A Mutual Influence Algorithm for Multiple Concurrent Negotiations Ka-man Lam and Ho-fung Leung Department of Computer Science and Engineering, The Chinese University of Hong Kong Shatin, N. T., Hong Kong, China {kmlam, lhf}@cse.cuhk.edu.hk Abstract Buyers always want to obtain goods at the lowest price. To do so, a buyer agent can have multiple concurrent negotiations with all the sellers. It is obvious that if the buyer obtains a good price from one of the sellers, the buyer should have more bargaining power in negotiating with other sellers. Then, other sellers should offer a lower price in order to make a deal. In this way, the concurrent negotiations mutually influence one another. In this paper, we present an algorithm to enable mutual influence among multiple concurrent negotiations. Introduction There are two main components in any negotiation: negotiation protocol, and negotiation strategies. One of the most widely used negotiation protocol is the alternating offers protocol (Osborne and Rubinstein 1994). In this protocol, a buyer is an agent that wants to obtain a good or service, while a seller is an agent providing the good or service. Each agent, buyer or seller, has its own reservation price. For a buyer, reservation price means the highest price at which it is willing to buy the good. For a seller, reservation price means the lowest price at which it is willing to sell the good. An agent starts the negotiation by making a proposal. In each round of negotiation, an agent can quit the negotiation, accept the latest proposal, or make a counter-proposal. To enable mutual influence, Nguyen and Jennings (2003, 2004) propose how to coordinate multiple concurrent negotiations. However, there are several problems with their model. In this paper, we introduce an algorithm to be used in the single-issue alternating offers protocol, which enables mutual influence in multiple concurrent negotiations. Most importantly, this algorithm ensures the buyer to obtain the good at a price within the lowest and the second lowest reservation price among all the sellers. A Motivating Example Consider a buyer b, which wants to buy a good. Its reservation price, RPb is $10, which means it would not buy the good at a price higher than $10. There are two sellers, s1 and s2. Their reservation prices, RPs1 and RPs2, are $7 and $5 respectively, which means the sellers would not sell the good at a price lower than $7 and $5 respectively. The buyer starts multiple concurrent negotiations with the two sellers by proposing, say, $8 for the good. Suppose sellers s1 and s2 reply with counter proposals of $9 and $8 for the good respectively. To enable mutual influence, buyer sends a message to seller s1, telling it seller s2 gives an offer of $8. Here, we assume there is no communication between the two sellers and time is not critical. Furthermore, if no agreement can be reached, then no transaction will occur, and all participating agents will effectively get zero utility. Since everyone prefers an agreement to no deal, after knowing that seller s2 offers the buyer $8, seller s1 rationally makes a counter proposal by offering a price lower than $8, say $7 to the buyer. To enable mutual influence again, the buyer sends a message to seller s2, telling it that seller s1 gives an offer of $7. Then, by knowing seller s1 offers the buyer $7, seller s2 rationally makes a counter proposal by offering a price lower than $7, say $6 to the buyer. Similarly, the buyer again sends a message to seller s1, telling it that seller s2 gives an offer of $6. However, since $6 is lower than s1’s reservation price, seller s1 cannot make a lower offer although it wants to make a deal. As seller s1 cannot make a lower offer, the buyer makes an agreement with seller s2 at a price of $6. Now suppose before the buyer makes an agreement with seller s2, there is another seller, s3, whose reservation price is $4. Then the buyer can send a message to this seller, telling it that someone offers the good at a price of $6. By mutual influence, this seller rationally makes a counter offer of $5. The buyer can again tell seller s2 that seller s3 offers $5. Then seller s2 will rationally replies with a counter offer of $5, but not a price lower than $5 as this is its reservation price. By mutual influence again, seller s3 will offer a price lower than $5, at which the buyer will make an agreement with this seller as this is the lowest price that the buyer can get from these negotiations. From this example, we can see that buyer can negotiate actively by enabling mutual influence, with which buyer can minimize the price in obtaining the good. Mutual Influence Algorithm Suppose there is a buyer, b, with reservation price RPb and there are n sellers s1 to sn, with reservation prices RPs1 to RPsn. The mutual influence algorithm can be formalized as follows. Step 1. Buyer starts negotiations by proposing a value v ≤ RPb to sellers s1 to sn, and waits for counter offers. Step 2. Suppose a counter proposal COik, from seller si, is the minimal among the received counter offers in round k. Buyer sends messages to seller sj, where j∈[1, n] and j ≠ i, announcing there is a counter offer COik from one of the sellers, and waits for counter offers. Step 3. Repeat step 2 until COjm ≥ COim–1. Step 4. Accept the counter offer COim–1 from seller si if COim–1 ≤ RPb, otherwise, announce RPb to sellers s1 to sn, and repeat step 3 to step 4. If RPb is already announced and COim–1 > RPb, then no agreement can be made. In step 1, buyer starts negotiations rationally by making an offer smaller than or equal to its reservation price. We assume, for simplicity, every seller must reply a counter offer even if the counter offer is the same as the previous one. In step 2, since everyone prefers an agreement than no deal, by knowing someone offers COik, those sellers rationally propose a counter offer greater than their respective reservation prices, but smaller than COik. Step 2 is repeated until the minimum counter offer the buyer received in round m, from seller sj, is greater than or equal to that received, from seller si, in round m – 1. This means that offer cannot be further lowered by mutual influence. So, in step 4, buyer accepts the counter offer COim–1, if this value is smaller than or equal to its reservation price. However, if this is not the case, buyer tries to lower the offer further by disclosing its reservation price. But if reservation price is already disclosed and COim–1 is still greater than its reservation price, which means that no seller can offer a price lower than or equal to the buyer’s reservation price, then no agreement can be made. This mutual influence algorithm can be applied if there is no communication between sellers. In this case, since every seller prefers an agreement to no deal, sellers are competing with each other and price can be lowered. However, if sellers are allowed to communicate with each other, they can collude not to lower the price. If this is the case, no mutual influence can be made. At the same time, the mutual influence algorithm can be applied if time is not critical. Obviously, if time is critical, e.g., buyer loses utility with time or deadline is short, there may not be enough time for buyer to benefit from mutual influence. Theorem 1 By mutual influence, the lowest price that the buyer can obtain lies within the lowest and the second lowest reservation prices of all the participating sellers. Conclusions and Future Work A good price with one seller should enable buyer to have more bargaining power in negotiating with other sellers. This paper presents a mutual influence algorithm in doing so. By simply sending messages to the sellers, telling them the minimum offer the buyer receive so far, the buyer can enable mutual influence among the sellers. On knowing someone is offering the buyer a lower price, sellers are put into competition. Since everyone prefers agreement to no deal, sellers rationally reply the buyer with a lower offer. One important contribution of this algorithm is that it can ensure the final price that the buyer can obtain lies within the lowest and the second lowest reservation prices among the sellers. Furthermore, if these two values are close, the final price will be close to the optimal, which is the lowest reservation price among the sellers. One important assumption in this algorithm is that sellers are assumed to believe all the messages from the buyers and the buyers never lies. In fact, buyer can always tell lies in messages, telling sellers that another seller is offering a low price. Then, the sellers may face the problem of having to choose whether to believe the message, and deciding what counter-offers to make. The issues about trust and honesty will be dealt with in future work. Besides, a mutual influence algorithm for multiissues can also be developed in future work. References Nguyen, T. D., and Jennings, N. R. 2003. A heuristic model for concurrent bi-lateral negotiations in incomplete information settings. In Proceedings of the Eighteenth International Joint Conference on Artificial Intelligence, 1467-1469. Nguyen, T. D., and Jennings, N. R. 2004. Coordinating multiple concurrent negotiations. In Proceedings of the Third International Conference on Autonomous Agents and Multi-Agent Systems, 1064-1071. Osborne, M. J., and Rubinstein, A. 1994. A Course in Game Theory. MIT Press.