In a new report the Government Office for Science has suggested risks in financial computer trading are potentially as severe as those facing the nuclear industry, and that markets may have to follow the safety standards found in the nuclear sector.
There has been a major nuclear "incident" somewhere in the world about every ten years involving reactors, but worldwide stock markets have recently seen regular market meltdowns, caused by network outages or problems with ultra-fast algorithmic trading systems.
A review of the risks involved in relying on technology to run financial markets - "Computer trading and systemic risk: a nuclear perspective" - was commissioned as part of the UK Government's Foresight Project, The Future of Computer Trading in Financial Markets.
Co-authored by City University London's Professor Robin Bloomfield and Dr Anne Wetherilt from the Bank of England, the report says the financial markets have "evolved to become complex adaptive systems highly reliant on the communication speeds and processing power afforded by digital systems".
It says "their failure could cause severe disruption to the provision of financial services and possibly the wider economy".
The report says "there are a very wide range of areas where the issues and practices in the nuclear industry might resonate with those raised by the evolution of computer-based trading".
These areas include the approaches to systemic risk definition and evaluation; and the definition of protection system parameters, risk controls and architecture.
The authors say, "We consider a serious nuclear incident that has the potential for the release of radioactivity with associated plant damage as a 'systemic event' and hence make the link to a financial market crash: an event that both damages the market and also potentially impacts the wider financial system and the broader economy."
They add that the development of the nuclear industry approach to safety has been driven by the need to engineer systems that provide social and economic benefits with "tolerable risks", to evaluate and explain the nature and extent of these risks, and to provide a framework that allows for scrutiny at varying levels of independence ranging from technical experts within the industry as well as pressure groups.
Many critics of the nuclear industry however would argue that it has been very poor at explaining the risks, with one of the worst public relations records in the business.
The paper says both industries have to consider the basic concepts of hazard, risk and accident; probabilistic safety analysis and the concept of a design basis; tolerability of risk and the "as low as reasonably practicable" (ALARP) principle; and numerical risk targets.
The authors say the nuclear industry has a formalised approach to defining the classes of consequence, the categories and frequencies of initiating events. It uses theory, models and experiments to justify the risk analysis.
This means that the industry can set risk targets for classes of accident and different classes of people, and discusses tolerability and proportionality in reducing them further.
In doing so, the authors say, the nuclear industry accepts that many things are hard to quantify, but there is nonetheless an emphasis on ranking risks, setting targets for risk reduction, and debating whether both the risks and the targets are accurate and acceptable.
They say the nuclear safety analysis framework allows systematic design of protection and mitigation systems that cover not only what they have to do, but also how much they have to be trusted. These systems, the authors say, use diverse mechanisms to ensure that the overall protection works when it is needed.
The paper asks whether the rapid development of computer-based trading in financial markets requires the adoption of additional risk concepts and tools, like those used in the nuclear industry.
The authors say, "Our analysis suggests that the following questions are worth asking: 'Is it possible to have a more precise description of risk categories?, is it possible to define precise tolerability criteria?, and is it possible to define numerical targets - if not, how does one define 'acceptable' risk?"
They also question whether the financial industry should work on the notion of a "design basis", which would characterise those adverse endogenous and exogenous events that systems should withstand?