2.3 Concurrent Bilateral Negotiation Strategies
2.3.1 The Case for Concurrent Bilateral Negotiation
2.3.1.4 Concurrent Bilateral Negotiation versus Auctions . 59
In the wake of success of many auction web sites, such as eBay.com, online auctions and exchanges are seen as an essential part of e-commerce (Kambil and van Heck 2002). Annual online auction sales exceed already a few years ago $30 billion (David et al. 2005). This popularity makes auctions a baseline in all types of electronic markets and any research in a new or less popular mechanism, like concurrent bilateral negotiation, must justify itself by showing that the mechanism in question can provide something that auctions can not. This is not necessarily easy, since the popularity of auctions is not a coincidence: Auctions do have many desirable properties. In particular:
• Legitimacy: Most auctions have very simple and clear rules. This creates two major benefits:
42With leveled-commitment contracts (see section2.2.3), there may be a cost involved in the latter action.
– Publicity: During an auction, the bids can be open or at the very least the winning bid is usually public.
– Controllability: Every participant can be certain that the rules are fol-lowed, since any deviation can be noticed. For example, the publication of the winning bid offers a way for the other participants to control that they were genuinely outbid.
– Neutrality: In most types of auction, the identity of bidder is not rele-vant, but only the bid.
• Analysability: The simple and clear rules also make it possible to analyse auctions mathematically. To this end, auction theory (see Krishna (2002) for an introduction) is one of the most active research areas in game theory.
• Efficiency: Auctions usually allocate the sold item or service to the person or group that values it most. In economics, it is usually assumed that the person with the highest valuation knows the best use for the service and therefore auctions promote efficient use of resources.
• Maximises profit of the auctioneer : When used properly, auctions can pro-vide the highest possible price for the seller or the lowest possible price for the buyer.
However, auctions are not a panacea. In particular, in dynamic markets such as ours, at least in some circumstances, the following properties might be considered drawbacks:
• Inflexibility: Clear and simple rules also mean that there are no exceptions or room for changes.
– changes difficult: Once the auction has started, it is usually very diffi-cult to change the rules of the auction43 or the details of the auctioned task.44
43An auction usually has either a fixed deadline or it closes after a certain period of inactivity (i.e. when nobody increases the standing offer for a specified time). In a dynamic environment, where requirements can change at any time, both are problematic. In the case of a deadline, there can be no agreement before it. The bidder may be bound to his offer for a long time, if the deadline is distant and he is not outbid. During this time his requirements may change.
On the other hand, if only a period of inactivity closes the auction, the exact end time may be difficult to estimate (Nguyen and Jennings 2005). Such open-endedness is very problematic for an auctioneer with a fixed deadline.
44For example, in eBay (http://www.ebay.com) a revision of the listing is only pos-sible if an item has received no bids and the deadline is at least 12 hours away
– formulaic information dissemination: When buying complicated and configurable tasks, the consumer must explain his requirements (winner determination rule) in detail to all potential providers, so that they can make meaningful bids. In some cases, companies or people may not want everybody to know what they need (Bajari et al. 2008) and so the only way to limit the flow of information, is to limit the number of bidders. Also the winning bid must usually be revealed, so that other bidders can ensure that they lost fairly. This may give useful information to the winner’s competitors.
– one-way information flow : Once the auction starts, the only informa-tion flowing between the parties is the bids. In some cases, the aucinforma-tion- auction-eer might notice some error in his call for bids or his prefences when bids start. At that point, changes are no longer possible.
– heavy planning phase: Since the auction itself is simple and clear, it often means that a lot of effort must be invested in the preceding phases (Bajari et al. 2008). For example, in a reverse auction, the consumer must find out what the relevant options and their characteristics are and what his preferences are, so that the winner determination rule can be written.
• Neutrality: In some cases, it might matter to the consumer, who the provider is and he may not want to make his preferences public.45
• Complexity: Although auctions are usually simple, they can become quite complex, especially computationally. In combinatorial auctions, for exam-ple, where bids on multiple services are allowed, the winner determination problem is NP-hard (for an introduction, see Lehmann et al. (2006)).
Concurrent bilateral negotiation, on the other hand, is very flexible, allows two-sided information exchange and also allows the party to control information dis-semination and complexity and freely discriminate between different opponents.
In addition, it can offer some strategic advantages. We will now discuss each of these benefits, one at a time.
(http://pages.ebay.com/help/sell/edit listing.html). Also cancelling the entire auction (for any reason) becomes impossible when there is an acceptable offer (i.e. one that exceeds the reserve price) and there is less than 12 hours until the deadline.
45Neutrality was also mentioned as a benefit of auction mechanisms. Whether it is a benefit or a drawback, depends on the circumstances.
First, flexibility means that in concurrent bilateral negotiation, the agent can change its goals easily at any time. Since the deadlines and other negotiation pa-rameters are private (requirement R6), changing them is always possible. More-over, the agent always has several options to choose from (the opponents’ last offers), if it suddenly needs to close the deal (Nguyen and Jennings 2005). In short, the agent can take into account any new information or change in environ-ment or requireenviron-ments very quickly.
Second, the possibility of two-sided information exchange has been recognised as one of the main advantages of concurrent bilateral negotiation (Nguyen and Jennings 2005; Rahwan et al. 2002). Another related advantage is the possibility of controlled information dissemination. Thus, the agent can start from a very general task description and reveal the details gradually. In addition, he can handpick the opponents to whom he gives any strategic information. The agent can also learn something from the opponents’ offers that enable it to re-evaluate its preferences. For example, the option that the consumer thought to be the best, can prove to be too expensive or an option that it earlier hardly considered can prove to be very interesting. The flexibility allows the agent to ‘change its mind’, since it is not bound by any call for proposals. This is particularly useful when the agent does not know what the relevant options are or how much they cost.
Third, concurrent bilateral negotiation offers a way to manage complexity in in-terconnected negotiations. In combinatorial auctions, all parties first send bids to the auctioneer, who then tries to find the optimal winners among these bids. In contrast, in concurrent bilateral negotiation each party manages its own depen-dencies. This alone reduces a big problem into several small ones and effectively distributes the problem. In addition, each player can decide how complicated the dependencies he wants to consider should be and other players’ dependencies do not have any effect on his problem.46 This could be a major advantage for con-current bilateral negotiation, which is why we will investigate also interconnected negotiations. However, the concurrent bilateral negotiation is no silver bullet: it does not necessarily offer an optimal solution, because suboptimal contracts be-tween parties are not only possible, but also likely. This means that there is no guarantee that each service or service combination will go to the consumer who values them most. Also negotiation strategies and pure luck have an impact on
46Moreover, there is no deadline for bids, but the process is continuous, new parties can enter and old ones can leave at any time (requirementR5). However, there is a specific kind of auction called a continuous double auction that is able to manage this type of dynamism (Krishna 2002).
Variations of this auction are used, for example, at all the major stock exchanges (Das et al.
2001).
the results. Therefore, concurrent bilateral negotiation will offer ‘only’ a solution that will produce reasonable results and will scale well to markets of any size.
Finally, the strategic advantages in concurrent bilateral negotiation come from two directions: a possibility to use different strategies with different opponents (Nguyen and Jennings 2005; Rahwan et al. 2002) and a possibility to use informa-tion from one negotiainforma-tion in others (Nguyen and Jennings 2005). Since we assume that the providers are heterogeneous (requirementR3), it may well be reasonable to use different tactics against different providers. The consumer agent may, for example, adopt a stubborn stance against a low quality or unreliable provider, but be more willing to compromise with a high quality or especially reliable provider.
And since the negotiation threads are separate (in the sense that the opponents do not know what is happening in the other negotiations), the opponents do not have to know that we consider them to be unreliable or to offer low quality. On the other hand, we can use information from the other negotiations. Thus, a good offer in one thread can be taken into account in others. Moreover, in a multi-issue negotiation, some new promising combination of properties that the agent did not think of, but an opponent did, can also be used in other negotiations.
Given these facts, it is hardly surprising that in practice private entities prefer ne-gotiation even in situations where the law requires the public entities to organise a competitive bidding process (Rothkopf and Harstad 1994).47 Since the private (unlike public) entities are free to choose, this result would indicate that negotia-tions are a more efficient means of transferring complicated assets than aucnegotia-tions and produce no worse results for the sellers (Rothkopf and Harstad 1994). There-fore we can be confident that our approach is a valid one. We will then discuss the details of this approach.
2.3.2 Managing Dynamism in Bilateral Negotiations
We start our discussion about concurrent bilateral negotiation strategies by in-vestigating the simplest basic component of concurrent bilateral negotiation, the
47For example,Bajari et al.(2008) discovered that in California only about 15 % of the private building projects were awarded after a competitive bidding process. This was in clear contrast to public building projects where Californian law (like the law in many parts of Western world) requires a competitive bidding process for all significant projects and an auction was organised for almost all building projects. Public procurement processes also have other goals than efficiency (such as transparency and equality), so this is probably not as surprising as it may sound at first.
bilateral negotiation itself. We discussed the basic approaches to bilateral nego-tiation earlier (section 2.2.1). As we mentioned then, these basic approaches do not explicitly (or often even implicitly) take into consideration the changes in the environment. In game-theoretic and other mathematical approaches the reason is simple: the possibility of environment changes makes the model more compli-cated and the possibility of unforeseeable or unexpected changes is very difficult to model mathematically. However, as dynamic environments are seen as ever more important, a number of approaches that allow the agents to adapt to changes have been suggested. We will first discuss heuristic approaches (section 2.3.2.1) and then machine-learning approaches (section 2.3.2.2).