OK, my article on agent-based modelling of the economy is now out in the Economist – you might be able to get it here, but if firewalls prevent that then here, naturally, is the original thing. And I’m interested that the reader comments don’t seem by any means as adverse to this sort of thing as I’d imagined regular economists would be. Encouraging. Some feel that the economy, or people, are too complex to be captured by any kind of modelling. I don’t believe there is any reason to think that (and some good evidence to the contrary), although it is surely right that we must keep all models in perspective. And we have to remember that social science is the hardest science of all.
For economists, the most serious deficit of the credit crunch may be in credibility. Vocal critics such as Nassim Nicholas Taleb are demanding to know why, when they failed utterly to foresee the crisis – indeed, apparently endorsed the conditions that created it – we should have the slightest faith in their capacity to mend it. And the diametrically opposed views of professional economists on what the remedy should be scarcely commands trust.
Yet there is little sign of discomfort or self-reflection in the citadel of orthodox economic theory. Much the same people, using much the same tools, are guiding economic policy after the crash as before it. Forecasting at the Federal Reserve, for example, is still being done using the so-called dynamic stochastic general equilibrium (DSGE) models that led one of its governors, Frederic Mishkin, to deliver an assessment of the downturn in the US housing market in summer 2007 that now looks grotesquely optimistic. The message seems to be ‘if you don’t fix it, it ain’t broke.’
Mainstream economics has always had its dissidents. But the seeds of change have never before found such fertile soil. Heavyweights such as Joseph Stiglitz and Paul Krugman are calling for radical rethinking. The Institute for New Economic Thinking (INET) in New York, which had its inaugural conference in April, boasts Stiglitz and Amartya Sen on its advisory board, and is bankrolled by George Soros. A hearing of the US House of Representatives Committee of Science and Technology in July called on distinguished witnesses such as Robert Solow to ‘build a science of economics for the real world’.
Critics tend to concur about what is wrong with the tools currently used for macroeconomic forecasting and policy – DSGE models were targeted in the House hearing, for example, while the INET has attacked many of the assumptions, including the efficient-market hypothesis and rational expectations, on which these models are predicated. But there is less agreement about what should replace the old techniques.
The hearing aimed to ‘question the wisdom of relying for national economic policy on a single, specific model when alternatives are available.’ One of the most promising and popular of these alternatives was on display at a workshop in Warrenton, Virginia at the end of June, funded by the US National Science Foundation and attended by a diverse bunch that included economists from the Fed and the Bank of England, social scientists, policy advisors and computer scientists. They explored the potential of so-called agent-based models (ABMs) of the economy to help us learn the lessons of the current financial crisis and perhaps to develop an early-warning system for anticipating the next one. Better still, this non-traditional approach might offer prevention rather than cure: not the false promise of a crisis-free economy, but a way of identifying systemic vulnerabilities and mitigating their effects.
Agent-based modeling  does not assume that the economy can achieve a settled equilibrium. The modeler imposes no order or design on the economy from the top down, and unlike many traditional models, ABMs are not populated with ‘representative agents’: identical traders, firms or households whose individual behaviour mirrors the economy as a whole. Rather, an ABM uses a bottom-up approach which assigns particular behavioural rules to each agent. For example, some may believe that prices reflect fundamentals while others may rely on empirical observations of past price trends.
Crucially, agent behaviour may be determined (and altered) by direct interactions between them, whereas in conventional models interaction happens only indirectly through pricing. This provision of ABMs enables, for example, the copycat behaviour that leads to “herding” among investors. The agents may learn from experience or switch their strategies according to majority opinion. They can aggregate into institutional structures such as banks and firms. These things are very hard, sometimes impossible, to build into conventional models. But in an agent-based model one simply runs a computer simulation to see what emerges, free from any top-down assumptions. As economist Alan Kirman has put it, ABMs ‘provide an account of macro phenomena which are caused by interaction at the micro level but are no longer a blown-up version of that activity.’
Agent-based models are not exactly an alternative to conventional approaches, but a generalization of them: just about any economic theory could be expressed as an ABM, including the DSGE models now used for forecasting by most central banks. While those models are also based on microeconomic foundations, they accept the traditional view that there exists some ideal equilibrium towards which all prices are drawn. That this is often approximately true is why DSGE models perform well enough in a business-as-usual economy.
But DSGE models are useless in a crisis, as even advocates such as Robert Lucas admit. Last year, Lucas responded in this magazine to the criticism that these theories had failed to foresee the credit crunch by saying that such events are inherently unpredictable. All that can be reasonably expected of economic models, Lucas implied, is that they work well in ‘normal’ times. Crashes must forever be anomalies where theory breaks down.
That’s true of DSGE models because their ‘dynamic stochastic’ element amounts to minor fluctuations around an equilibrium state. Yet there is no equilibrium during big market fluctuations such as crashes – one can say that DSGE models thus insist that such events never occur.
ABMs, in contrast, make no assumptions about the existence of efficient markets or general equilibrium. The markets that they generate are generally not in equilibrium at all but are more like a turbulent river or the weather system, subject to constant storms and seizures of all sizes. Big fluctuations and even crashes are an inherent feature.
That’s because ABMs contain feedback mechanisms that can potentially amplify small effects, such as the herding and panic that generates bubbles and crashes. In mathematical terms the models are nonlinear, meaning that effects need not be proportional to their causes. These nonlinearities are absent from DSGE models, but they were evidently central to the credit crunch.
For example, in Virginia Andrew Lo of MIT’s Laboratory for Financial Engineering presented a model of the US housing market, inspired by ABM approaches, which showed how a fateful conjunction of rising house prices, falling interest rates and easy access to refinancing created high systemic risk, amplifying the housing downturn into an awesome burden of debt . And John Geanakoplos of Yale University explained how the leverage cycle in remortgaging – high leverage during booms, low during recessions – can bloom into instability like an out-of-control pendulum, unless carefully managed . The web of interdependencies forged from the buck-passing of risk using complex derivatives may create the potential for propagating nonlinear instabilities analogous to those that crashed the power grid of the North American eastern seaboard in 2003, and are precisely the kind of thing that ABMs are well suited to capturing. Sujit Kapadia of the Bank of England is attempting to uncover and model these network-based vulnerabilities in financial systems ,
While all of these culprits have been fingered in the voluminous post-mortems of the current crisis, there has been barely any discussion of the way nonlinear feedbacks gave them such impact. As a result, the understanding on which any preventative regulation and ‘macroprudential’ strategies might be based is still thin.
Another of the key lessons of the crisis is the role of interactions between different sectors – housing and finance, say. While concentional macroeconomic models can incorporate these, ABMs might be better tailored to each specific sector – for example, including banks in financial markets, which DSGE models do not. In principle, ABMs can include as much of the economy as you like, with all the sector-specific structures and quirks. Indeed, the organizers of the Virginia workshop – physicist-turned-economist Doyne Farmer of the Santa Fe Institute in New Mexico and social scientist Robert Axtell of George Mason University in Virginia – wanted to explore the feasibility and utility of constructing an immense ABM of the entire global economy by ‘wiring’ many such modules together.
What might be required for such an enterprise in resources and expertise, and what might it hope to achieve? One vision is a real-time simulation, fed by masses of input data, that would operate rather like the traffic models now used for forecasting on the roads of Dallas and the Rhine-Westphalia region. But it might be more realistic and useful to employ a suite of such models, in the manner of global climate simulations, which project various possible futures and thus give an aggregated forecast – and show how our actions, laws and institutions might influence it.
In either case, the models would need much more data on the activities of individuals, banks and companies than is currently available. Gathering such information will be one of the key tasks of the US Office of Financial Research instituted by the 2010 Dodd-Frank Act to reform Wall Street. While this plan has raised privacy fears, such data-gathering is no less essential for understanding the economy than are meteorological observations for understanding climate, or geological monitoring to anticipate earthquakes.
And although seismologists may never be able to make precise forecasts, it would be deplorable if they were to shrug and resign themselves to modelling just the regular, gradual movements of tectonic plates and faults. Instead they have developed methods for mapping the evolution of stress patterns, identifying areas at risk, and refining rough heuristics for hazard assessment. Why should the same not be done for the financial system? It won’t be cheap or easy. But to deny the very possibility merely to absolve the conventional models of their severe limitations is starting to look unforgivable.
1. B. LeBaron & L. Tesfatasion, Am. Econ. Rev. 98(2), 246-250 (2008).
2. A. E. Khandani, A. W. Lo & R. C. Merton, Working Paper, September 2009.
3. A. Fostel & J. Geanakoplos, Am. Econ. Rev. 98(4), 1211-1244 (2008).
4. P. Gai & S. Kapadia, Bank of England Working Paper 383 (2010).
Excellent article(s) - both the version here and the highly edited one in the Economist.
I'm a physicist turned Wall St. risk manager and have been using behavioral economics lessons for a while, but have not found ABMs mature enough to use just yet. Glad to hear that they're gaining popularity.
Post a Comment