Central banks will need to refine their crisis models
Current equilibrium models fail to predict business cycles correctly, claims University of Amsterdam professor Cars Hommes. Non-linear models incorporating human behaviour are the future.
Cars Hommes has a predilection for modelling complex processes. Hommes studied mathematics at the University of Groningen and obtained his PhD with a thesis on chaos theory applied to the economy. In 1992, he switched to the University of Amsterdam (UvA). Since 1998, Hommes is attached to the Center for Nonlinear Dynamics in Economics and Finance (CeNDEF), a research group for complex economic systems. He is currently its director. Hommes teaches master’s students of econometrics at the Amsterdam School of Economics (ASE) and PhD students at the Tinbergen Institute.
At the beginning of 2018, Hommes was awarded a four-month Wim Duisenberg Fellowship at the European Central Bank (ECB). He completed a research paper on macroeconomic behavioural models and experimental economics. The paper will soon be published in the prestigious ECB Working Paper Series. The fellowship is named after Wim Duisenberg, a Dutchman who, at the beginning of this century, was the ECB’s first president.
Non-linear complex systems are a central theme in Hommes’ scientific research. ‘Standard models take as their point of departure that markets always tend to reach an equilibrium as economic actors are aware of the fundamental equilibrium price and acting rationally. The reality is entirely more complex. People are not rational individuals, but respond to each other and learn from each other’s behaviour. This means that rather than there being one single equilibrium, the situation can switch from one equilibrium to another.’
The credit crisis ten years ago made us understand that the financial supervisors have substantially underestimated the so-called system risk – i.e. the extent to which problems in a group of banks or in a sector will spread to other banks or sectors. ‘It is not fair to conclude that the standard models used by central banks failed to predict the credit crisis as it is simply very difficult to predict such a crisis. But what is relevant here is that these models had not even incorporated the possibility of a crisis of this magnitude, while behavioural models do.’
To date, the ECB and other supervisors, such as the Dutch central bank (DNB), base their policies primarily on traditional models. ‘But we know that the standard model is incorrect. I am still worried about this system risk, even though the banks have been placed under stricter supervision and even though the economy has recovered. The ECB still does not have a good model that describes the system risk of all European banks, and is not putting much effort into developing one.’
While crises are an intrinsic part of business cycles, Hommes considers that extremes must be avoided. ‘The crisis of ten years ago was unprecedented and has had a huge damaging impact. Substantial groups of people have been unemployed for a long time or have been forced out of the labour market, especially in the southern European states.’
Meanwhile, Hommes notes an increasing interest in non-linear models, as is illustrated - among other things - by the ECB awarding him the Wim Duisenberg Fellowship. ‘The interest in complex systems is growing, especially at the central banks in the UK and Canada, but the ECB has begun to show a growing interest as well.’ The DNB, too, has joined in. ‘I worked with DNB to investigate the way the prices of properties have developed in various countries. We used a non-linear model which turned out to be able to explain bubbles.’
Much time will pass before financial supervisors will commit themselves to the transition that Hommes believes is needed. ‘I believe in the practical application of behavioural models. The central banks currently employ thousands of people who took received their PhD on the standard theory. Instead, they should employ hundreds of people who are trained in behavioural models. With the two approaches being complementary, a full transition is not needed. But even so this will take at least ten years.’
The fundamental difference between traditional and behavioural models is their approach to expectations. ‘The standard model starts from rational expectations and from prices always moving to their real equilibrium. Behavioural models work with expectations based on learnt behaviour.’
Hommes believes traditional models are too narrow. ‘They only have eyes for the rational equilibrium. In behavioural models, the ‘almost self-fulfilling equilibrium’ plays an important role. While this type of equilibrium is not precise, it does by and large reflect what people expect to be the equilibrium. This is the reason why several equilibriums can co-exist.’
So-called positive feedback can cause the path towards these equilibriums to fluctuate wildly. ‘When speculators believe that prices are rising, demand will rise and prices will rise too. So, equilibriums emerge whose rationality is bounded. Such models allow us to explain and predict bubbles and extreme crises very well.’
Experiments have increasingly become a part of behavioural economics. ‘Physical processes can be tested quite well, but in economics science, carrying out experiments is a difficult story.’
Hommes uses the laboratory to set up experiments that explore the influence of human behaviour on the price formation process. ‘This involves dynamic economic models, with the price expectations being determined by a so-called feedback system. This means that the current price is a function of tomorrow’s average prediction. So, the fundamental equilibrium price plays a minor role in the price formation process.’
Hommes set up groups of students trading shares. They were provided ex ante with information on the future dividend and market interest rates, data from which a certain equilibrium price can be deducted. A trading computer will intervene when the price formation diverges too much from this equilibrium price.
‘Although shares have a certain fundamental value, every experiment has its own dynamics.’ In one experiment prices may fluctuate wildly around an average price, in another they may remain close to the equilibrium and in yet others prices may swing up and down nervously at first before gradually slowing down. ‘The structures that emerge reinforce each other and the number of patterns that can be observed remains limited. The size of the group is irrelevant’, says Hommes. We’ll need more research to find out what exactly makes these patterns emerge.’ In the future, the lab can also be used to test policy analyses.