Agent-based modeling of financial systems

How do you model the behavior of financial markets in a crisis? The answer does not lie in the equilibrium–based models of mainstream economics. A new book, The End of Theory¹, by Richard Bookstaber, describes an alternative approach.

If not the traditional models, then what?

There's a growing realization that modeling investment markets as if they followed the laws of physics is inadequate for many purposes, especially when analyzing extreme events. While it's relatively easy to point out the problem, coming up with alternative approaches is more difficult. Richard Bookstaber's latest work does both: starting with a broad overview of the reasons that traditional models fall short, he goes on to describe how agent–based modeling may offer a better framework.

The book concentrates on the context of financial crises, a context in which the traditional models of mainstream economics are obviously inadequate. John Maynard Keynes famously observed that "economists set themselves too easy, too useless a task if in tempestuous seasons they can only tell us that when the storm is long past the ocean is flat again."² Almost a century after the words were written, that criticism continues to ring true.

The approach advocated by Bookstaber is agent–based modeling, in which individuals are assumed to act based on what they observe in their immediate environment, and in which their behavior in turn changes the environment, creating a dynamic and evolving system. (A fuller definition is available here.) This approach is the one used in the Santa Fe artificial stock market (described here) and it’s an approach that has proved useful in understanding complex systems such as traffic flow or the motion of a flock of birds.

No definitive answer

Bookstaber argues that a definitive model of the market is unattainable: "markets are systems based on gaining an informational advantage, on gaming, on action and strategic reaction, and thus a space that can never be expressed with a fully–specified model or with well–delineated possibilities." Hence any model will fail to capture some aspect of the full range of market behavior. And in extreme events such as financial crises, those limitations really bite.

Meaningful modeling of markets cannot simply assume away complexity and ambiguity. The agent–based approach instead accepts them and offers a different angle of attack: "The point isn’t to crank out and act on a number. It is to set up a model to see what light can be shed on a real–world problem, and to see if it can fit a larger, intuitive narrative about what is going on." In other words, models should serve as an aid to understanding and not a replacement for it.

Agent–based models aren’t a panacea but, in a sense, that’s the author’s point. As he puts it: "if you’ve got one model, you are a dinosaur." The financial services industry is prone to becoming too attached to a single way of looking at the world, a single way of measuring risk. But it’s a mistake to marry a model.

The book's primary focus is on how regulators and others might have at least a fighting chance of usefully modeling the course of a financial crisis. The agent–based approach seems to offer promise beyond that specific context to the modeling of risk in general, making this essential reading for any institutional investor who is serious about risk management.