Agent-Based Modelling of High Frequency Traders
The phenomenon of the 'flash crash' is inextricably linked with the emergence of fragmented electronic financial markets in recent years, where each asset can now be traded on a number of different venues (i.e. exchanges and dark pools). The Flash Crash differs from the familiar long-standing type of financial crash in three ways. First, as its name suggests, it happens very, very fast - usually in under a second. Second, it is often accompanied by a similarly fast post-crash recovery. Third, there appears to be a lot of them. This suggests that market structure and behaviour are at the very least different from what we understand from history - which raises questions about our understanding of the stability and fragility of these markets, and possibly about their efficacy and efficiency as price discovery tools. There is a lack of rigorous scientific analysis and evidence on these questions.
The first aim of this research project is to use agent-based modelling to categorise the different types of high frequency trading strategies, and use simulations to measure their effect on observed market patterns. The developed simulation platform is then validated by comparing the results against patterns observed in real high frequency data. Once the agent-based system is developed and validated, the second aim of the project is to use the platform as a testbed to evaluate market policies which discourage harmful high frequency trading behaviour, and to compare interventions by financial exchanges to prevent systemic risk.
Algorithms and computational methods: Agents
Programming languages and libraries: Python
Transdisciplinary tags: NGCM