Finding sense in finance
Dietmar Maringer, professor of computational management sciences, University of Basel
After losing a fortune in the South Sea Company bubble of 1720, Sir Isaac Newton said: ‘I can calculate the motion of heavenly bodies, but not the madness of people.’
As the world sits on the edge of a financial precipice it may be a good time to reflect on the fact that little fundamental research has been done on how financial markets work. Clearly there are non-linearities and very large numbers of variables, but logic would suggest that such situations are better handled with the help of powerful computers than simply the ‘gut instincts’ of traders.
The field of computational finance is well established and has contributed to the development of – as well as the understanding of – banking and money. But what is less well understood is the basic operation of markets. People buy things because other people are buying it or vice versa, creating huge swings in value that have no basis in underlying asset strength. Looking back at various significant events in the financial markets over the years indicates that there are some fundamental laws describing the markets, analogous to the laws of physics – it’s just that nobody knows what they are.
Dietmar Maringer, professor of computational management sciences at the University of Basel, is devising complex models of the markets and using heuristics to derive some kind of structure as to how the markets work. You would think, given the current situation in the world of finance, that the major banks would be beating a path to his door. But as yet, this area of fundamental research is only beginning to make its mark.
Manfred Gilli, professor of econometrics at the University of Geneva, says: ‘Dietmar is a real leader in the field of heuristic optimisation as applied to finance problems. It is a new field and more interest will come. It is not always easy to convince an industry about the trend; they tend to stick to what they have learned.
‘The particular point about these methods is that you do not have to build a model that fits your classical optimisation tools; you can optimise whatever you want. It leaves you free to model what you think is adequate and pertinent rather than the other way around, where you have to stick to a framework or else you cannot optimise it. This is the new approach. In many cases the problem in finance is not to find the absolute optimum; you are looking for a good solution. You always have a lot of noise in the data, so an absolute optimisation makes no sense. This approach allows you to model extremely large problems without the models becoming too complex.
‘People will realise that these methods are useful, maybe more quickly given the current situation, and maybe they will realise that you need a computer in order to do something properly. People will always need to make the final decision, but the computer can provide something on which they can rely. The emotional component of investment decisions is very important.
‘Dietmar has become a full professor at Basel at a very young age, which shows how successful he has been. I think he will realise big things and I fully support him. He is a bright guy and a very hard worker. He is an excellent presenter at conferences and really captures the audience.
‘Dietmar has a feeling for the practical aspects of a problem, but he is not arrogant in telling people that he knows how to do something, what he is providing is a tool, you also need input from the practitioner. You need both components to be successful.’
Maringer was born in a small town near Vienna, Austria. Both his parents were in business, but his interest in his early years lay in sport, maths and science. He studied at the University of Vienna and the University of Technology of Vienna in a combination of economics, computer science and business administration, which was styled ‘business information systems’. He was not sure what he wanted to do, but started to develop an interest in economics at university. Computers were not used much in economics at the time, but Maringer had studied operational research, database management and some artificial intelligence, and became interested in combining the two.
He says: ‘I realised that computing was not just about the computers themselves. There were also benefits that could be applied in economics and finance. Initially I worked in option pricing and some more mathematics areas. Eventually I moved on to econometrics, where my research was on estimating risk that had a data mining element to it. It seemed to me a very good idea to get to the computers to do the number crunching.’
He graduated in 1993 and his skills with computers secured him a job at the University of Vienna’s Finance Department while he continued his studies. He was offered the chance to go to Cambridge University in 1996 to do a masters degree in finance. The Cambridge economics department was very mathematical in those days and Maringer started to look at more advanced mathematical techniques and how they could be applied to economics and finance. It was the beginning of using advanced mathematics in option pricing, and a lot of people were being recruited into banking from a physics or applied mathematics background. ‘Of course in those days we worked on the assumption that everything was normally distributed and we have seen in the last few months that there is no such thing as a normal distribution. One of the most important things I learned at Cambridge was the extent to which we are making assumptions and simplifying the real world.
‘If we assume everything is normally distributed and linear we are probably missing most of what happens in the real world. If you look at the FTSE (London Stock Exchange Index) over the past 25 years there are 17 events that have been more than four standard deviations from the mean, 10 events with five standard deviations and one event at eight standard deviations.’
Maringer returned to Vienna to complete his PhD and started taking an interest in using artificial intelligence methods to solve problems in finance. In particular he used heuristics and evolutionary methods applied to portfolio management. He started attending computational finance meetings and, while at one of these, he met Peter Winker, who had just been appointed professor of economics at the University of Erfurt. He was also interested in the same type of research and he invited Maringer to join him. Winker is a mathematician by background and they have collaborated on many published papers.
After three years at Erfurt he moved to the University of Essex, where there was an established research programme in computational finance and using artificial intelligence to solve business-related problems. In 2008 the University of Basel decided to set up a new chair in computational management science and Maringer landed the position. It was the chance to set up a new research group, which he hopes will be international in composition, in a country famous for its banking industry.
He says: ‘To my knowledge it is the first chair in computational management science. There used to be a chair of computer science within the economic department, but that was around database management and business systems. They wanted to get involved in the newer fields like simulation and optimisation. I am now setting up the team and determining the research agenda here, which will be computational economics and finance.’
The whole field of computational finance was very small when Maringer began his research and there were many people who believed that this kind of research would never really get anywhere. Maringer says: ‘Five or 10 years ago we simply did not have the technology to ask the kind of questions and do the kind of investigations we are doing today. Now it is regarded as a very useful tool. We can rely on standard hardware and standard software packages and still investigate the non-linearities.’
Part of Maringer’s research has involved agent-based modelling of financial markets. The behaviour of markets has always been difficult to model if you assume that people are making decisions for good reasons. In financial markets there are events that create herd behaviour where investors flock to or from a particular investment in the same way that wildebeest move between waterholes on the Serengeti, and usually for similar reasons.
Maringer’s approach is to create a number of independent agents and give half of them a ‘trend following’ behaviour and the other half ‘fundamentals’. The first group makes decisions based entirely on what happened yesterday and the second calculates a price for a security based on knowledge. The behaviour of the market alters wildly according to the proportion of trend followers and fundamentalists.
He says: ‘We can come up with reasons why certain things happen. We have found that extreme events usually happen when the composition of the market changes. If you have a bull market or a bear market, things can suddenly switch when there is an outside influence that changes people’s behaviour. This can often be events taking place in another market. What we are doing is modelling the individuals in the market rather than the market as a whole; and those individuals can change its behaviour. What people tend to do is assume that everyone behaves the same and averages the market, but in fact everyone is different and the average person does not really exist.
‘To understand the markets, you also have to understand the machine trading, which accounts for about 50 per cent of the volume of trades on the London Stock Exchange. One of the explanations for the crash in the late 1980s was that the trading programs were not very sophisticated and the sell orders from one machine trigger sell orders from others.
‘A lot of the hedge funds now have trading algorithms that are their own and so take the place of the human trading decisions. We need to know what effect this algorithmic trading has on the markets and modelling is very useful for this. We can set up a model of different algorithms trading against each other. We can then run some experiments without ruining our stock exchanges. It is not enough to just suspend machine trading when certain events happen, because that in itself completely changes the composition of the market.’
He says that one of the complications of financial markets is that the most rewarding outcome is often obtained by those whose strategy is adopted by the minority rather than the majority. For example, those who are buying a security while the majority are selling stand to gain the most from the restoration of an equilibrium or fair value. He says that this is very difficult to analyse using conventional game theory, but it does lend itself to modelling.
One of the most surprising aspects of this work is that it is not besieged by hedge fund managers offering grants for research. Maringer believes that people are waiting to see if this work is consistent with actual recorded market data first. There have been plenty of false dawns in the past. Secondly there is a self-belief among traders that their experience is far superior to knowledge gained from computer modelling.
He admits that this research is at an early stage and the models are actually still quite small in terms of participants and securities compared to a major financial trading system. But interest is growing and when the markets recover from the latest troubles he expects to find many banks beating a path to his door.
He adds: ‘What we have not quite understood is how things correlate; traditional linear correlations don’t work, because we don’t have a normal distribution. We don’t have a correlation that is constant, because things change over time and sometimes things are strongly correlated and sometimes they are not. But we do know that correlation is stronger for negative events and that it is not symmetrical. For example, when oil prices rise, airline stocks go down, but when the oil price falls again, not all airline stocks rise at the same speed. We have days when all stocks crash in price, but rarely are there days when they all rise and when they do it is not in proportion to their profits.
‘We are getting some interest in our research from central banks which want to understand more of the macroeconomic consequences and the way that things are changing. I get phone calls every now and then from hedge funds and people from banks, but at the moment people from the industry do not really know what to do with these methods. Economists often believe their work will have a dramatic effect on the industry, but in fact it’s usually many years before new ideas become accepted and widely used, then people flock to their office door.
‘Having said that it is very difficult to find out officially what people are doing in the industry. They don’t publish, because if what they are doing is successful they don’t want other people copying it, and if it’s not then they want to forget it.’
Maringer is hoping that he can build relationships with the Swiss financial institutions and international organisations based in the country. For example the Bank for International Settlements HQ is a short walk from his office. He is hoping to inspire not just computer scientists, but also mathematicians looking for new and interesting problems to work on.
He says: ‘Our approach does not circumvent or avoid mathematics, it creates a need for new mathematics that you would not see if you only did pen and paper work. The mathematicians and the computer scientists in economics have traditionally been a bit suspicious of each other, in that the computational people think the mathematicians make too many assumptions and the mathematicians think we are not creating real solutions. Hopefully we can go some way towards closing this difference and use modelling to steer us towards mathematical methods.’