ReputationDistributionandConsumertoConsumerOnlineAuctionMarketStructureAStudyinTaobao.com

Reputation Distribution and Consumer-to-Consumer Online Auction Market Structure: A Study in TChang Boon Patrick Lee, Lingyan Maggie ZhangFaculty of Business Administration,University of Macau, MacauEMAIL: cbleeumac.mo; maggie.ld Abstract: The purpose of this research was to analyze the sellers reputation scores in the Chinese online auction market and relate the results to the market structure. The study examined four sub-markets in Taobao namely, clothing, cosmetics and jewelry, computers and computer accessories, and food and health products. The first three sub-markets are the most popular while the last is the least popular. By using logarithmic transformations and the Wald-tests, the results obtained in this study showed that the reputation scores in the four sub-markets did not follow a lognormal distribution. Moreover, using the methodology employed by Hart and Oulten 12, the results found that there were some differences among sub-markets. The paper discussed the implications of the results.Keywords: Reputation System, Taobao, Gibrats Law, Market Structure I. IntroductionMany years ago, it was difficult for people who havent seen the goods provided by strangers living far afield to carry out business transactions with each other. Electronic-commerce (e-commerce), however, helped to overcome this problem by bringing together buyers and sellers. Although there are many fraud cases in e-commerce, they have failed to undermine the growth of the consumer-to-consumer (C2C) online auction markets. One reason is possibly the use of online reputation scoring system (or reputation system) to assist users in making transactions in the Internet 23.A reputation system is one that collects, distributes and aggregates feedback about participants past behavior 23. In this way, potential customers can see the reputation scores to assess the risk of trading with the sellers. The online reputation system has attracted researchers from various disciplines such as the behavioral sciences and economics. From the behavioral sciences perspectives, researchers have focused on the effects of online reputation on trust 3 as well as price 24. In the area of economics, there has been a series of research that investigated the relationship between online reputation and market structure. For example, Lin et al. 18 19 used reputation scores as a measure of a firms capacity and analyzed them to study the electronic market structure. Lin et al. 18 found that smaller sellers grew faster than larger sellers. Their results are not consistent with Gibrats Law of proportionate growth, which posits that a firms growth should follow a lognormal distribution and independent of firms initial size.A review of prior research has found that there is little research on the market structure of C2C auction market in China. As at the end of 2007, there were 137 million Internet users in China and 23.6 percent of these users had used C2C online shopping services 7. As the Chinese economy develops and its social status increases, it is hard to ignore Chinas e-commerce market. Therefore, this paper focused on Taobao a C2C marketplace which has more than 80 percent of the online auction market share in China 5. This study is based on a large dataset from Taobaos reputation system in three best selling categories (clothing, cosmetics and jewelry, computers and computer accessories) and one worst selling category (food and health products) 6. By using stochastic models such as the Gibrats Law and the Galton-Markov model, this paper examined the reputation scores in the sub-markets to address the following questions:1) What is the distribution of the sellers reputation scores in the four sub-markets in Taobao?2) What is the difference in the market growth for the four sub-markets? The findings gathered from this paper would provide a market structure profile of the four sub-markets in Taobao. They would also provide information about which group of sellers contributes most to the growth of each market and if there is any difference in the best and worst selling categories. This paper is organized as follows. The next section examines Taobaos reputation system, and this is followed by a review of related literature. We then go on to describe the data collection and the data analysis and results. The paper then ends with a conclusion.II. Taobaos Reputation System The name, “Taobao”, has a strong cultural meaning which is interpreted as “treasure rush”. It is founded in May 2003 under the flagship of the largest Chinese B2B e-commerce company, Alibaba. Taobao has become the most famous and biggest online C2C auction market in China today. The success of Taobao is due to many things, such as management, innovative service, and cooperation with many partners. A healthy reputation system is also the foundation that contributes to Taoboas success. The reputation system in Taobao is similar to the reputation feedback forum in eBay. As show in Figure 1, the reputation system provides positive, neutral, negative and total scores. Figure 1: The Reputation System of a Store in Taobao(Source: )Taobaos reputation system is different from eBays in three ways. First, Taobaos reputation system is divided into two parts, one for the seller and the other for the buyer. Second, there are scores for one week, one month, six months, and total. Third, Taobao uses 15 symbols to represent total scores. Figure 2 shows the symbols. Using these symbols, a buyer can distinguish the reputation of the sellers quickly.Figure 2: The Rating of Reputation Scores(Source: )III. Literature Review Market structure and Gibrats law An important literature related to market structure started with Robert Gibrats Inégalités Économiques 11. Gibrat suggested that the expected value of the increment of a firms size in each period is proportional to the current size of the firm 22. Denoting the size of the firm at time t by St and the proportional growth between period (t-1) and period t by a random variable t, so that t = (St - St-1)/ St-1 (1)then St = St-1(1+t) = S0 (1+0)(1+1)(1+t) (2) Taking logs and using approximation that log(1+t) t, then log St = log S0 + 0 +1 +t (3)When t, log S0 become very small compared to log St. By assuming the increment with mean and variance 2, log St can be approximated by a normal distribution with mean t and variance 2t 26. Gibrats Law or the Law of Proportional Effect pointed out that no matter the firms initial size, its growth rate follows a lognormal distribution. Gibrats Law initiated economic research in growth pattern with regard to firm size and revenue distributions for more than 70 years. Sutton 27 summarized the “legacy” research developed since Gibrats work. This area of research, called “Growth-of-Firms” includes research related to verification of proportionate firm size growth 17, stochastic growth model 16 and multi-sectional growth rate of firms among industries 26.Although Gibrats Law provided a basic model for investigating market growth, studies have shown that it is not satisfactory in explaining the evidence from empirical data. Mansfield 21 collected data in three American industries in different time periods and concluded that Gibrats Law does not satisfy different situations. Evans 8 found that firms growth was not independent of firms size smaller firms grew faster than their larger counterparts. Hart and Oulton 16 found that for larger firms, there was no relationship between growth and size. According to them, the growth of the firm was affected by large stochastic shocks “which will outweigh the systematic forces in so many cases that the resulting skew size distribution of firms by output will appear to be generated by a multiplicative stochastic process”. Their work extended Gibrats Law by using new enhancements in stochastic models.Reputation as an indicator of online business capacityThere are many different economic signals about a conventional firms capacity. Previous research has used firm size, revenue, value added, payroll and new capital expenditure to measure firm capacity and had consistent but not identical conclusions regarding growth of a firm 28. For virtual firms in an auction environment, it appears that there is not much information available regarding the firms. For example, there are many C2C websites that remain anonymous. Oftentimes, virtual sellers can register in different names and they buy their own products to increase the reputation scores. To overcome problems related to authenticity of the firms, many C2C websites use third-party authentication service to assure the information and minimize deceit at the auction market. The adoption of online reputation scores implemented by a trusted third party can help to reduce transaction risk 2. This service helps to increase the validity of the data in the reputation system and is a trend observed in the marketplace.Reputation can be regarded as the impression and assessment of a social entitys esteem or desirability 10. Bromley 4 pointed out that a social entity builds its reputation based on past behaviors. From the macro level, reputation has been recognized as an indicator of social stratification 25 and industrial stratification 9. Reputation scores, therefore, are perhaps the only publicly accessible measure that can be used as a proxy for the business capacity of a firm 19. This paper, therefore, used reputation scores as the measure of an online firms capacity. Reputation scores are also used here to identify the evolution of the general market structure.Market structure research in online auction marketLin et al. 18 were among the first to study the relationship between online reputation distribution system and market structure. They conducted a research based on data from eB and found that the reputation scores approximate a geometric distribution. Lin et al. 22 then conducted research on the structure and dynamics of electronic markets. The rationale of their method is that reputation can be regarded as the impression and assessment of a social entitys esteem or desirability. They use reputation scores as the indicator of online business capacity, assuming that each individual trader, a seller in particular, can be treated as a business unit or a virtual firm. An investigation of the transition pattern of trader reputation scores in the C2C online auction market may well represent the evolution of the general market structure. By using the Gibrats Law and Central Limit Theorem, their proposition leads to the result that the size of a firm is lognormally distributed for sellers. For buyers, the net reputation data fail to pass the lognormality test, while the six-month total reputation feedback is approximately lognormal. Combined with the theory of entry and exit, they explained why the whole reputation system does not follow the Gibrats Law of proportionate growth but the sellers reputation distribution is still lognormal. They found that the net population of traders in eB is growing and the total transaction volume is increasing. Moreover eBays C2C market is becoming a B2C market, although C2C was an important and main business activity in eB.Lin and Li 20 had also conducted a comparative study between Taobao and eBay. The comparison found that in Taobao, (a) the distribution of total reputation scores is lognormal, which is the same as ebay, (b) the negative feedback rates (NFR) and neutral feedback rates (NEU) are also lognormal, and (c) the overall neutral feedback rate is higher than that in ebay. They found the structural change of Chinas online auction market in Taobao is similar to eBay. In addition, they had some new ideas such as they used the NFR as the best indicator of risks. They found that neutral feedback rates of sellers in both Taobao and ebay are lognormally distributed, while sellers in Taobao have a higher overall neutral rate. The results showed that Taobaos reputation system can effectively provide information to buyers on the risks of their transactions with a seller.IV. Data CollectionAccording to the C2C research published by CNNIC 6, the most popular product categories in Taobao are clothing, cosmetics and jewelry, and computers and computer accessories. The least popular category is food and health products. Rather than studying the reputation system for the whole market, this research focused on the reputation system in these four sub-markets. Following the data collection procedures reported in Lin et al. 19, the data for this study were collected from the Taobao website in two periods, in January 2007 and March 2007. For the first period, data for the trader IDs and reputation data were collected. The information of trader IDs consists of registered name and opening date of store. The reputation data included positive, neutral and negative reputation scores in recent six months and in total. The study used simple random sampling to collect data for each sub-market. Note that we used the equation, n=p*(1-p*)(z/2/E)2 1 to determine the appropriate sample size. With p*=0.50, E=3% margin of error, and z/2 =1.96, the sample size, n = 1067. To ensure sufficient numbers, we collected 1500 samples.In the second round of data collection in March 2007, 2500 samples of data, including those contained in the previous 1500 samples were collected. The different sizes of datasets enabled the verification of the consistency of data analysis outcomes at two different time period. The datasets collected at the first and second rounds were labeled 070121 and 070315 respectively. For convenience, the numbers 1 to 4 were used to represent the different sub-markets. Category 1 is for clothing, category 2 for cosmetics and jewelry, category 3 for computer and computer accessories, and category 4 for food and health products. V. Data Analysis and ResultsDescription of data Table 1 shows the descriptive data for the samples collected for this study. For the clothing industry, most sellers had business in the past six months which is about 94.5% and 91.0%, and those who had never sold a single good is very small which is less than 0.2% and 0.3%. The table shows that the tendency of the other three industries is the same as the clothing industry. Table 1: Descriptions of reputation datasetsInitial Collection DateSample sizeTraders whose total reputation scores>0Traders whose six-month total>0123401/21/200701/21/200701/21/200701/21/200711981190150011291196(99.8%)1187(99.7%)1403(93.5%)1116(98.8%)1132(94.5%)1085(91.2%)1261(84.1%)1031(91.3%)123403/15/200703/15/200703/15/200703/15/200723422304230023252336(99.7%)2259(98.0%)2097(91.2%)2311(99.4%)2132(91.0%)2012(87.3%)1860(80.9%)2154(92.6%)Histograms of reputation distribution Figures 3 to 6 shows the histograms of logarithm-transformed reputation data for this research. In the histograms, Y=Ln(X), where X is the original reputation data and Y is the transformed reputation data. Figure 3: Histograms of clothing market(a1) Total reputation score (Jan 07) (a2) 6-month reputation score (Jan 07)(a3) Total reputation score (Mar 07) (a4) 6-month reputation score (Mar 07)Figure 4: Histograms of cosmetics and jewelry market (b1) Total reputation score (Jan 07) (b2) 6-month reputation score (Jan 07)(b3) Total reputation scores (Mar 07) (b4) 6-month reputation scores (Mar 07)Figure 5: Histograms of computers and computer accessories market(c1) Total reputation scores (Jan 07) (c2) 6-month reputation scores (Jan 07)(c3) Total reputation scores (Mar07) (c4) 6-month reputation scores (Mar07)Figure 6: Histograms of food and health products market(d1) Total reputation scores (Jan 07) (d2) 6-month reputation scores (Jan 07) (d3) Total reputation scores (Mar07)(d4) 6-month reputation scores (Mar 07)Several findings emerged from the histograms. First, for all the four sub-markets, the distributions of the reputation scores deviate from the bell-shaped normal distributions shown with a solid line. Second, for the six-month period, all the four sub-markets have a large number of sellers who had original reputation scores of 1, and log-transformed reputation score of 0. Third, for the total reputation scores, the number of sellers with original reputation scores of 1 is much higher in computer and computer accessories market for both time periods. Moreover, the number of sellers with original reputation scores of 1 had a great increase in March 2007 in cosmetics and jewelry industry.Wald testThe Wald test was conducted on the dataset collected at two different time points using the formula,Wald-stat=degree_of_freedom*(skewness2/6+kurtosis2/24) where the critical value 2 (d=2, =0.05) = 5.99, and the degree of freedom is the sample size of the dataset 5. If the Wald-stat is less or equal than 5.99, then the distribution of reputation system is lognormal. Table 2 shows that for the six-month total reputation score of the clothing industry collected at January 2007, the Wald-stat is 2.479 which is less than 5.99. However, all the other 15 samples failed to pass the Wald test. This means that the distributions of the reputation systems are mostly not lognormal. Table 2: Descriptions of reputation scoresDatasetType of ScoreSample SizeMean WaldStat070121-1Total Scores6-month Total119611324.4473.42625.2172.479070121-2Total Scores6-month Total118710853.9563.16962.99231.824070121-3Total Scores6-month Total140312613.2452.68524.09528.395070121-4Total Scores6- month Total111610313.5653.08953.00132.315070315-1Total Scores6-month Total233621323.9543.006126.75844.691070315-2Total Scores6-month Total225920123.7412.96820.98352.780070315-3Total Scores6-month Total209718603.4632.82324.80640.105070315-4Total Scores6-month Total231121544.1923.54716.15118.918The following, therefore, are the results from both the graphical and lognormal tests. Seller reputation data, including the total scores and six-month total reputation scores are not lognormally distributed. The results mean that the four sub-markets in T did not have proportionate growth. One reason maybe the limitation of Gibrats Law. Previous research had indicated that Gibrats Law did not work consistently in different situations and may be interpreted in several ways, depending on how the data were analyzed 21. The other reason is the booming industry in China. As the great development of T in recent years, it is not surprising to see there are many new sellers registered in these four sub-markets and the volume of successful transactions is increasing. Consequently, the growth of industry is greater than proportionate. Third, there is a traditional Chinese festival in the spring festival in February and people will have holidays and buy gifts and other things for celebration. Hence, the number of transactions might have increased substantially. The growth of each categoryMuch research has revealed the limitation of Gibrats Law. Among them, the work by Hart and Oulton 12 showed the growth pattern of firms using stochastic models. To determine differences among the sub-markets, this