PASCAL - Pattern Analysis, Statistical Modelling and Computational Learning

Artificial Agents and Speculative Bubbles
Yann Semet, Sylvain Gelly, Marc Schoenauer and Michele Sebag
In: Computational Finance and its Applications, April 2004, Bologna.


Pertaining to Agent-based Computational Economics (ACE), this work studies a community of computationally simulated agents, and is centered on exchange price fixing under the double auction mechanism. Two distinct settings are presented. The first one is based on a simple simulation of a goods market, where two populations, buyers and sellers, interact to iteratively set an exchange price for a single consumption good (an asset cannot be stored and later resold). The growth and steadiness of exchange prices as well as the global surplus are both found to consistently converge to the predictable equilibrium. The influence of supply, demand and informational noise is studied in detail. The study is then carried on to a network of interconnected such markets, where the reciprocal influences of the different markets with respect to their connecting topology is observed. The second setting is an elementary artificial stock market where the focus is on the relationship between individual rationalities and the birth, rise and panicky downfall of speculative bubbles. Goods are no longer perishable and agents can be both buyers and sellers. Speculative bubbles, as described by Smith or Ruffieux, appear in two contexts that differ in complexity: First, with "Zero-Intelligence" traders and a finite time horizon that makes hopes for dividend collecting decrease with time; Second in an horizon-free context where agents make use of the "greater fool" hypothesis and base their behaviour on an endogenous estimation of risk. Ongoing research is inspired by the {\em El Farol bar} and the {\em Santa Fe market}: agents are endowed with evolving sets of rules and the goal of the experiments is to investigate how speculative bubbles and market behaviour in general can be explained by a particular ecology of strategies. This work is, additionally, conducted on a peer-to-peer distributed computing platform, the DREAM (Distributed Ressource Evolutionary Algorithm Machine), that permits the gathering of computational power by interconnecting voluntary CPUs across the world in a transparent and convenient fashion. This choice is intended to allow for an inquiry into the influence on market behaviours of large numeric scales in geographically distributed market interconnections and asynchronous stochastic high frequency trading.

PDF - Requires Adobe Acrobat Reader or other PDF viewer.
EPrint Type:Conference or Workshop Item (Paper)
Project Keyword:Project Keyword UNSPECIFIED
Subjects:Learning/Statistics & Optimisation
ID Code:617
Deposited By:Marc Schoenauer
Deposited On:29 December 2004