## Information Economics in the Age of e-Commerce: Models and Mechanisms for Information-Rich Markets

##### Abstract

The internet has dramatically changed the landscape of both markets and computation with the advent of electronic commerce (e-commerce). It has accelerated transactions, informed buyers, and allowed interactions to be computerized, enabling unprecedented sophistication and complexity. Since the environment consists of multiple owners of a wide variety of resources, it is a distributed problem where participants behave strategically and selfishly. This led to the birth of algorithmic game theory, whose goal is to understand equilibria arising in these strategic environments, study their computational complexity, and design mechanisms accordingly. More recently, the prevalence and access to information about consumers and products has increased dramatically and changed the landscape accordingly. In this thesis I focus on two simple yet fundamental observations which require further investigation as the field of algorithmic game theory progresses in the context of information economics. Specifically: 1) Access to information widens an agent's strategy space, and 2) the generation and exchange of information between agents is itself a game. There are three specific problems we address in this thesis. Informed Valuations: Increasingly sophisticated consumer tracking technology gives advertisers a wealth of information which they use to reach narrowly targeted consumer demographics. With targeting, advertisers are able to bid differently depending on the age, location, computer, or even browsing history of the person viewing a website. This is preferable to bidding a fixed value since they can choose to only bid on consumers who are more likely to be interested in their product. This results in an increase in revenue to the advertiser. Notice, however, that this implies information has changed the distribution of bids. With this change, common assumptions, which are reasonable in the absence of information, no longer hold. Thus, the mechanisms currently in place are no longer optimal. Using historical bidding data from a large premium advertising exchange, we show that the bidding distributions are unfavorable to the standard mechanisms. This motivates our new BIN-TAC mechanism, which is simple and effective in extracting revenue in the presence of targeting information. Bidders can "buy-it-now", or alternatively "take-a-chance" in an auction, where the top d > 1 bidders are equally likely to win. The randomized take-a-chance allocation incentivizes high valuation bidders to buy-it-now. We show that for a large class of distributions, this mechanism outperforms the second-price auction, and achieves revenue performance close to Myerson's optimal mechanism. We apply structural methods to our data to estimate counterfactual revenues, and find that our BIN-TAC mechanism improves revenue by 4.5% relative to a second-price auction with optimal reserve. Information Acquisition: A prevalent assumption in traditional mechanism design is that the buyers know their precise value for an item; however, this assumption is rarely accurate in practice. Judging the value of a good is difficult since it may depend on many unknown parameters such as the intrinsic quality of the good, the saving it will yield in the future, or the buyer's emotional state. Additionally, the estimated value for a good is not static; buyers can "deliberate", i.e. spend money or time, in order to refine their estimates by acquiring additional information. This information, however, comes at some cost, either financial, computational or emotional. It is known that when deliberative agents participate in traditional auctions, surprising and often undesirable outcomes can occur. We consider optimal dominant strategy mechanisms for one-step deliberative setting where each user can determine their exact value for a fixed cost. We show that for single-item auctions under mild assumptions these are equivalent to a sequential posted price mechanism. Additionally, we propose a new approach that allows us to leverage classical revenue-maximization results in deliberative environments. In particular, we use Myerson (1981) to construct the first non-trivial (i.e., dependent on deliberation costs) upper bound on revenue in deliberative auctions. This bound allows us to apply existing results in the classical environment to a deliberative environment; specifically, for single-parameter matroid settings, sequential posted pricing is a 2-approximation or better. Exchange Networks: Information is constantly being exchanged in many forms; i.e. communication among friends, company mergers and partnerships, and more recently, selling of user information by companies such as BlueKai. Exchange markets capture the trade of information between agents for profit, and we wish to understand how these trades are agreed upon. Understanding information markets helps us determine the power and influence structure of the network. To do this, we consider a very general network model where nodes are people or companies, and weighted edges represent profitable potential exchanges of information, or any other good. Each node is capable of finalizing an exactly one of its possible transactions; this models the situation where some form of exclusivity is involved in the transaction. This model is in fact very general, and can capture everything from targeting information exchange to the marriage market. A balanced outcome in an exchange network is an equilibrium concept that combines notions of stability and fairness. In a recent paper, Kleinberg and Tardos introduced balanced outcomes to the computer science community and provided a polynomial time algorithm to compute the set of such outcomes. Their work left open a pertinent question: are there natural, local dynamics that converge to a balanced outcome? We answer this question in the affirmative by describing such a process on general weighted graphs, and show that it converges to a balanced outcome whenever one exists. In addition, we present a new technique for analyzing the rate of convergence of local dynamics in bargaining networks. The technique reduces balancing in a bargaining network to optimal play in a random-turn game, and allows a tight polynomial bound on the rate of convergence for a nontrivial class of unweighted graphs.