Abstract

Recent years and the outbreak of world's economic crisis in 2008 proved the crucial importance of reliable analysis of market dynamics. However, werarely apply models of proper detail level (the global prosperity forecast of 2007 can be seen as a grim proof). The behaviour of individuals and companies is far from being ideal and rational. Many claims that the economic paradigm of rational expectations (coming from J. Muth and R. Lucas) should be extended by theory of irrationality. Moreover, the knowledge of market is usually limited, as the price of intelligence is high. Therefore, typical model of supply-demand is applicable in rare cases only. On the other hand, complicated theoretical economic models usually lack support from flexible simulation tools. That way usable simulation playgrounds for testing user- based scenarios are rare. In this article authors propose a dynamic relationship network designed for a computer simulation of market behaviour. The core assumption is, that decisions of market subjects are not based by objective knowledge, but rather on associative memory (information gathered by subject and by trustworthy partners). The obvious example: one would rather buy from a supplier recommended by a friend, than try to check pricing (and quality) in all available sources. Moreover: all actions between partners (including information exchange) influence the power of mutual relation (this is similar to Flood & Dresher Prisoner's Dilemma model and the outcomes of Axelrode Tournament).The proposed platform is based on a newly designed and implemented (in C++) library. It consists of three main layers: autonomic agents (including both abstract models of behaviour and scripts), network management routines (rules for creating and maintaining the relationships between agent) and measurement interface. This model is flexible enough to implement variety of agent-network configuration, but special attention is placed on purely economical scenarios.In the classical supply-demand model, the market price is fixed at the level set by rational decisions of all involved parties. However, in real life subjects (especially individual customers) operate in limited knowledge (no access to all sources or the access is expensive). The lack of full information is balanced by an associative memory stored in relation network. This network can be utilised in price establishment. However, the result is not fixed, but rather fuzzy (the transaction price can differ on one market). In this article, authors present the model of a relationship-drivenmarket and its working implementation. There is also a discussion ofoutcomes - i.e. the impact of network parameters on actual price evaluation. Platform proposed in the paper should be seen a result of synergy of knowledge combined from three areas of science. The IT model and implementation details comes from computer science and mathematics (graph theory). The market behaviour models and simulation's output definition are taken from economics. Finally, the agent scripts are based on observations from psychology (i.e. theory of relationships and decision making ).

Authors (2)