Identifying and Forecasting Economic Regimes in TAC SCM - Agent-Mediated Electronic Commerce - page 113

Identifying and Forecasting Economic Regimes
in TAC SCM

Wolfgang Ketter, John Collins, Maria Gini, Alok Gupta , and Paul Schrater

Department of Computer Science and Engineering
Department of Information and Decision Sciences
University of Minnesota, Minneapolis, MN 55455, USA
{ ketter, jcollins, gini, schrater } @cs.umn.edu,
[e-mail]

Abstract. We present methods for an autonomous agent to identify dominant
market conditions, such as over-supply or scarcity, and to forecast market changes.
We show that market conditions can be characterized by distinguishable statisti-
cal patterns that can be learned from historic data and used, together with real-
time observable information, to identify the current market regime and to forecast
market changes. We use a Gaussian Mixture Model to represent the probabilities
of market prices and, by clustering these probabilities, we identify different eco-
nomic regimes. We show that the regimes so identified have properties that corre-
late with market factors that are not directly observable. We then present methods
to predict regime changes. We validate our methods by presenting experimental
results obtained with data from the Trading Agent Competition for Supply Chain
Management.

1 Introduction

In the Trading Agent Competition for Supply-Chain Management [1] (TAC SCM), six
autonomous agents attempt to maximize profit by selling personal computers they as-
semble from parts, which they must buy from suppliers. The agent with the highest bank
balance at the end of the game wins. Availability of parts and demand for computers
varies randomly through the game and across three market segments (low, medium, and
high computer prices). The market segments are affected not only by the random vari-
ations in supply and demand, but also by the actions of other agents. The small number
of agents and their ability to adapt and to change strategy during the game makes the
game highly dynamic and uncertain.
During the competition, an agent has to make many operational and strategic deci-
sions, ranging from how many parts to buy, to when to get the parts delivered, how to
schedule its factory production, what types of computers to build, when to sell them,
and at what price.
The problem we address in this paper is how an agent can detect and exploit market
conditions, such as oversupply or scarcity of products. We show that market conditions
can be characterized by distinguishable statistical patterns, that we call regimes ,we
show how such patterns can be learned off-line from historical data, how they can be
identified on-line during the game, and how future regimes and times of regime transi-
tions can be forecast during the game.