Interesting article in The Economist (viewable for free) on evolutionary economics. The article mentions that Paul Krugman has been critical of the area but from my reading he’s not entirely opposed especially given he’s written a book about The Self-organizing Economy. His scepticism and how he sees it as being useful seems to be outlined in this talk he gave on What Economists can learn from Evolutionary Theorists. Which seems pretty well balanced even though I think he’s a little harsh on Stephen Jay Gould.
Anyhow I think there is a lot of value in the type of modelling described below, even if it’s not of the explicitly predictive kind. Certainly valuable qualitative knowledge and statistical relationships can be found by such modelling not to mention insight into how relatively simple relationships between groups of individuals can lead to incredibly complex behaviour and structures.
…more unsettling than the ideas are the techniques and tools Mr Beinhocker advocates. He argues that economists should abandon blackboard deduction in favour of computer simulation. The economists he likes do not “solve” models of the economy—deducing the prices and quantities that will prevail in equilibrium—rather they grow them “in silico”, as he puts it.
An early example is the sugarscape simulation done in 1995 by Joshua Epstein and Robert Axtell, of the Brookings Institution. On a computer-generated landscape, studded with “sugar” mountains, they scattered a variety of simple, sugar-eating creatures, which compete for this precious commodity. Some creatures move faster than others, some see farther, and some burn sugar at a higher metabolic rate than their rivals.
Surprisingly, the results of their myopic lives can be gripping. Even simple rules of behaviour result in collective patterns that are impossible to foresee yet easy to recognise. The sugarscape, for example, is quickly beset by a division between haves and have-nots, which bears a strong statistical resemblance to the distribution of income in real economies. These macro-results cannot be deduced from the micro-rules simulators write. Rather, they emerge from the interactions of the creatures in the model, just as “wetness” emerges from the interaction of water molecules, rather than being a property of the molecule itself.
Criticality 
There sometimes seems to be a prejudice for top-down (eg analytic) solutions to problems rather than bottom-up (eg simulations) – no doubt this is due to the role of equation heavy classical physics.
Simulations are very interesting, if only to show that complicated overall behaviour can emerge from simple local rules.
Sacha,
Part of my thesis was on self organized criticality. So one of the things I looked at for quite a chunk of it was simulations on cellular automata and how we can possibly model more complicated phenomena with them. Now one of the more interesting things about them is the idea of universality classes. That you can change the rules, governing the cells behaviour, considerably in some cases, without changing the global behaviour stastitical behaviour. So in essense all these models are the same, which also implies that you can, if you get the right simplifications, make a very much reduced model which may capture the essense of much more complicated system.
So without going so far as Stephen Wolfram and thinking that CA should be the replaced as the foundation for all science, I think there is a vast amount of understanding that can be gained from them and will be in the future, but quite possibly much of this knowledge won’t be quite in the same form that we are used to i.e A deterministic equation governing a relationship between quantities.
should read “CA should replace current theory as the foundation for all science”
Thanks Steve.
“That you can change the rules, governing the cells behaviour, considerably in some cases, without changing the global behaviour statistical behaviour.”
This is interesting… something that comes to mind, is the extent to which “global behaviour statistical behaviour” captures the behaviour of the system, ie, are there significant aspects of behaviour that aren’t captured? This this effectively describe the behaviour of the system globally?
Sacha,
Ususually that is how you characterise the system, by looking at distributions of activity or whatever depending upon what it is. A bit difficult to explain what I mean in short. Perhaps I’ll post about what I mean later.