Exploring Assignment-Adaptive (ASAD) Trading Agents in Financial Market Experiments

Automated trading systems in the global financial markets are increasingly being deployed to do jobs previ- ously done by skilled human traders: very often a human trader in the markets simply cannot tell whether the counter-party to a trade is another human, or a machine. Clearly, automated trading systems can easily be considered as "intelligent" software agents. In this paper we report on experiments with software trader- agents running the well-known "AA" and "ZIP" strategies, often used as reference benchmarks in previously published studies; here we suggest disambiguated standard implementations of these algorithms. Then, using Exchange Portal (ExPo), an open-source financial exchange simulation platform designed for real-time be- havioural economic experiments involving human traders and/or trader-agents, we explore the impact of intro- ducing a new method for assignment adaptation in ZIP. Results show that markets containing only assignment- adaptive (ASAD) agents equilibrate more quickly after market shocks than markets containing only "standard" ZIP agents. However, perhaps counter-intuitively, in mixed heterogeneous populations of ASAD agents and ZIP agents, ZIP agents outperform ASAD agents. Evidence suggests that the behaviour of ASAD agents act as a new signal in the market that ZIP agents then use to beneficially alter their own behaviour, to the detriment of the ASAD agents themselves.