Paul Jefferies

Application of multi-agent games to the prediction of financial time-series

We report on a technique based on multi-agent games which has potential use in the prediction of future movements of financial time series. A third-party game is trained on a black-box time series, and is then run into the future to extract next-step and multi-step predictions. In addition to the possibility of identifying profit opportunities, the technique may prove useful in the development of improved risk management strategies.

Crowd–anticrowd theory of the minority game

The Minority Game is a simple yet highly non-trivial agent-based model for a complex adaptive system. Here, we provide an explanation of the games fluctuations which is both intuitive and quantitative, and which applies over the entire parameter range of interest. The physical idea behind our theory is to describe the interplay between crowds of like-minded agents and their anticorrelated partners (anticrowds).

Generalized strategies in the minority game.

We show analytically how the fluctuations (i.e., standard deviation sigma) in the minority game can decrease below the random coin-toss limit if the agents use more general, stochastic strategies. This suppression of sigma results from a cancellation between the actions of a crowd, in which agents act collectively and make the same decision, and those of an anticrowd, in which agents act collectively by making the opposite decision to the crowd.

Dynamics of the time horizon minority game

We present numerical and analytic results for a new version of the minority game (MG) in which strategy performance is recorded over a finite time horizon. The dynamics of this time horizon minority game (THMG) exhibit many distinct features from the MG and depend strongly on whether the participants are fed real, or random, history strings. The THMG equations are equivalent to a Markov Chain, and yield exact analytic results for the volatility given a specific realization for the quenched strategy disorder.

Anatomy of extreme events in a complex adaptive system

Large changes tend to dictate the long-term dynamical behaviour of many real-world complex systems in both the natural and social sciences. We provide an analytic, microscopic analysis of extreme events arising in an adaptive population comprising competing agents (e.g. species, cells, traders, data-packets). Our results reveal a taxonomy of such extreme events, and provide a microscopic understanding as to their build-up and likely duration.

Managing Catastrophic Changes in a Collective

We address the important practical issue of understanding, predicting, and eventually controlling catastrophic endogenous changes in a collective. Such large internal changes arise as macroscopic manifestations of the microscopic dynamics, and their presence can be regarded as one of the defining features of an evolving complex system. We consider the specific case of a multiagent system related to the El Farol Bar model and show explicitly how the information concerning such large macroscopic changes becomes encoded in the microscopic dynamics. Our findings suggest that these large endogenous changes can be avoided either by pre-design of the collective machinery itself or in the postdesign stage via continual monitoring and occasional “vaccinations.”

Toward an Understanding of Financial Markets using Multi-agent Games

We report on our use of multi-agent games to understand financial market behavior. In addition to discussing the background to the multi-agent games themselves, we report a technique which may prove useful for forecasting future movements of financial time-series. A third-party game is trained on a black-box time-series, and is then run into the future to extract next-step and multi-step predictions. Such predictions have potential use as the basis for improved risk management and portfolio optimization strategies.