Critique of TAC/AA Games - Literature review Example

? TCA/AA Games TAC/AA Games Auctions are popular, distributed as well as independent safeguarding methods of assigning things among many agents (Sandholm 2000). They are fairly proficient in terms of outcome and process. The widespread use of the internet has caused the internet auctions (e-auctions) to become very popular so as to trade the increased number of items. The internet offers the perfect infrastructure and market information for carrying out auctions at low administrative costs. Internet advertising offers a considerable source of income for online publishers and this amount to billions of dollar yearly. Auctioning attracts much interest and has been employed in various market places. The mechanisms and approaches used by auction agents play an important role in trading transactions (Parsons, Rodriguez-A, & Klein 2011). The TAC Ad Auction (TAC/AA) game is a sponsored search scenario with features that are not common in more stylized environments. The scenario utilizes a usual ad auction method and a simple but structured model of a population of search users in a replicated retail market. Developing advertiser mechanisms for ad selection and bidding in this area is normally a challenging issue, which cannot be resolved by any known analytical solutions. In a TAC shopping game, every agent (a competition entrant) is considered a travel agent with the aim of assembling travel packages (TAC, 2011). Every agent normally acts for eight clients who have communicated their inclinations for numerous features of the trip. The travel agent aim is to make the most of the total satisfaction of the clients. In TAC, the travel packages consists of a hotel reservation, round-trip flight, tickets for museum, amusement park, and alligator wrestling (TAC 2011). One of the approaches for auction agents is the traditional auction format where an agent auctions items separately. The bidder is usually required to evaluate the other products obtained in other auctions (Bradshaw 1997). This approach necessitates a nonflexible look before the series of auctions. There is usually an enduring indecision because of partial news on the other bids by the partakers. The traditional auction format ignores important features relevant to the online consumer auctions (Binmore 1992). It offers bidders the chance to bid on a product or to purchase it immediately. The Buy It Now option in the traditional auction approach allows the seller to set the price he or she is willing to sell for and the bidders either place a bid for less than the set amount(however at or more than the starting price) or win the auction immediately by paying the set price. When the bidder accepts to pay the set price, the auction closes. When someone places a bid below the set price, the option disappears (Entrepreneur Press & Lynn 2007). The second approach is the combinatorial auction. This approach can be used to surmount the problem of uncertainty in the traditional auction approach (Sandholm & Suri 2003). The combinatorial auction is known as the bundle auction or multi-item auction. It denotes an auction that involves many products simultaneously. This approach keeps the bidders from the possibility of acquiring only some of the combinations that the wish for. These auctions are not popular because they offer the bidders more significant power through the substitute or complementary bids. Pekec & Rothkopf (2003) termed combinatorial auction as any auction that sell many items at the same time and allows nothing or all propositions on combinations of these items. There are numerous combinatorial auctions applications in the real world; airport landing slots auctions, scheduling and shipping among others. Trading in electronic markets has progressively more become a usual economic activity and an issue is special interest in electronic commerce, AI as well as Multi-agent Systems (MAS) research communities. TAC Ad auctions (AA) agents represent internet advertisers’ offer for search-engine ad placement in an array of interconnected keyword groupings. A back-end search-user model interprets position over every replicated day to impressions, clicks as well as conversion of sales bringing about returns for the person advertising. Advertising methods that combine bidding tactics and online data analysis compete to maximize returns over the simulated campaign horizons (Binmore 2008). On the other hand CAT or reverse TAC also known as TAC market design are software mediators that characterize market makers whose objectives are to draw prospective sellers as well as buyers as clients and to match the sellers with buyers. The makers of the market contend with each other in drawing and matching buyers and sellers for the business of the stock agents (Binmore 2008). The TAC supply chain management creates an active supply chain setting where the agents contend to get hold of consumers’ orders and constituents needed for production of orders. The games capture most of the intricacies of the usual supply chains as both supply as well as demand vary and every company has restricted manufacture capability. Partakers have the choice of registering two supply chain management challenge events: a prediction as well as a procurement challenge. The market games are particularly devised to offer agents hard choice setbacks and present an array of negotiation approaches and impending bidding (Binmore 2008). The TAC/AA represents a reasonable sponsored search environment for a virtual advertising situation. The advertisers act on behalf of the retailers in a basic home entertainment market and they bid to place adverts before the users searching using product keywords. The contestants of the game devise and put into practice bidding approaches for advertisers (Zeff & Aronson 1999). The search engine and users behaviors are simulated by the server. TAC/AA scenario creates a reasonable simulator in which participants can come up with tactics that could apply to real sponsored search auctions. Brendan Kitts, the developer of the Pay Per Click Bidding Agent Competition 2, assured realism by developing software for actual-sponsored search interfaces. The TAC/AA situation is set to incorporate most of the remarkable strategic features of search auctions that are sponsored while being repeatable as well as computationally agreeable to pragmatic analysis. Non-linearity in value per click by the advertiser gives an explanation for the interdependence of keywords. The linearity of values is dependent on the way clicks and queries are made by users. Advertisers who represent retailers’ of home entertainment products struggle for ad replacement across numerous interrelated keywords. A game instance signifies a simulated ad campaign with a constant number D of bidding times referred to as days. Every day, and for every keyword, the N advertisers chosen between generic and targeted ads, and settle on the amount to bid for placing an ad. According to Lahaie & Pennock (2007), the search publishers place ads, collect bids and charge advertisers depending on the family of ranking algorithms. Queries are generated by the changing population of search users. The search users normally examine the search results and decide the action to take such as clicking on ads or buying products from the advertisers according to their predilections (Mcafee & McMillan 1987). The user purchases generate sales profit for the advertisers where the user clicks generate revenues for the publisher. Initializing information is received by the advertisers’ agents when they connect to the game server while starting a game instance. The publisher is stimulated by the user population as well as the publisher. By the end of the D-day campaign, agents are assessed depending on their collective surplus: sales profit-advertising costs. The TAC/AA scenario had three kinds of agents: publishers, users and advertisers. The publisher and search users follow preset policies that incorporated in the game environment. The advertiser agents, save for dummy agents, offered for testing normally follow policies set by competition entrants. The interface among the types of agents can be summed up as follows: user, search engine queries, advertiser, agent action, purchases products, clicks on ads, bids for ad placement, publisher, processes user clicks and queries, chooses ads for display, runs auction for every user query, receives analytics reports, queries search engines and delivers daily query reports to the advertisers . The behavior of advertisers, users and publisher interact to yield search advertising events during every simulated day. Every interaction sequence follows a user as it clicks, queries and purchases an item. Every advertising agent chooses a particular ad to represent a given query. The bid for the query class is also set by the publishers. The ads and bids are used by the publisher to establish the placement of the slot through the use of an ad auction. When a query is submitted by a user, the auctioneer runs the auction and the slot replacement results are returned to the user in the form of an impression which is a ranked list of ads (Lahaie 2006). The user examines the impression and evaluates to click on every ad or not to. When the user clicks on a particular ad, he or she is taken to the respective landing page of the advertiser. Additionally, the advertiser pays a cost per every click that was set by the publisher when the ad auction was run. Each time a customer clicks on an advertiser’s ad, he or she resolves whether or not to buy an item from that particular advertiser. In case the user makes a purchase, the event is referred to as a conversion and the advertiser makes a profit. The same process is repeated for every user throughout every simulation day. The chain of activities that surround a query are: a bid on the keyword is made by the advertisers, the ads are ranked by the publisher, the user clicks on an ad, he or she views the landing page and converts its interest to a sale. Electronic commerce has continued to play a progressively more significant function in most businesses as it provides chances for improving the dealings among the suppliers as well as the customers. There have been numerous successful auctions applications in electronic business, for instance 3G mobile-phone license, electricity markets as well as commodity trading. The different kinds of auctions are effective mechanisms for allocating resources. Judged against the other auctions styles, continuous double auctions (CDAs) are complex and usually create competitive products easily (Phelps, Parsons, & McBurney 2004). Both the buyers as well as the sellers are permitted to present their ‘asks’ or ‘bids’ at any particular time when the trading is underway. The CDA is a method of matching buyers and sellers of a particular product and to ascertain the prices at which trades are carried out. The traders have the liberty of placing limit orders in form of bids that is buying orders and asks which is selling orders. An order book if maintained for outstanding orders. Each and every trader can at any time place a market order for buying or selling straight away at the market price, this is determined by the set of orders in the order book (Phelps Parsons, & McBurney 2004). Buyers and seller execute trades when a new limit order comes in and the highest ‘bid’ is equal to or more than the highest ‘ask’ price. Trade is also executed when the order book has orders with corresponds to the market order. Software agents refer to programs which are independent, personalized, long term and interactive (Nicolaisen Petrov, & Tesfatsion 2001). The agents thus take action on behalf of the auctioneers, seller or buyers to attain some auction goals. An agent in continuous double auction can be referred to as a user’s delegate to attain a good performance which normally implies good returns. In CDA markets, buyers or sellers are usually allowed to leave or enter the market without restraint. Thus the buyers or sellers normally vary every time. CDA markets are thus vibrant. CDA are normally associated with traders’ portfolios and cash accounts. Once trade is executed, cash and goods can be exchanged in trader accounts as laid out in the trade. If a trader‘s account has inadequate goods or funds, the trade is considered infeasible and the offending orders are disregarded. In CDA, for every piece of product sold, there exists a withheld price that is only familiar to him (Toft & Bagnall 2004). The other bidding strategy used by auction agents is Gjerstad-Dickhaut auction strategy which was provided by Gjerstad and Dickhaut (Ma & Leung 2008). This strategy is founded on belief functions, which point toward the likelihood of certain bid being agreed to. This is attained by studying the earlier recorded market data –that is, the presented ‘asks’ as well as ‘bids’ rate of recurrence and the rate of recurrence of asks and bids that result into a transaction. Given that current asks as well as bids are of more significance that the old ones, introduced by the authors have a sliding window role in the agent’s memory, which only considers the recent L shouts in the market history (Ma & Leung 2008). Gjerstad –Dickhaut sustain that functions are founded on the assumption that, if one ‘ask’ is accepted, all ‘asks’ will also be accepted but at a lower price however if an ‘ask’ is rejected, all asks, which have a more expenive price will also be rejected. Likewise, if a ‘bid’ is accepted, each and every ‘bid’ at an increased price will similarly be agreed. If a ‘bid’ is rejected, every ‘bid’ at a reduced price will similarly be rejected (Ma & Leung 2008). The Roth-Erev (RE) auction strategy mimics the people actions in games with mixed approach equilbria (3). Backup learning algorithms are used on the profit margins of the agents so as to adjust them to the prevailing market situation. This approach is dependent of the auction mechanism itself because it relies on the direct feedback of the agents with the market mechanisms. RE strategy does not require any data like ZIP and GD, however it relies on the interaction of the same agent with the market mechanism. Both ZIP and GD necessitate the trading agents to reach to the ‘bids’ and ‘asks’ history and all the accepted business transactions as well as their prices. The advantage of RE is that it is adequate to be used in all auction environments. This strategy is founded on prejudiced reinforcement –learning in which the agents raise their predisposition to bid at a certain mark up depending on the profit earned in the earlier round (Nicolaisen, Petrov, & Tesfatsion 2001). Zero Intelligent Plus (ZIP), an artificial trading agent employs simple machine training to adapt to function as sellers or buyers in an online open outcry auction market environment. Technicians optimize the eight factors functioning as governors of the ZIP trader actions using a standard genetic algorithm. This is an advantage because previously it was set manually. This algorithm has been basically employed for auctioning agents with many buyers as well as sellers (Neumann 2011). The auction market parameters setting are optimized by applying the genetic algorithm. ZIP demonstrated that the genetic algorithm can do better in evolving the bidding agents’ parameters and also strategies. The core aspect of ZIP agents is that they train to a bidding method which is solely based on information offered by the results of earlier shouts or auctions and their private price. The disadvantage of ZIP is that agents are usually outperformed if they fail to integrate time as an important aspect; in spite of its high effectiveness, ZIP agents timing strategy is not usually optimal. The agents are at a risk of accepting sub-optimal deals and also trading too early (Neumann 2011). The advantage of ZIP is that agents normally operate without any memory and with a limited market information and they manage to attain a high value of allocative effectiveness when put solely in a market discipline (Neumann 2011). Zip agents normally have random very low profit margins at the start of a trading day. The profit margin begins to increase when a transaction showing that they could receive a unit at an enhanced process takes place. Nevertheless, the agents always have a backup plan that protects them from setting their profit margin on a higher scale. When a shout of a buyer is rejected, the shout price is increased. When a shout of a seller if rejected, the shout price decreases. Likewise, the agents look out as transactions are performed by competing sellers and they decrease their profit margins if required to do so, in order to avoid being undercut by competing buyers or sellers (Reeves, MacKie-Mason, & Osepayshvili 2005). The agents are usually profit oriented and they are thus enhanced with a refined timing strategy to investigate the agents’ performance. Time is used as a strategic element in Zero-Intelligence –Plus. The traders are not prudent; they don’t remember past activity in the market and can as a result not learn from experience. They also do not attempt to maximize any profit. Nonetheless their timing approach necessitates them to place their orders regularly. The truth telling strategy is an auction strategy whereby agents provide offers that are equate their valuation for the resources being traded. This strategy is dominant in a strategy proof market. This strategy is dominant in a second price auction (Meersman & Tari 2008). The CAT game is commonly based in the terms of the Trading Agent Competition or TAC. There is no fixing of the markets in this game; however they are creates by every competing agent. It is thus referred to as ‘CAT’ or ‘reverse TAC’. In the CAT game, various trading agents are produced by the game itself whereas the aim of the contestants is to devise specialists’ agents. Every specialist agent operates in a single market and also lays down its rules. The trading agents are allowed to select just one market in every operating day in addition to the idea that they can only sell or purchase in the market they pick. The agents in the market are either purchasers or sellers for the whole game (Iwasaki & Yokoo 2004). They purchase or sell identical products in one auctioning unit. However, the agents are allowed to trade numerous products every day, placing a new ask or bid after a concluded transaction. On-line auctions, particularly the ones controlled by independent agents have been expansively used in the real world. The double auction strategies mostly applied in the real world is a generalization of the more frequently popular single sided auctions, like the English ascending auction, which comprises of one seller trading with many buyers. In double action, many traders are allowed on both sides of the market. The agents solicits offers to buy an item from buyers, to be exact bids, they also solicit offers to sell an item from sellers also known as asks. Double action variants are mostly employed in most world market places like stock exchanges in situations where demand as well as supply is extremely self-motivated. References Binmore, K 2008, Game Theory: A Very Short Introduction, UK, Oxford University Press. Binmore, K 1992, Fun and Games: A Text on Game Theory, D.C. Lexington, MA, Health and Company. Bradshaw, J 1997, Software Agents, Cambridge , MA, The MIT Press. Entrepreneur Press & Lynn, J2007, Start Your Own Business: Business on eBay, California, Entrepreneur Press Iwasaki, A & Yokoo, M 2004, Stability of the Truth-telling Strategy in Multi-unit Option Allocation Auctions, Laboratory Experimentation. Klemperer, P 1999, Auction Theory, Journal of Economic Surveys. 13, 227–286 Lahaie, S 2006, An Analysis of Alternative Slot Auction Designs for Sponsored Search, In 7th ACM Conference on Electronic Commerce (EC’06), Ann Arbor, Michigan, USA. Lahaie, S & Pennock, D 2007, Revenue Analysis of a Family of Ranking Rule for Keywords Auctions, In ACM Conference on Electronic Commerce (EC’07), San Diego, California, USA. Ma, H & Leung, H 2008, Bidding Strategies in Agent-based Continuous Double Actions, Switzerland, Birkhauser Basel Mcafee, P & McMillan, J 1987, Auctions and Bidding, Journal of Economic Literatire, 25(2), 699-738 Meersman, R & Tari, Z 2008, On the Move to Meaningful Internet Systems: OTM 2008 Confederated, Springer. Neumann, M 2011, Price Information of Continuous Double Auction Agents using Time as a Strategic Element, Retrieved from http://michaelkaisers.com/publications/2011_BT_MNeumann.pdf Nicolaisen, J., Petrov, V & Tesfatsion, L 2001, Market power and efficiency in a computational electricity market with discriminatory double-auction pricing, IEEE Transactions on Evolutionary Computation, 5(5), 504–523 Parsons, S., Rodriguez-A, Juan, A & Klein, M 2011, Auctions and bidding: A guide for computer scientists, ACM Computing Survices, 43, 10:1–10:59. Phelps, S., Parsons, S & McBurney, P 2004, An evolutionary game-theoretic comparison of two double auction market designs, In Agent-Mediated Electronic Commerce VI. Springer-Verlag. Reeves,D., MacKie-Mason, J. W & Osepayshvili, A 2005, Exploring bidding Strategies for market-based Scheduling, Decision Support Systems. Sandholm, T2000, Approaches to Winner Determination in Combinatorial Auctions, Decision Support Systems, 28(1), 165-176. TAC/Ad Auctions 2011, Retrieved from http://aa.tradingagents.org/?page_id=34 Toft, I & Bagnall, A 2004, Adaptive Agents for Sealed Bid Auctions, Technical Report CMP- C04-03. Zeff, R & Aronson, B 1999, Aronson Advertising on the Internet, New York, John Wiley & Sons. Read More