Stock prices and volume in an artificial adaptive stock market

Abstract

we present an application of neural networks to financial markets, experimenting with various learning mechanisms that may describe reasonable behavioral rules followed by agents acting under incomplete information about the environment. Each agent is described by a neural network who decides the price she is willing to pay for an asset, and the quantity she wants to buy or sell. Agents differ as to a number of dimensions, and in particular the trading strategy that may be used to divide the traders in two different categories. The interactions among the different agents determine every day the market price and the volume of transactions. We analyze the behavior of the market as a function of the proportions of traders in the two categories, showing that increasing heterogeneity positively affects the market volume, without visibly increasing the volatility of prices. We also experiment with different learning mechanisms, that can be interpreted in economic terms as forcing on the agents differing degrees of risk aversion.