My last post supporting the use of nonlinear models (“You Do Need A Weatherman”) generated some thoughtful responses, mainly along the lines of this post by Ari Andricopoulos entitled “A View on the Economic Model Debate from a Non-economist (but someone who builds models for a living)”. The basic argument is that a full nonlinear model of any significant economic process would be too complicated, and that it was better therefore to stick with tractable linear models, while keeping in mind that the real world is nonlinear:
I build models with data for a living, and I am acutely aware of the problems with using non-linear models to make any sort of accurate predictions – even with huge volumes of data to calibrate it with.
It is not that the systems are linear. They are hugely complex. My problem is that they are too complex to model even with non-linear models. My belief is that linear models do have to be used but with a full understanding of the non-linearity of real life. Also, the whole building of macro-models from first principles, based on ‘rational’ agents, is a complete joke of a way to design a model that is supposed to be used in the real world.
While these points have some validity (especially Ari’s jibe at “rational agent” models), this criticism approaches complex systems from the wrong end—the “complicated” as opposed to “complex” end. A core lesson from complex systems analysis (dating right from its first discovery by Poincare back in 1899, and manifest in the first simulation of a complex system by Lorenz in 1963) is that a simple system can demonstrate complex behaviour. And a simple complex system—yes, I know that sounds like an oxymoron, but bear with me—can tell you most of what you need to know about a complicated complex system.