In this tutorial, we will describe the mechanisms currently used for ad auctions, and present some central issues encountered in the design of ad auctions: Choice of pricing and allocation scheme, click fraud, budget constraints, and online nature of auctions.

We will conclude with a list of open problems motivated by ad auctions.

For more Information (PDF)

This tutorial will begin with a brief review of auction basics and of the “high theory,” i.e., game theory models, of single isolated auctions. It will also give a brief overview of the contributions and approaches of different disciplines—operations research, economics, sociology, and computer science—to the studies of auctions.

The main body of the tutorial will cover the wide variety of practical issues that affect auctions, ways of analyzing these issues, and their effects. The issues considered will include: the structure of information available to the bidders, bidders’ attitude toward risk, the effects of known asymmetries among bidders, decisions by bidders about entering auctions and their effects, bidders’ concerns about secrecy, bidders’ limited access to capital, auctions stopping rules, bidders’ interests in bending and breaking the auction rules, bid preparation costs, other transaction costs, the interrelationship of multiple auctions including Lotting decisions by bid takers, the internal and external effects of bidding in sequential auctions, the effects of bidding in simultaneous independent auctions, and the many design issues involved in combinatorial auctions.

The rapid and recent advent of automation in practically all aspects of modern financial markets has created unprecedented focus on algorithms and modeling. This tutorial will examine some of the most interesting scientific challenges presented by the markets, with particular emphasis on computational finance topics that are accessible to the EC, theoretical computer science, machine learning and AI communities.

Topics to be discussed include: 1) Market Microstructure. Modeling and exploiting high-frequency intraday limit order data, including statistical and game-theoretic approaches; 2) Optimized Execution. Algorithms for executing very large, fast or otherwise challenging trades via control-theoretic and other approaches; 3) Online Portfolio Optimization. Universal portfolio and exponentiated gradient algorithms; risk, portfolio concentration, and other considerations; 4) Recent Challenges to the Efficient Market Hypothesis. "Agent-based" models; profitability of technical indicators; 5) Proprietary Trading. Developing autonomous automated trading strategies for profit: basic concepts and approaches, including momentum and mean reversion; consideration of risk measures.

Annually, the Association for Trading Agent Research hosts the Trading Agent Competition (TAC). TAC is an international forum where researchers can evaluate trading agent strategies in simulated, electronic marketplaces. Over the past seven years, TAC has attracted participants from institutions in dozens of countries around the world.

At present, there are two running versions of the Trading Agent Competition: TAC Travel and TAC SCM. In TAC Travel, travel goods---flights, hotels, and entertainment tickets---are sold in three types of electronic markets. TAC SCM stands for TAC Supply Chain Management. In TAC SCM, agents are computer manufacturers who buy components from suppliers and sell finished goods to customers.

Trading agents face a challenging task. To play the market effectively, they must make decisions in real-time in uncertain, dynamic environments. Successful agents rapidly assimilate information from multiple sources, forecast future events, optimize the allocation of their resources, anticipate strategic interactions, and learn from their experiences.

This tutorial will provide an overview of the design features of some of the most successful TAC agents. It will include an anecdotal component, highlighting some of the quirks in early TAC game designs. It will also include a research component, describing how to apply knowledge gained building TAC agents to generic trading agent design and analysis.

For more information about TAC, visit http://www.sics.se/tac