Co-evolutionary Dynamics in a Simulation of Interacting Financial-Market Adaptive Automated Trading Systems

In present-day major financial markets around the world, adaptive automated trading systems are responsible for many more transactions than are human traders, and human traders have largely been replaced by trading machines that can process super-human quantities of data and react to market events at super-human speeds: this paper reports new results from a high-fidelity simulation of such highly automated financial markets, populated by minimal adaptive trading strategies, and presents results showing that competition between trading systems leads to unstable market dynamics and convergence on economically suboptimal outcomes. The simulations reported here used the public-domain open-source market simulator BSE, which offers an accurate model of a contemporary limit-order-book financial exchange, which was then extended by adding a collection of individual adaptive automated trading entities (‘traders”), each of which uses a simple stochastic hill-climbing optimizer to adapt its zero-intelligence trading strategy over time, trying always to improve profitability. The traders interact and transact with each other on sub-second timescales, but this paper focuses on the simulated market's dynamics over extremely long periods of times (hundreds of days, during which time many hundreds of millions of transactions can occur) and shows that while the system can remain in seemingly stable states for protracted periods, the overall long-term dynamics of the system can be unstable and economically inefficient. The extended BSE source-code written for these simulations is freely available on GitHub, for use by other researchers