Re: [Swarm-Modelling] decision modelling

[Steve Railsback]

address@hidden wrote:
> 
> Basically I would like to simulate or "predict" the voting of creditors about
> the reorganization plan.
> 
> When a firm goes bankrupt it is possible to start a reorganization. Therefore
> you create a reorganizantion plan. The creditors of this firm now have the
> right to say whether they are in favor of the plan or whether they are not.
> 
> Each creditor has his own mind/decision rules/set of variables that determines
> their behavior. The creditors are put into different groups, e.g. bank. Within
> one group the decision rules should be pretty much the same.
> 
> The decision about the plan is made within a meeting that all creditors may
> attend. In fact they do communicate with each other and so one creditor might
> affect another creditor through this communication. At the end of the meeting
> each creditor votes.
> 
> The modell I think of shows each creditor as one agent. Each agent has his own
> set of variables/rules (his own "mind"). According to this they have something
> like a "first opinion" about the reorganization plan and how they would like
> to vote. Then the agents communicate with each other for a certain time by
> message passing. During the communication an agent can possibly change his
> mind. At the end of the process all agents have their "final opinion" about
> the plan. According to this they vote at the end of the process. If you add
> the individual decision you get the final result of the voting.

So it looks like the creditors' decision on how to vote is the critical
behavior to model. There are several ways people model decision-making
in agent-based models:

a. Fitness-seeking: the agents do some calculations to figure out which
choice is best for their well-being ("fitness" in ecology, "money" in
economics). To me, this seems like a reasonable approach for your
problem- I can easily see the creditors sitting there with calculators
figuring out which alternative gets them the most money back. 

b. Game-theory (about which I do not remember enough to comment)

c. Evolved traits: a number of models in economics set up a problem (as
you do), then use genetic algorithms to artificially evolve behavior
rules that work well. This approach is especially interesting when
agents compete against each other, as they do in your problem...but this
approach kind of assumes the real agents have highly-adapted,
finely-tuned traits. I'm not sure that's realistic for your problem.

Places to start getting an idea of what has been done:

a. Robert Axelrod's books, especially Axelrod, R. 1997. The complexity
of cooperation: agent-based models of competition and collaboration.
Princeton University Press, Princeton, New Jersey.

b. There are one or two books on modeling economics in Swarm. One is: F.
Luna and A. Perrone, eds. Agent-based methods in economics and finance:
Simulations in Swarm. Kluwer Academic Publishers. I think there is also
one edited by Benedikt Steffanson (should be on the www.swarm.org site
somewhere). 

c. Leigh Tesfatsion at Iowa State University maintains a huge web site
on agent-based economics: http://www.econ.iastate.edu/tesfatsi/ace.htm
In fact, I would not do anything without first contacting Dr.
Testfatsion- she has done all kinds of work and built some nice
software.

Steve Railsback
-- 
Lang Railsback & Assoc.
250 California Ave.
Arcata CA  USA 95521
707-822-0453; fax 822-1868