Understanding flash crash contagion and systemic risk: a calibrated agent-based approach

The global financial system is a sociotechnological complex network, in which millions of economic agents interact over timescales ranging from months to milliseconds. The decade since the near-collapse of this system has been characterised by the meteoric rise of computerised algorithmic trading. This transition has resulted in markets that are vulnerable to new forms of systemic risk, as exemplified by the Flash Crash of May 2010, and related events. The failure of extant models to predict, or mitigate, either the Global Financial Crisis, or the Flash Crash, has led financial regulators to increasingly engage with academia in search of new approaches through which to understand such systems. Agent-based modelling, rooted in complex systems theory, has emerged as a compelling method. However, there remains a pressing regulatory need to understand how systemic (macro- level) network effects are influenced by (micro-level) algorithmic trading. To address this, we developed and assessed a novel combination of macro- and micro-level agent-based models of financial markets, demonstrating the benefits and applicability of the agent-based methodology. We con- structed quantitatively validated network representations of all liquid as- set portfolios held by investors in US equities, and then simulated the exchange of assets using quantitatively calibrated and validated market mechanisms, incorporating key regulatory interventions such as price limits and circuit breakers. Our microstructure calibration dataset spans a range of assets and dates that far exceeds previous approaches. By subjecting the validated model to counterfactual scenarios, such as price shocks, alternative network topologies, and alternative agent behaviours, we presented the first in-depth analysis of systemic risk over intraday timescales. We assessed the effectiveness of current, and proposed, regulations at mitigating this new form of high-frequency financial distress. We find that the actions of distressed market participants facilitate new propagation channels for intraday systemic risk, and flash crash contagion, across the financial network. We also find that systemic risk has a complex non-linear dependence on network topology. We further identify stable and unstable network and agent-behaviour scenarios. The scope and complexity of the model allows the first quantitative analysis of the systemic effect of price limits and circuit breakers. Our insights include the fact that current regulation is inadequate in certain circumstances, and that its effectiveness strongly depends on fund liquidation behaviour. We perform the first ever quantification of the effectiveness of a novel pol- icy proposed by a major market participant, and find it could potentially reduce systemic risk. Based on our work and results, we argue that a hybrid micro-macro approach offers a powerful new lens, through which regulators can assess novel threats to financial stability in the modern, computerised, market- place.