Credit_Rating_System_in_C2C_E-Commerce

Credit Rating System in C2C E-Commerce: Verification and Improvement of Existent Systems with Game Theory Youbei Huang, Mingming Wang Economic Information Management Department Renmin University of China, Beijing, 100872, China youbeihuang@yahoo.com.cn Abstract—China C2C e-Commerce has developed at high speed during recent years. Whether the credit rating systems of C2C e-Commerce is rational or not directly influences its development as the systems is a key way for risk-averse. This paper analyzes the C2C users’ fraud behavior with repeated game theory, checks the validity of existent C2C e-Commerce credit rating systems, especially the new one applied in Baidu “Youa” C2C e-Commerce platform. Finally some strategies that can be adopted to improve the systems are pointed out. Keywords- C2C market; e-Commerce; repeated game theory; credit rating systems I. INTRODUCTION China C2C online market has developed at high speed. The trade turnover of 2008 has reached 113.887 billion RMB [1] [2] with an expectation of 388.3 billion RMB in 2011[3]. Meanwhile, frauds in C2C trade occur frequently, which seriously obsess the users and the operators of C2C e- Commerce platforms. Most C2C online platforms have established various systems to reduce or avoid frauds, among which online reputation system is a significant way. Take eBay.com for example. On the platform, every member including both buyers and sellers has a feedback score to reflect their behavior during trades in general. Buyer and seller can rate each other after each trade with three kinds of feedback: positive, negative or neutral rating. These three feedbacks correspond to different scores: a positive rating increases the score by one point; a neutral rating leaves the score the same; and a negative rating decreases the score by one point. Most C2C online platforms, such as Taobao, have the same system. Therefore, the reputation system is a potential deterrent to both parties, encourages honesty, as well as restrict fraud behaviors because it has close relationship with members’ online behaviors due to its extensive reputation openness[4][5]. At the end of 2008, Baidu launched “Youa” C2C e-Commerce platform (http://youa.baidu.com) and improves the traditional credit rating systems: the turnover of each trade is included in its credit rating system. This improvement induces a hot discussion about credit rating systems on C2C e-Commerce. The foundation of designing a credit rating system is the behaviors of the both parties in C2C trade, which is a continuous process of identifying each other, gaining information and fighting fraud. The scholars both at home and broad deeply research on this problem with game theory. A number of scholars investigated the credit rating systems (a kind of reputation system) in C2C e-Commerce through various ways. Some empirical studies on this field were held, showing reputation systems have significant impact on transactions [6] [7]. In addition, trust models were built by conceptualizing, computerizing and other methods. Game Theory is introduced to this field as well. Qing Tang [8] regards one transaction in C2C online market as a static game of complete information. She elaborates the importance of honesty and information openness and transparence by analyzing a mixed strategy Nash equilibrium model. Xianfeng Zhang [9] incorporates “trade cost” in the payoff matrix. However, the analysis above overlooks that the result of existent credit rating systems: the trades on C2C online platforms are dynamic games. Haiyan Wang [10] introduces the conception of dynamic behaviors into her research and builds a trade model based on evolutionary games. Hitoshi Yamamoto [11] sets up a computer simulation model based on the theory of prisoner’s dilemma to reveal the effeteness of positive reputation system, whereas Axelrod [12] simulated repeated prisoner's dilemma by computers, and uses “the shadow of the future” as the metaphor of the evolution of cooperation in the 1980s. “The shadow of the future” shows an important apocalypse that current behaviors will influence the counterpart’s further behaviors. Specific to C2C online trade, sellers’ frauds will reflect in his/her credit record. That is it is less likely for online buyers to choose the seller who has worse reputation, namely lower feedback score, than the one who has better reputation [13]. However, few specific suggestions about improvement of the current credit rating system in C2C e- Ecommerce were put forward recently. This paper borrows the concept “the shadow of the future” to discuss the credit rating systems in C2C e- Commerce. The article is structured as follows: the first section is the introduction, including a short literature review of the researches on buyers and sellers’ behaviors in C2C online trade from the perspective of game theory. The second section describes a repeated game model of C2C online trade and indicates the key points to validate credit rating systems. The third section check the validity of existent C2C e-Commerce credit rating systems and the suggestions of improving these systems are pointed out, while the final section presents implications for further research. 2009 International Conference on Management of e-Commerce and e-Government 978-0-7695-3778-8/09 $26.00 © 2009 IEEE DOI 10.1109/ICMeCG.2009.52 36 2009 International Conference on Management of e-Commerce and e-Government 978-0-7695-3778-8/09 $26.00 © 2009 IEEE DOI 10.1109/ICMeCG.2009.52 36 2009 International Conference on Management of e-Commerce and e-Government 978-0-7695-3778-8/09 $26.00 © 2009 IEEE DOI 10.1109/ICMeCG.2009.52 36 2009 International Conference on Management of e-Commerce and e-Government 978-0-7695-3778-8/09 $26.00 © 2009 IEEE DOI 10.1109/ICMeCG.2009.52 36 2009 International Conference on Management of e-Commerce and e-Government 978-0-7695-3778-8/09 $26.00 © 2009 IEEE DOI 10.1109/ICMeCG.2009.52 36 2009 International Conference on Management of e-Commerce and e-Government 978-0-7695-3778-8/09 $26.00 © 2009 IEEE DOI 10.1109/ICMeCG.2009.52 36 2009 International Conference on Management of e-Commerce and e-Government 978-0-7695-3778-8/09 $26.00 © 2009 IEEE DOI 10.1109/ICMeCG.2009.52 36 Authorized licensed use limited to: Tsinghua University Library. Downloaded on May 19,2010 at 01:39:59 UTC from IEEE Xplore. Restrictions apply. II. A REPEATED GAME MODEL OF C2C ONLINE TRANSACTIONS In one trade, one buyer and one seller carry out the deal. Let the turnover be π. Actually, buyers pay not only the price of the good sold, but also shipping cost which is totally paid to transportation firms. That is, shipping costs have nothing to do with decision-making of sellers as they get nothing from the costs. Therefore, π is equal with the price of the good so as to simplify the model. As buyers in C2C online market are individuals who lack information seriously and are usually at a disadvantage, they have little fraud behaviors. Therefore, this paper supposes that the buyer does not defraud at all in order to focus on the key points of the model. However, the buyer has to pay for identified cost c, as a result of distinguishing whether a seller may be honest or not. If the buyer holds the viewpoint that it is almost or definitely possible that the seller will be honest, she/he will enter a trade, or the buyer will give up in spite of paying the identified cost. A seller can have two actions, honesty and fraud. In order to gain maximum yield, a seller may choose to defraud to get the “black yield” which is the difference between the turnover and the good’s true value. According to the actual situation in C2C online trades, a seller usually waits for buyers’ active inquiries and does not need to pay identified cost. Based on the analysis above, the payoff matrix for one trade in C2C online market is built as below: TABLE I. PAYOFF MATRIX FOR A REPEATED GAME Seller Be Honest Defraud Buyer Purchase (π-c, π ) ((1-θ) π-c, (1+θ) π-c) Not Purchase (-c, 0 ) (-c, 0 ) If the buyer does not purchase, he/she only pay identified cost c, while the seller pay no cost. If the buyer purchases and the seller chooses to be honest, the buyer has to pay identified cost c as well as the price of the good. Usually, a buyer and a seller have different yields. The yield of buyer is more than that of seller for one good. In the model of this paper, the difference of the yields has no influence to the question itself. Therefore, it is purposed that the yield of buyer and the yield of the seller in one trade is the same. Each one’s yield is the turnover of this trade π. If the buyer purchases but the seller chooses to defraud, namely the good sold to the buyer is fake or of bad-quality, the buyer not only pays for identified cost c and the price of the good, but also takes the loss θπ which is the difference between the price of the good and its true value. So, the final value that the buyer gets is (1-θ) π. Meanwhile, the seller gains additional “black yield” so that the final value the sell gets is (1+θ) π. In the existent C2C e-Commerce credit rating systems, the buyer will rate the seller after the trade. Thus, the next consumer of the seller will know the history of the game record. Besides, trades in C2C online market are independent of each other and have the same structure. So, the action of sellers and buyers forms a dynamic repeated game. Let the probability that a consumer buy one good is p. Actually, different consumers have different decisions to one good, therefore p in fact represents the proportion of the trade times that take places to the total number of purchase decisions of one kind of good. If the seller defrauds, the purchasing probability of the next consumer will decrease as he/she observes the seller’s fraud behaviors. Suppose that after each fraud, the purchasing probability of the next consumer changes from p to δp（0≤δ≤1）. δ is called fraud-influence coefficient. Suppose that the total number of the trades of the seller is n. The yield of the seller when he/she choose to be honest is 1R , while the yield when to defraud is 2R . Thus, ππππ nppppR =+……++=1 (1) πθδπθδπθδπ )1()1()1( 122 ++……+++++= − ppppR n ] 1 )1()1(1[ 1 δ δδθπ − − ++= −n p (2) Whether the seller defrauds or not is up to which is larger between 1R and 2R . If 1R is greater than 2R , the seller will choose to be honest because the yield of honesty is more than that of fraud. On the contrary, if 2R is greater than 1R , the seller will choose fraud as fraud brings more yield. Now, we will discuss the parameter δ, π and n: A. π and n Are Assumed to be Constants; δ Has Impact on a Seller’s Behavior When 1R > 2R , θδ δδδ + − < − − = − 1 1 1 )1()( 1 nf n . The larger the difference between )(δf and θ+ − 1 1n is, the higher opportunity cost of fraud is and the larger the possibility of being honest is. When ]10[ ，∈δ , )(δf is a monotone increasing function1. Thus, under the circumstance that π and n are assumed to be constants, the smaller δis, the smaller )(δf is, and smaller the difference between )(δf and θ+ − 1 1n is. Therefore the possibility that the seller choose to be honest is larger. 1 2 1 )1( 1)1()( δ δδ δ δ + +−− = ∂ ∂ −nn nnf Let 1)1()( 1 +−−= −nn nng δδδ 0)1)(1()( 2 <−−= ∂ ∂ − δδδ δ nng n ∴ )(δg is a monotone decreasing function And when 1=δ , )(δg =0. ∴when ]1,0[∈δ , )(δf >0, and therefore, )(δf is a monotone increasing function. 37373737373737 Authorized licensed use limited to: Tsinghua University Library. Downloaded on May 19,2010 at 01:39:59 UTC from IEEE Xplore. Restrictions apply. According to the analysis above, in order to restrain sellers’ behaviors, the builders of credit rating systems in C2C online market should lower the value of δ as much as possible, which means the systems should make the sellers who defraud pay for their behaviors as much as possible,. The seller’s fraud cost is his/her lessened future yield as the result of the decreasing possibility of buyers’ purchase. The value of the possibility is greatly influenced by how much the seller’s information the consumers have gotten, including the seller’s trading history record, credit rating and so on. If the information of seller who defrauds is easier to get, consumers’ will learn more about his/her fraud behavior so that they will reduce the possibility to purchase the goods of the seller. Henceforth, the value of δ would lower when the transparency of sellers’ information is improved and the information asymmetry is reduced. B. n is Assumed to be Constant; π and δ Have Impact on a Seller’s Behavior Suppose there is another kind of good with its turnover 'π which is larger than π. When δ = 'δ which means the fraud-influence coefficient does not change no matter how much the turnover is, '2R is larger than 2R . In other words, the larger the turnover of one good is, the larger the total yield is, and thereby it is more likely for the seller to defraud. In order to make that the turnover of one good does not has impact on the seller’ behavior, it is required that 22 ' RR = . According to Eq. (2), δ > 'δ are requested. That is, the larger the turnover of one good is, the smaller the relevant fraud-influence coefficient should be. C. π and δ are Assumed to be Constant; n Has Impact on a Seller’s Behavior The sellers in C2C online market do not need to pay rent and have no limit to the space of their shops because currently C2C e-Commerce online platforms in China are almost free and the network is virtual world. So, the sellers can keep the information of the goods in the virtual shop indefinitely with paying nothing [10]. According to Eq. (2), if the turnover and the fraud- influence coefficient are two constants, the larger the value of n is, the larger the value of 2R is. Therefore, if one seller has defrauded for many times, he/she will pay no additional cost for defrauding one more time and get more benefit. In addition, as the seller has got bad reputation, he/she do not care about his/her feedback score. So, it is of great possibility that the fraud action will happen again if a sell has defrauded for many times and pursue yield maximization. Based on the analysis above, the builders of C2C online credit rating systems should rationally set up the value of π, δ and n so as to restrict sellers' behaviors. III. THE VALITY OF EXSITING CREDIT RATING SYSTEMS The existent credit rating systems in C2C online market mainly refer to rating on the trades as well as the turnover. That is, the systems influence the sellers’ behaviors by π and δ. A. Credit Ratings and Positive Ratios: Enlarge the Impact of δ From the existent credit rating systems, buyers decide whether to purchase or not based on the information about seller’s credit rating and positive ratio. One seller’s credit rating is the number of the trades that get positive ratings minus negative rating number. Generally speaking, the longer a seller’s honestly operating history is, the larger the sales volume is, and the more the number of positive ratings he/she gets. In order to eliminate the influence of time, positive ratio namely (the number of positive ratings)/ (the total number of ratings), is involved to rate sellers’ honesty. Positive ratio reflects seller’s credit status in general. This way, even though a seller cheat only one time, his/her fraud behavior is presented in the positive ratio. Thus, it is easier for consumers to get the information of sellers as δ is reduced. B. Turnover is involved in credit ratings The price of the good sold, namely the turnover is determined by the good’s cost and its profit as well as market competition. The behavior adopted by sellers depends on whether defrauding brings more money or not. According to the analysis above, the more the turnover is, the larger the possibility of defrauding is. Henceforth, it is necessary to increase the punishment for sellers who defraud. At the end of 2008, Baidu launched “Youa” C2C online platform with the highlight that the turnover of each trade is involved in credit ratings. That is, feedback score = credit rating × the weight of the turnover. The weight varies by the turnover for each trade. For example, if the turnover is below 1 RMB, this trade has no score as its weight is 0; if the turnover is above 1 RMB and below or equal with 200 RMB, the weight is 1, and when the seller gets a negative rating, the feedback score will reduce 1; if the turnover is above 200 RMB and below or equal with 1000 RMB, the weight is 2, and the feedback score of the seller who gets negative ratings will reduce 2. Although rating is after trading and has no impact on current trade, it does influence future trades because consumers will make decisions based on seller’s trade records namely feedback score. As turnover is involved in credit ratings which means the more the turnover is ,the larger the weight is and the more serious the consequence of defrauding is, sellers will rebalance between gaining more “black yield” and losing more feedback scores or further yields. Thus, the possibility of defrauding will decrease. To this point, the new measure adopted by Baidu “Youa” is effective. IV. SUGGESTION FOR IMPROVEMENT OF CREDIT RATING SYSTEMS Although credit rating systems are improved continuously, the existent ones have not considered n (total number of trades including the past and the future) which has influences on sellers’ behaviors. According to the analysis in Section 3, if n is too large, the punishment brought by δ will provoke more serious 38383838383838 Authorized licensed use limited to: Tsinghua University Library. Downloaded on May 19,2010 at 01:39:59 UTC from IEEE Xplore. Restrictions apply. frauds instead of helping punish the fraud behaviors. Therefore, this paper suggests that the builders of the credit rating systems set limits on the value of n. Two strategies suggested are as following: • set the maximum value of n, maxn • set the minimum value of positive ratio, maxr Thereinto, positive ratio= the number of positive ratings/ the number of transactions happened= (1- the number of defrauding) / the number of transactions happened. If one seller’s number of frauds is more than maxn or the positive ratio is less than maxr , the platform should shut up his/her shop forevermore and forbid him/her to reopen a new shop. Thereby, the possibility of defrauding will decrease as sellers will consider more seriously when they deicide whether to cheat or not. However, it's important to note that that the value of maxn should not be too large. According to Eq. (2), when ∞→n , 2R converges to ]1 )1(1[ δ δθ π − + +p . That is, as n is large enough, when the seller carry out a transaction one more time, his/her total yields will hardly increase. Therefore, the seller has no motive to trade one more time and maxn has no meaning. In addition, sellers usually input costs in the initial stage of their businesses, such as the cost for the first batch of goods. If maxn is too large, it would not restrict sellers’ behaviors as they have recovered all the costs and even gain much profit. Thereby, maxn should be small enough so that it could bring impact on sellers’ decision- making. In a similar way, maxr should be large enough. However, this paper does not discuss the numerical examples of maxn and maxr as they should b