This review was commissioned as part of the UK Government’s Foresight Project, The Future of Computer Trading in Financial Markets. The views expressed do not represent the policy of any Government or organisation.
Jean-Pierre Zigrand
London School of EconomicsIntroduction
A driver for future risk and catastrophes lies in the fact that the seemingly large bio-diversity of traders may be illusory and that in a stress situation many algorithms quickly and unwittingly coordinate, act in unison and feed on each other in a feedback loop, thereby leading to a disproportionate value destruction.
The first of the two main themes concerns uniformity (or its antonym diversity) and synchronisation. Even though various strategies and algorithms can appear to be genuinely diverse, there is the risk that a market event - even one quite unrelated to the strategies underlying the algorithms - can under a given set of circumstances make the actions underlying the strategies nearly instantaneously uniform and synchronised. This paper explores a number of possible situations that could lead to such unintended coordination. Furthermore, the trigger leading to lock-stepping can be very nonlinear in the driving variables, which means that to an outside observer who does not see the inner workings of the markets, the ganging-up appears very unpredictable.
The second, and often complementary, theme concerns feedback effects or feedback loops, mainly positive. Positive feedback effects act as a mutually reinforcing propagation mechanism. This mechanism is strengthened precisely the more players act upon it at the same time, i.e. the more they synchronise, which is the first theme. In finance, positive feedback effects often arise from dual roles played by endogenous variables. For instance, prices both reflect fundamentals and actions, and drive actions, too. In standard models, the latter comes from various constraints, such as leverage caps, VaR constraints, solvency constraints, policy responses (such as inflation targeting by central banks) and so forth. But with computer driven trades, the link may be more mechanical, systematic and immediate, and hence more direct and more powerful. The feedback loop emerges when both aspects of endogenous variables (as signal and as imperative to action) are in place.
In our mind there is little doubt that positive feedback loops are underlying forces behind many, if not most, financial crises. The fact that neither in the Findings Regarding the Market Events of May 6, 2010 (The Staffs of the CFTC and SEC (2010)) nor in the Recommendations regarding regulatory responses to the market events of May 6, 2010 (Joint CFTC-SEC Advisory Committee on Emerging Regulatory Issues (2010)) there appears to be any explicit mention of the feedback loop dynamics inherent in such events is a cause of some worry. Such nonlinearities,1 and the inherent unpredictability and speed of unravelling that go with such feedback nonlinearities, underlie many surprising financial crises and deserve to be singled out as a major issue. In principle positive feedback loops could lead either to welfare improving or to welfare destroying outcomes, although economic logic more often than not leads to deleterious effects.2 We believe that a very simple – though necessarily superficial – criterion that market regulators ought to keep a watchful eye on is the minimisation of net positive feedback loops, those naturally present in markets as well as those caused by the very regulations aimed to prevent crises.
In this paper, we outline some of the risk drivers from such deleterious loops in computer-based environments. The seeds of such crises episodes are to a non-negligible extent already sown and exist today as potentials, even if the conditions or realisations that would bring them to the forefront have not materialised yet. In that sense many of the risk drivers of the next 5 to 10 years are the same as the ones present now, with the caveat that given the perceived trends in technology, the unravelling may become more forceful and more rapid. For other risk drivers, the technological conditions that would make them dangerous are yet to be fulfilled through potential future developments.