Dresden 2003 – scientific programme

Parts | Days | Selection | Search | Downloads | Help

AKSOE: Physik sozio-ökonomischer Systeme

AKSOE 4: Mikro-ökonomische Modelle und Multi-Agenten-Systeme

AKSOE 4.3: Talk

Tuesday, March 25, 2003, 10:30–11:00, BAR/205

Cognitive Agents with Utility Functions — •Hans-Georg Zimmermann, Christoph Tietz, and Ralph Grothmann — Siemens AG, Corporate Technology, CT IC 4

A financial market consists of a large number of interacting agents. Thus, a natural way of explaining and predicting market prices is to analyze the decision making of the agents on the microeconomic level and to study their interaction on the macroeconomic side of the market. We present a new approach of multi-agent market modeling. The decision making of an agent is modeled with a homeostatic dynamic system. Homestasis means, that the system maintains an internal equilibrium. On this basis, we derive the properties perception, internal processing and action of a cognitive system. As a structural representation of homeostasis, we propose so-called zero-neurons within a time-delay recurrent neural network. As an extension to this concept, we integrate a utility function into the cognitive system. By this, we are able to explain the decision behavior of an agent by a detailed analysis of his long-term objectives. The aggregation of agents decisions leads to a multi-agent based approach of market modeling. Merging the economic theory of multi-agent market modeling with neural networks, our models concern semantic specifications instead of being limited to ad-hoc functional relationships. In an empirical study, it turns out that our approaches are superior to a traditional econometric forecasting techniques.