Stopping financial crashes with physics
How can we stop glitches & cyber-attacks crashing stock markets? Imperial postgraduate student Vincent Sebag developed a solution with econophysics.
In August 2012, an incomplete software update at Knight Capital caused around 4 million rogue stock-market orders to be executed. Over the next 45 minutes, some $457 million was lost as a result.
Whilst this event was caused by a combination of a software glitch and human error, it’s very possible that a cyber-attack could have the same effect.
Finsec - the cyber-security of financial systems – is a thematic focus for the Institute for Security Science and Technology. We caught up with Vincent Sebag to hear about his recent MSc. project at the ISST, applying econophysics to finsec.
What did your project look at?
We created a risk-mitigating tool which detects and stops financial crises as they are happening.
Past events have shown that existing crisis-mitigating regulations are not always effective, particularly with crises led by single financial actors, for example the Knight Capital glitch.
Our novel solution targets such crises; those stemming from a single (or a few) financial actor, such as exchanges, brokerage companies or high frequency traders. It’s a last resort protection, a kill-switch, to stop the financial actor continuing to fuel the crisis.
Why is this important?
It can help stop the immediate financial loss, but more importantly, a single financial actor has the potential to cause a world crisis. So detecting crises from single actors early is extremely important.
How can a single actor have such a big impact?
Firstly, markets are highly and increasingly correlated; shocks on one market tend to quickly propagate to all others with no attenuation. Secondly, the potential impact of individuals on a market has grown; single orders can be greater than $500k per exchange. Thirdly, financial actors have an increasing risk of malfunction, be it a through a trading software glitch or cyber-attack.
All these taken together means it is increasingly likely that a single actor can generate a crisis, which will propagate through all markets, having a high-impact.
What was the solution from your research, and how is it different?
Our solution differs from other crisis-mitigating-systems as it targets market actors rather than financial products.
We measure the impact of each financial order sent by an actor and deduce whether it is fueling a crisis. However, rather than starting with orders, we detect financial shocks and analyze how they propagate across the different markets. From this we can understand who triggered, who is fueling and who is dampening a crisis, and then operate a kill-switch on the potential rogue actors.
How do you detect a shock?
We analyze the co-movement of highly correlated pairs of stocks, such as fungible stocks. These are stocks that are traded on two different markets, for example you could buy Toyota shares on the NASDAQ and on the London Stock Exchange.
The two prices are highly correlated. In unperturbed times they are perfectly equal due to arbitrage. In perturbed times however, discrepancies in their co-movement show the existence of a shock.
The lag and difference in magnitude of the pair’s co-movement provide key information. The lag corresponds to the cause of the shock, who is fueling it. The amplitude of the lag, and the difference in magnitude of the co-movement, provide insight to the size of the shock we are dealing with.
Does this mean you have to monitor every financial instrument across multiple markets? Sounds intense!
Yes! This analysis requires high performance computation; it must operate in real-time on all fungible stocks. It also operates through time scales ranging several orders of magnitude – for example we process tick data in sub-milliseconds, and lags ranging from seconds to minutes.
These requirements echo fields of physics like thermodynamics and shock physics, so we took an econophysics approach.
Where is the research now?
We have developed the core of the algorithm and proved its concept using historic market data from the Knight Capital Glitch. We successfully detected the shock, exposed Knight Capital’s involvement, and drew a contamination map displaying the propagation of the shock over multiple markets.
This approach could be turned into a full-scale solution and be used by any financial actors, from exchanges and regulators to brokering companies and hedge funds.
More surprisingly, this technology can be applied to many other sectors, like detecting a digital shock in a network of sensors, or detecting the effect of a geopolitical event on the commodities or forex markets.
Vincent Sebag is a recent MSc. graduate from Imperial College London. You can reach him via email or LinkedIn.
For more information on our work in Finsec please contact our Deputy Director, Dr Deeph Chana, and Honorary Principal Research Fellow, Dr PJ Beaghton.
Article text (excluding photos or graphics) available under an Attribution-NonCommercial-ShareAlike Creative Commons license.
Photos and graphics subject to third party copyright used with permission or © Imperial College London.