Bereiche | Tage | Auswahl | Suche | Downloads | Hilfe

AKSOE: Physik sozio-ökonomischer Systeme

AKSOE 12: Postersitzung

AKSOE 12.5: Poster

Montag, 11. März 2002, 16:00–18:00, D

Modeling FX-Markets based on Intelligent Agents — •Georg Zimmermann and Ralph Grothmann — Siemens AG, Corporate Technology, Otto-Hahn-Ring 6; 81735 Muenchen

We present an uncommon approach of multi-agent modeling based on the idea of an elementary cognitive system with three functionality features (perception, internal processing and action). At first, we introduce these features inductively.Our considerations should be seen as necessary conditions for a cognitive system.

Second, we deal with a deductive approach to the functionalities. As we will show, all functionalities of the inductive approach can be derived deductively from homeostasis. Homeostasis may be seen as a dynamical stable state of the internal processing.

In order to derive a structural representation of the functionalities, we propose so-called zero-neurons in combination with recurrent neural networks. Zero-neurons are input / output neurons with fixed target values of zero. On this basis, we are able to model homeostasis and thus, all other features of the cognitive system.

The cognitive system can be seen as a model of the decision behavior of a single agent. Now, the interaction of many agents can be used to explain the price formation of a market. Fitting real-world data of a FX-market, it turns out, that our approach is superior to conventional forecasting techniques.