Abstract:
Computational advancements today have made modeling of some complex
social systems like stock markets possible by providing conditions for controlled
experiments. Such a control has proved helpful in exploring and studying dynamics
of these complex markets, by isolating the cause and effect relationship of different
phenomena in here. Extensive research in multi-agent simulation of financial
markets has been carried out to study various market models. Much attention has
not been paid to changing artificial market properties during the simulation, which
is actually the real world scenario. We hypothesize that dynamically controllable
simulating environment shall prove much more beneficial for more sophisticated
and richer experimentation by the market specialists. This research contributes
towards the artificial market paradigm by letting the financial scientists control
simulated market properties at two levels. Macro level is controlled by enabling
change in the composition of market participants during simulation. Micro level
control describes defining a wide set of individual properties of participating agents
that is made possible through Agent Construction Framework (ACF). This
framework provides extensibility in defining new agent types with varying
capabilities thus supporting wide range of heterogeneity in market participants.
This research presents an extensible architecture for building such a dynamic
controllable market that effectively imitates the real world. This extends the
usability and effectiveness of agent-based simulating environments to new levels.
