Rapid evolution of financial markets and return of ACE

By Robert Hillman, PhD, Chief Investment Officer, Neuron Advisers LLP

Published: 25 October 2017

The rapid growth of algorithms in the world of finance is giving regulators, economists and investment professionals plenty to think about. They may need to look toward approaches more in common with Minecraft than with traditional methods of economic analysis.

In the seven years since the May 2010 ‘flash crash’, multiple theories have been proposed as to the source of that and subsequent crashes.No simple explanations are readily available because conventional research methodologies are not well suited to analysing these events. Economists typically take a two-pronged approach to analysis, collecting and analysing historical data, and building toy models of the situation at hand. But while there is no shortage of data in terms of sheer volume of information (because the events take place at such high-frequency that there are often millions of records per minute), there are very few distinct events to study and from which to generalise. In terms of building models, economists are well versed in modelling human decision making, but few have until recently considered the implications of trading that takes place so fast that European regulators have had to recently propose a synchronisation of clocks to within a millionth of a second to reduce ambiguity about the order and sequence of trading orders.2

With conventional methodology falling short, some economists are looking towards alternative methods including an approach developed in the late 1980s and 1990s in an effort to build an alternative way to study financial markets via the simulation of people (often called ‘agents’) with computer programs. Often labelled ‘agent-based computational economics’ or ACE, it built on 1950s work by pioneers like Herbert Simon who studied human behaviour as computational processes and vice versa. These models offered new explanations for concerning phenomena like bubbles and crashes, but they had little impact on economics or investment management professions at the time.

ACE techniques are ideal for simulating today’s technologically-driven automated markets. Within artificial markets, researchers can explore what types of algorithms, external shocks and exchange rules might be prone to generate flash crash type behaviour. Exchanges can explore the effects of defensive mechanisms such as circuit breakers for example, and on the other side of the fence, investors can explore what the effect of circuit breakers might be on their operations. For these experiments, computer simulation is not merely conveniently aligned to the reality of modern trading technology, it is vital. Two lessons from the earlier period of ACE were that the interaction of simple algorithms can lead to complex dynamics, and that macro level behaviour can be impossible to predict from analysis of the components in isolation – the whole being greater than the sum of the parts.

Simulation of artificial markets can also generate unlimited quantities of data and suggest behaviour not previously recorded in the real world. Interesting parallels can be drawn here between innovative simulation-based approaches now being used in weather forecasting and the modelling of extreme and ‘unseen’ weather events by the Met Office in the UK3 and how similar techniques are being used in finance.4 During periods of rapid structural and technological change, artificial simulation can complement the limited historical data available. For example, risk and technology managers might find ACE techniques helpful in suggesting stress scenarios that might lead to market or operational risks, and in doing so take steps to future-proof their business.

New uses for ACE are not likely to be limited to high-frequency contexts. The last few years have seen the spread of algorithms right across the investment management landscape. Many fund managers may nowadays best be described as designers and guardians of rules-based strategies. The proliferation of exchange traded products alongside the commoditization of dynamic portfolio management techniques like risk parity and smart beta are only accelerating this trend. The original ACE approach approximated human investor behaviour by modelling agents as simple algorithms, and one reason ACE failed to disrupt mainstream economics was because economists viewed the approach as ad hoc and lacking logical foundations. But twenty years of actual evolutionary processes at work have brought us to a situation where today, real markets closely resemble the algorithmic markets of ACE. Science fiction has become fact.

There is some urgency around these issues. The August 2015 market crash prompted questions about the role of lower frequency algorithms, specifically risk-parity strategies and ETFs, in contributing to market volatility and disrupting the relative pricing of cash and derivative instruments. ACE models offer one means of studying these issues as part of a wider simulation or ‘virtual markets’-based approach and financial regulators have been developing similar techniques, operationalising agent-based models and other network approaches to study the effect of increased connectedness of the propagation of risks throughout systems and networks.5 If economists are reluctant to embrace the power of virtual environments, there is a new generation of Minecraft players who may be more at ease with the concepts.

To contact the author: 
enquiries@neuronadvisers.com

Footnotes
[1] See for example ‘Findings Regarding the Market Events of May 6, 2010’ Report of the Staffs of the CFTC and SEC to the Joint Advisory Committee on Emerging Regulatory Issues. September 2010. https://www.sec.gov/news/studies/2010/marketevents-report.pdf
[2] Under Article 50 of MiFID II, see for example https://www.thetradenews.com/Regulation/Clock-synchronisation%E2%80%A6time-is-ticking/?l=tl
[3] For more on this see the Met Office website, in particular https://www.metoffice.gov.uk/news/releases/2017/high-risk-of-unprecedented-rainfall
[4] See Hillman (2017), ‘Extreme Weather and Extreme Markets – Computer Simulation Meets Machine Learning’ on http://www.neuronadvisers.com/
[5] For example, see Braun-Munzinger, K, Liu, Z and Turrell, A (2016), ‘An agent-based model of dynamics in corporate bond trading’, Bank of England Staff Working Paper No 592, available at http://www.bankofengland.co.uk/research/Pages/workingpapers/2016/swp592.aspx

Disclaimer
Unless clearly indicated otherwise, this article represents the opinions of the author based on his research experience.