Benchmarking Competition for Trading Agents

Found this linking from Vorobeychik, a student of the current sigecom chair, M. Wellman. He wrote a *great* survey 5-pager at http://www.eecs.umich.edu/~yvorobey/2002/YABackground.pdf

PhD Graduates

(Chronological order)

name (defense date) [current affiliation]

thesis title

John Cheng (Jan 1998) [Capital Networks]

Essays on Designing Economic Mechanisms
co-advised with Carl Simon

Chao-Lin Liu (May 1998) [Taiwan Nat’l Chengchi University]

State-Space Abstraction Methods for Approximate Evaluation of Bayesian Networks

Tracy Mullen (Jan 1999) [Pennsylvania State University]

The Design of Computational Markets for Network Information Services

David Pynadath (Jan 1999) [USC Information Sciences Institute]

Probabilistic Grammars for Plan Recognition

Junling Hu (Jun 1999) [Talkai, Inc.]

Learning in Dynamic Noncooperative Multiagent Systems

Peter Wurman (Jul 1999) [North Carolina State University]

Market Structure and Multidimensional Auction Design for Computational Economies

David Pennock (Sep 1999) [Overture]

Aggregating Probabilistic Beliefs: Market Mechanisms and Graphical Representations

William Walsh (May 2001) [IBM Research]

Market Protocols for Decentralized Supply Chain Formation

Trading Agent Competition – TAC Classic – Game Description

In the TAC shopping game, each “agent” (an entrant to the competition) is a travel agent, with the goal of assembling travel packages (from TACtown to Tampa, during a notional 5-day period). Each agent is acting on behalf of eight clients, who express their preferences for various aspects of the trip. The objective of the travel agent is to maximize the total satisfaction of its clients (the sum of the client utilities).

Travel packages consist of the following:
A round-trip flight,

A hotel reservation, and

Tickets to some of the following entertainment events

Alligator wrestling

Amusement park

Museum

Illustration of the environment a TAC agent operates within. To the left are its eight clients and their preferences, in the middle all its competitors lined up (7 competitors/game), and to its right are all the auctions (28 simultaneous auctions of three different types).

There are obvious interdependencies, as the traveler needs a hotel for every night between arrival and departure of the flight, and can attend entertainment events only during that interval. In addition, the clients have individual preferences over which days they are in Tampa, the type of hotel, and which entertainment they want. All three types of goods (flights, hotels, entertainment) are traded in separate markets with different rules.

A run of the game is called an instance. Several instances of the game are played during each round of the competition in order to evaluate each agent’s average performance and to smooth the variations in client preferences.