We delve into Neuroeconomics to find the true cause of the credit crunch and ask if we should start treating money as a valuable commodity...
The credit crunch in 2008 was blamed on bankers' bonuses, dodgy mortgage derivatives and algorithmic trading, but evidence is stacking up to suggest that at least part of the answer may lie in incorrect assumptions about how people behave.
In a recent report, the US Financial Crisis Inquiry Commission (FCIC) concluded that the crisis was the result of 'human action and inaction, not of Mother Nature or computer models gone haywire'. But the fact remains that the two types of computer models that are still used by banks and governments to forecast market trends – dynamic stochastic general equilibrium (DSGE) models and econometric models – failed to predict the crash.
These models make their predictions based on aggregate data such as gross domestic product (GDP) and unemployment. As with much of mainstream economics, they assume that everyone is rational with equal access to market information and that markets will naturally reach equilibrium.
Physicist J Doyne Farmer, a professor at the Santa Fe Institute in New Mexico, began his career researching what became 'chaos theory' and later founded Prediction Company, a quantitative trading firm that has now been sold to the United Bank of Switzerland. He thinks that the ability of agent-based models to handle chaotic, non-linear interactions may help us to understand trends and potential crises rather better than the models used today.
DSGE models produce such complex equations for different elements in the economy, including consumption, savings and labour supply, that it is necessary to use linear approximations to solve them. 'Prior to the crisis these equations didn't include liquidity or the housing markets, which are the very components that constituted the crisis,' Farmer points out. Trillions of dollars of mortgage-related securities were packaged up from mortgages that could never be repaid and then the housing bubble burst. 'Once this hit the real economy, people started to be laid off or their salaries dropped, which reduced their demand for houses and house prices were driven down even more,' explains Farmer. 'These are inherently non-linear effects that are difficult to capture with DSGE models.'
Econometric models did no better. These statistical models are built on aggregate data from prior economic behaviour over a series of quarters or years. 'Clearly, their usefulness is limited if you hit a situation that you have never seen before, such as the existence of mortgage-backed securities,' Farmer says. During the crisis, the FRB/US econometric model used by the Federal Reserve Bank indicated that if house prices dropped by 20 percent not much would happen.
Farmer says an agent-based approach would have made it easier to see before the crisis that a significant fall in house prices would cause a big problem. Rather than use aggregate data, an agent-based model tries to represent individual agents, such as people, companies and governments, in order to simulate the cumulative effects of small changes in agents' interactions and to predict situations that haven't been seen before.
In an agent-based model of motorway traffic, say, such detailed interactions might produce the emergent behaviour of a queue forming in the opposite lane of an accident because drivers are gawping at the scene.
Practically, the way to do this would be Monte Carlo sampling, which involves modelling thousands or millions of typical people as in a statistical survey but with enough detail to produce their full profile in terms of income, savings, mortgage, where they are employed and so forth.
You would also need models of real individual companies or representative sample companies, including how much they depend on credit, as well as the effects of government and banking decisions. 'If credit is withdrawn, as happened in the financial crisis, such a model could indicate whether a certain company would have problems and people be laid off,' says Farmer.
Today's computers have the power to handle this kind of approach but efforts have been limited in scale. 'To be productive, agent-based-modelling needs to be tackled on a large scale, rather like weather forecasting,' says Farmer. To this end, he and researchers from the University of Amsterdam, the Catholic University of Milan, McKinsey Corporation, the Potsdam Institute of Climate Change, Ecole Polytechnique in Paris, and the Hungarian software firm AITIA have made a €3.5m research proposal to the European Commission to fund developing a proof of principle.
An emotional response
If the results of agent-based models are to be trusted, any assumptions they make about various agents' decision-making behaviour must match reality. What's clear from the upcoming book 'Minding the Markets: An Emotional Finance View of Financial Instability' by David Tuckett is that money managers are not entirely rational.
Tuckett, a fellow of the Institute of Psychoanalysis and a professor at University College London, spent much of 2007 conducting research interviews with fund managers and found that they developed emotional attachments to shares they bought. Whether they were describing success or failure, emotion was a palpable feature of what appeared to be 'a story of a dependent relationship with a beginning, continuation and ending'.
Functional magnetic resonance imaging (FMRI) is also providing evidence that emotion may influence everyone's financial decisions in important ways.
A few years back, Kevin McCabe, professor of economics, law, and neuroscience, and director of the Center for the Study of Neuroeconomics, at George Mason University in America, worked with colleagues to gather some insight into how people learn from price information when making portfolio decisions in a stock market.
Employing real data from sources including the crash of 1929 and the bubble of the late 1990s, they scanned the brains of people during the decision-making process, and showed them market changes in the portfolio prices over time to see if they would evaluate their decisions in a 'what-if' mode. This is known as the fictive learning signal. 'In fictive or counter-factual learning, agents learn from the outcomes what rewards they could have received, had they chosen differently. One could call the signal an ex-post 'regret' signal,' says McCabe.
It turned out that there was a separable learning system that uses this signal and it largely accounts for over-trading in bull markets. 'The fictive signal's presence generally causes overreaction,' says McCabe, 'as people overreact to recent information, causing them to buy at over inflated prices.'
The price mechanism, McCabe suggests, is a good example of a feedback system where the processes inside the brain can lead to particular decisions in markets and vice versa. Because the processes are quite precise and computationally identifiable, he says they can be added as constraints to the usual models of economic behaviour, as well as within agent-based models for larger simulations of events within a total market.
More recently, Charlene Wu and Brian Knutson at the SPAN Lab (Symbiotic Project on Affective Neuroscience) at Stanford University in California have found that 'skewness' (large and asymmetric but unlikely outcomes) seems to influence financial decision-making. 'Skewness' preferences are not considered by many finance theories but it seems that certain types of positively skewed gambles – like the lottery – attract almost everybody.
People's propensity to follow the crowd is a better-known phenomenon implicated in stock-market bubbles and bank runs that may not be as rational as once thought. 'In economic theory, the idea consistent with the view of rationality is that 'herding' is Bayesian-learning. In other words, people are learning from the actions of others because of a probabilistic decision, because the actions of others are informative,' explains Michelle Baddeley, an economist at Cambridge University who has been working on locating a neural basis for herding with Wolfram Schultz, famous for his work on reward and dopamine.
Their findings, published in Frontiers in Human Neuroscience, suggest herding has an emotional basis. Having first correlated 19 people's tendency to herd with certain personality traits, such as empathy and impulsivity and found individual differences, Baddeley and her colleagues used FMRI to scan their 19 subjects as they made decisions on whether to buy stocks after observing others' buying decisions. It turns out that by following others, areas of the brain associated with reward processing – mainly the ventral striatum, which is rich in dopamine neurons – light up in the scanner.
Significantly, they found that the signal in the ventral striatum did not track non-human, non-social control decisions. 'We are still analysing our data from the imaging experiments. But clearly herding has its own reward, i.e. people are stimulated by following others' behaviour,' says Baddeley.
Money, it's a gas...!
Market models assume people function like rational robots. But what if human nature doesn't make nearly as much difference to the big-money picture as we were beginning to think? What if we need to start treating money like energy?
Perhaps the most thought-provoking of the efforts to engineer reality into economic models comes from Victor Yakovenko, professor of physics at the University of Maryland. A specialist in condensed matter theory, for the last decade Yakovenko has also been applying statistical physics to the distribution of money in an economy.
Yakovenko has become known for the simple idea of the conservation of money. 'Unlike physical objects such as cars, houses, computers and food, money is not consumed and is abstract – when transactions are done, digits are transferred between computer accounts,' he says. Money can be studied as the 'informational layer' of the economy.
Models of financial markets tend to ignore whether the players have enough money to buy the assets they want to buy, he says. 'The usual response is that market players can always borrow money if they need to.' However, borrowing cannot go on forever. 'Ignoring constraints due to money conservation, debt liability, and demographic fluxes of money led to spectacular failure of 'financial engineering' during the recent crisis,' he says.
Yakovenko has plotted the distribution of money in various countries and found that they all show a lower class distribution that tends to be extremely stable and an upper class that is dynamic and follows a power law, known as the Pareto law. Having plotted the IRS income data from 1983 to 2007 in his recent paper in the New Journal of Physics (2010), he found that the fraction of the total income going to the upper tail is highly correlated to market bubbles.
'In 1983, only 5 per cent of income was going to the upper class, then from 1985 to 1995 it was about 10 per cent but then from 1995 to 2000 it shoots up to about 19 per cent during the dotcom bubble. Then inevitably the bubble crashed and the fraction of income went back down to around 12 per cent. The housing bubble started in around 2003 and the upper class proportion of income went to above 22 per cent in 2007, an all-time high.'
One of the origins of the bubbles is debt, he says. 'Because of the conservation law, to become super-rich on the positive side you need to put lots of other people into debt on the negative side. During the housing bubble, money was lent to people with nothing and the loans were then pushed onto someone else.'
Yakovenko describes the chopping up of loans until no one has any idea what belonged where, in terms of chaos theory. 'This process was heralded as risk-spreading and, thus, stabilising the system. However, from a physics point of view, this process destroys information by disconnecting the informational layer from the physical layer (i.e. lenders and borrowers) and thus ultimately destabilises the system. In the end, governments and central banks have to step in and inject new money into the system to cover losses of failed banks, because only the central authority has the monopoly of issuing money.'
Doyne Farmer thinks the 'conservation' theory produces interesting predictions about the distribution of income that provide good reality checks that any theory about income and wealth should satisfy. 'For instance, the fraction of wealth the richest 1 per cent of the population owns differs from one country to another and through time. Nobody knows why. Potentially with agent-based models we can answer questions that do not have good answers in economic theory at present, such as: how does the fraction depend on the tax-code, or inheritance taxes, or birth rate?' says Farmer.
By combining demographic data with the principle of money conservation, Yakovenko thinks it may be possible to predict, to some degree, the macroeconomic behaviour of the economy, based on the big flows of money in and out of the stock market as the baby boomers retire.
The news, though, is not good. Yakovenko thinks the market will collapse completely and precipitously, based on demographics. 'According to the economic forecaster Harry Dent, we will have to wait until 2023, when the children of the baby boomers reach 45 and start to invest again,' he says.
Brace yourself for the next 15 years or so. *