Inferring the Composition of a Trader Population in a Financial Market

TL;DR

This paper presents a novel method for predicting financial market movements by inferring the diversity of trading strategies among agents, accommodating more complex strategies and reducing bias in parameter estimation.

Contribution

It extends previous frameworks to include more intelligent agents, broader strategy sets, and a bias removal mechanism for better heterogeneity inference.

Findings

Enhanced prediction of market movements.

Ability to incorporate diverse agent models.

Reduced bias in heterogeneity estimates.

Abstract

We discuss a method for predicting financial movements and finding pockets of predictability in the price-series, which is built around inferring the heterogeneity of trading strategies in a multi-agent trader population. This work explores extensions to our previous framework (arXiv:physics/0506134). Here we allow for more intelligent agents possessing a richer strategy set, and we no longer constrain the estimate for the heterogeneity of the agents to a probability space. We also introduce a scheme which allows the incorporation of models with a wide variety of agent types, and discuss a mechanism for the removal of bias from relevant parameters.