[Swarm-Modelling] [Fwd: Re: [ABMs in finance]

[Steve Railsback]

-------- Original Message --------
Subject: Re: [Swarmfest2006] Keynote speakers
Date: Fri, 21 Apr 2006 09:09:16 -0600
From: Marcus G. Daniels <address@hidden>

Btw, the idea of using ABM as tool actually trading seems potentially
useful.  The idea would be to identify a network of traders (e.g. a
number of large companies like Goldman Sachs, Merill Lynch, etc.) that
would tend to show large back and forth trading patterns that would move
the price in the short term.  Then use data mining techniques (e.g.
CART) to identify some likely decision variables, and then study that in
detail and build an ABM of the players reactions dependent on the
environment, using observed estimates of their inventories, and having
the notion of those player's recent memory of observed trading.  The
goal would be to use simulation to catalog trading scenarios involving
large volume and extended trading periods such that the price direction
would change, could be predicted with better than average accuracy,
while giving time to jump in and out with a buy and sell.   But the data
mining is hard because the big investment banks can have dozens of
automated or semi-automated trading strategies in place all conflated
under one name in the data.

This is in contrast to what Doyne has been doing which considers
aggregate order flows, market order rates, deletion rates, and posits
some order splitting and distribution of inventories.  (At least as of a
year ago.)   The price predictions don't come from simulations, but have
been tested by them and also by real world exchange datasets (the two
papers that Pietro cites, respectively).

This practice of trading I describe would consider micro-ecologies with
local inefficiencies using a potentially wide area of variables (and
brute force) whereas what Doyne is interested in is more about
predicting global, or scale free price phenomena from simple aggregate
indicators.