Agent Based Modeling by Paul Cottrell

Agents are the atoms of complex systems.  These complex systems can be considered adaptive and exhibit emergent behavior.  Agent–based modeling is a bottom–up strategy.  These agents can be programmed to interact with external and internal environments.  For example, researchers are using agent–based modeling to enhance stress testing in the banking sector.  Standard orthodoxy for stress testing is to ask the individual banks how their balance would be affected by a ten percent reduction in assets or a thirty percent increase in loan defaults.  The problem with the current orthodoxy model approach is that it does not capture a realistic contagion in the financial markets.  Individually the bank many seem to pass a prescribed test, but in fact will actually fail if one considers the amplification of financial stressed counterparties.  Therefore, modeling in an agent–based manner will allow for counterparty feedback loops and a more realistic simulation. 

Complex behavior can emerge from agent–based modeling with just simple interaction rules.  In our bank stress example, these agent rules and parameters could be total loans outstanding; payment dates and amounts between counterparties; capital reserve requirements; and short term funding rates.  This might not be an exhaustive list of agent rules and parameters, but the main idea can be understood.  With these simple rules the banking system will show weaker links than the current stress testing methodology—revealing that stronger banking firms might have much more counterparty risk than their risk models suggest.  Through agent–based modeling of the banking sector you will notice how fragile the financial system is to asset depreciation.

Agents should be able to morph their behavior.  Allowing for a morphing of behavior will add functionality to the complex adaptive system, producing a more anti–fragile system (Taleb, 2012).  When banks are assessing their counterparty risks and determining what reserves are required this is a brittle strategy.  They are trying to harden their balance sheet to absorb a negative shock, but this strategy does not allow them to adapt to a changing environment—leading to the needs of a lender-of-last-resort, i.e. central bank intervention.  An adaptive banking system, in theory, would increase their balance sheet reserves in good times and drawdown their reserves in bad times, which reduces the need for central bank intervention.

When agents can morph their behavior, evolutionary pathways are revealed and allows for heterogeneous populations.  This morphing of behavior might be in the form of agents assessing their agent neighbor’s environmental successes and trying to copy them.  Within a large state–space there will be diverse groups optimized to their respective local environments.

Another promising field in agent–based modeling is the use of virtual reality.  Agent–base modeling is a simulation, albeit of multiple dimensional–space.  It is a natural progress to apply agents’ interactions in a virtual world, whereby a researcher can explore the interactions in an immersive way.  Game developers have been programming agents in virtual worlds for many years, but usually as first-person-shooting games.  To apply this gaming technology to modeling financial markets is not that hard to envision. One just needs to program the agent with certain rules and allow the different agents to interact.  The virtual world it just a visualization tool to understand the complexity that emerges.