Novel computational models to predict and prevent the next financial crisis
EU-funded researchers have developed novel mathematical models and tools to better prepare financial markets for the unexpected, addressing many of the shortfalls of classical modelling approaches that dramatically failed to predict the 2008 financial crisis.
© Sergey Nivens - fotolia.com
Work in the STRIKE project has resulted in the creation of a Computation Finance Toolbox comprising numerical modelling software that will help researchers, policymakers and financial market participants to forecast trends and risks. To be made freely available for industry and academia, the innovative tools provide more realistic methods of estimating stock- and energy-market prices. The aim is to gauge the contagion effect of financial shocks across markets and countries, and potentially to predict and prevent future large-scale crises.
“STRIKE developed state-of-the art pricing software, now being made available with a user-friendly interface, to tackle the increasing computational complexity of numerical models employed in financial mathematics that are imperative for modern applications in the financial industry,” says project coordinator Professor Matthias Ehrhardt at Germany’s University of Wuppertal.
One of the STRIKE team’s key achievements was to successfully combine different modelling methodologies to generate complementary results and forecasts.
Until the 2008 financial crisis, the most commonly used numerical models were simplified by incorporating assumptions known to be relatively unrealistic, such as the classical linear Black-Scholes-Merton model of financial markets including derivative investment instruments. This approach enabled models to be generated that were easier and faster to use but contained data that was approximate rather than entirely accurate.
As demonstrated in 2008, such models are often unable to forecast extreme events outside of median trends, even though statistical and probability analyses of past financial crises show that such incidents are more common than these classical linear models presume.
Stock-market traders, for example, might use Black-Scholes-Merton models to calculate the probable time within which a security may reach a specific value and what deviation might occur. However, the linear approach is unable to take into account all relevant effects, including irrational investor behaviour. As trading decisions are based on such data, small errors can have major and exponential impacts on the market.
On the other hand, so-called non-linear partial differential equations can generate models based on a much more diverse set of real data, including transaction costs, illiquid market effects and the impact of very large dealers such as central banks, thereby providing a more accurate basis for pricing options.
A realistic, non-linear approach to financial modelling
The STRIKE project focused on developing complementary non-linear extensions of the Black-Scholes-Merton model and entirely new modelling techniques based on mathematical analysis, stochastic simulation and deep qualitative and quantitative modelling of financial market data. The approach extends beyond financial market effects towards a better consideration of the potential social impact of systemic events.
As a theoretical framework, researchers focused on modelling the European financial crisis as the product of contagion and herding behaviour, an approach not possible with classical linear mathematical models. This contagion effect, exemplified in the way the Greek debt crisis impacted Spanish and Portuguese markets, was analysed using real data from companies and banks, helping the STRIKE team to calibrate their models.
Ehrhardt points out that a similar technique could be used to gauge the impact of a country either leaving or joining the EU, or how political instability in one euro-zone country could trigger financial effects in other Member States.
“By using dynamic models of market players, non-linear partial differential equations and stochastic optimal control techniques, our tools can provide suggestions for the avoidance of financial crises on a purely mathematical basis,” he explains.
To address the numerical complexity of these non-linear models, the STRIKE team implemented advanced computational techniques relying on parallel processing via multiple graphics processing units. This is the same technology underpinning recent advances in machine learning, artificial intelligence and data analytics.
As a training network under the EU’s Marie Curie programme, the project has given a significant boost to the careers of early-stage researchers, including 12 PhD students. It has also led to the creation of an ECMI Special Interest Group on Computational Finance, publication of a STRIKE handbook of research outcomes, and the launch of the biennial ICCF - International Conference on Computational Finance series.
“Moreover, a key result has been the establishment and strengthening of new networks, with new links created to companies and academic groups and long-lasting structures established,” says project manager Dr Jan ter Maten at the University of Wuppertal.
“Several EU network and bilateral proposals have emerged from the STRIKE consortium, leading to new STRIKE sub-networks and the launch of a follow-up training initiative focused on applying novel modelling approaches to deal with challenges in energy-trading markets.”