Measuring Bidder Loyalty in Online Auctions with a Bidder-Seller Network
https://arxiv.org/pdf/1010.1636.pdf
In this paper, researchers Wolfgang Jank and Inbal Yahav examine the idea of buyer-seller loyalty in online auctions. For sellers, loyal customers are a huge benefit to the business and in online auctions, it is even more important because buyers can simply switch to a different seller with a simple click. Jank and Yahav use a data set of 30,000 auctions for one item and create a bipartite graph of bidder and seller nodes. The nodes are linked when the bidder purchases an item from a seller node. The strength of the tie increases if the bidder buys more from that same seller. From this graph, they derive loyalty distributions and discuss the effect of loyalty on the result of an auction. Although it sounds simple, problems arise in analyzing bidder-seller networks when a few sellers dominate the market (which happens in most cases). This creates mega-clusters which are hard to deal with from a statistical standpoint as they violate standard models such as the ordinary least squares model. Attached underneath is an example of a bidder-seller network showing the mega-sellers only. The bidders are the red dots while the sellers are the white triangles. We can see there are many isolated clusters as shown on the outskirts. However, there are also a decent amount of shared bidders as well, as shown in the center between the sellers.
Regarding loyalty, for each seller there are values that describe both the proportion of bidders loyal to that seller and the degree of loyalty of each bidder. And for each seller, a distribution of loyalty with frequency is produced, which categorizes the popularity of the seller in the bidder-seller network. Through this distribution, the researchers find relationships between bidder loyalty and the seller’s price. They found that loyalty benefits the seller only up to a certain point. Below this threshold, prices tend to be competitive with a few loyal buyers. However, with many loyal buyers, competitive prices go down because the buyers already know the seller well enough. To me this was a very interesting conclusion because I assumed that loyalty would always be better for the sellers. I enjoyed reading this paper because of its direct connection to our lectures and my personal interest in data analysis. Similar to our class, the paper looks at clusters (in this case with mega-sellers) and discusses their importance on the network. I also found myself thinking about structural balance in this case. Would triads be unbalanced since if two bidders are loyal to a seller, then the bidders may compete against each other for the same seller? Moreover, in terms of STCP, two bidders don’t need to know each other to compete for the same seller online. All in all, this paper was a dense but worthy read into insights on online auctions.