons’: Quality Uncertainty and the Market Mechanism." Quarterly Journal of Economics 84:488-500.

Avery, Chris, Paul Resnick, and Richard Zeckhauser. Forthcoming. "The Market for Evaluations." American Economic Review.

Barrett, Daniel J. 1996. "Buying and Selling on the Net." Pp. 83-98 in Bandits on the Information Superhighway. Cambridge, MA: O’Reilly & Associates.

Burrows, Dave. 1997. Dave’s Magic Trading Post. Web site: http://www.cobweb.net/~quail/magic.htm

Chervitz, Darren. 1998. "IPO First Words." CBS MarketWatch (October 26). Web site: http://cbs.marketwatch.com.

Dawes, Robyn. 1980. "Social Dilemmas." Annual Review of Psychology 31:169-193.

Friedman, Eric and Paul Resnick. 1998. "The Social Cost of Cheap Pseudonyms: Fostering Cooperation on the Internet." Proceedings of the 1998 Telecommunications Policy Research Conference.

Hayashi, N. 1993. "From Tit-For-Tat to Out-For-Tat: The Dilemma of The Prisoner’s Network." Sociological Theory and Methods:8, 1(13), 19-32. (in Japanese)

Hughes, G. Patton. 1997. Auction Land Online Report. Web site: http://www.neomax.com/ (now defunct).

Jaycobs, Richard. 1998. "The Internet: The Ultimate Solution?" Futures Industry 8(4):11-13.

Kakiuchi, R. and Toshio Yamagishi. 1997. "The Dilemma of Trust." Seventh International Conference on Social Dilemmas. Cairns, Australia

King, Dan. 1996. The Usenet Marketplace FAQ. Web site: http://www.fmn.net/FAQ/index.html

Kollock, Peter. 1994. "The Emergence of Exchange Structures: An Experimental Study of Uncertainty, Commitment, and Trust." American Journal of Sociology 100(2):313-345.

Kollock, Peter. 1998a. "Social Dilemmas: The Anatomy of Cooperation." Annual Review of Sociology 24:183-214.

Kollock, Peter. 1998b. "The Economies of Online Cooperation: Gifts and Public Goods in Computer Communities." In Communities in Cyberspace, edited by Marc Smith and Peter Kollock. London: Routledge.

Kollock, Peter. 1998c. "The Value of Reputation." Working paper, UCLA.

Kuhn, Jeff. 1997. Official Frequently Asked Questions for alt.music.bootlegs. FTP: ftp://ftp.visi.com/users/astanley/ambfaq.txt

Lerner, Abba. 1972. "The Economics and Politics of Consumer Sovereignty.: American Economic Review 62(2):258-66.

Mak, Innate. 1997. Beyond Dominia. Web site: http://www.bdominia.com/

Messick, David M., and Marilynn B. Brewer. 1983. "Solving Social Dilemmas." Pp. 11-44 in Review of Personality and Social Psychology (Vol. 4), edited by L. Wheeler and P. Shaver. Beverly Hills, CA: Sage.

Miller, Mark S., and K. Eric Drexler. 1988. "Markets and Computation: Agorics Open Systems" In The Ecology of Computation, edited by Bernardo Huberman. Elsevier Science Publishers/North-Holland.

Rapoport, Anatol, Andreas Diekmann, and Axel Franzen. 1995. "Experiments with Social Traps IV: Reputations Effects in the Evolution of Cooperation." Rationality and Society 7(4):431-441.

Raub, Werner, and Jeroen Weesie. 1990. "Reputation and Efficiency in Social Interactions: An Example of Network Effects." American Journal of Sociology 96(3):626-54.

Wilson, Robert. 1985. "Reputations in Games and Markets." Pp. 27-62 in Game-Theoretic Models of Bargaining, edited by Alvin Roth. Cambridge: Cambridge University Press.

Yamagishi, T., N. Hayashi, and N. Jin. 1994. "Prisoner's Dilemma Networks: Selection Strategy Versus Action Strategy." Pp. 233-250 in Social Dilemmas and Cooperation, edited by U. Schulz, W. Albers, and U. Mueller. New York: Springer-Verlag

Yamagishi, Toshio, and Midori Yamagishi. 1994. "Trust and Commitment in the United States and Japan." Motivation and Emotion 18(2):129-166.

Footnotes

[29]  Sometimes this restriction exists only as a norm; other times it is programmed into the software