"Agent-based modeling offers a viable alternative to calculus-based representative agent models in economics", is something I have said to people in the past. In response, I often get blank stares, cries of indignation or just plain 'WTF?'. So what's the deal?

There are several reasons why individuals are choosing to adopt agent base modeling. Let's highlight a few of them. First, the concept of the representative agent, or applying methodological individualism always and everywhere, can be deeply problematic. Who exactly should this 'representative agent' represent? Should we not instead have a distribution of heterogeneous agents interacting in manners that replicate to a greater or lesser degree the real world, either establishing their own social networks or using geographical lattices for interacting with other agents? Should we not have something more realistic than random matching?

The next question that we should ask is, do agents optimize or do they adhere to rule-based behaviour? Or is it a mix of both contingent on knowledge or information? Leijonhufvud parodies it well in his 'Three Items on the Macroeconomic Agenda', "The economist of today expects to see a solitary representative agent, under the mathematical spotlight on a bare and denuded stage, asking agonizing questions of himself: `What does it all mean?, [...] or `I know I have optimized, but is that all there is?'" The problem of optimizing agents has been examined exhaustively over the past century, with origins in the work of Herbert Simon and others.

Another favourable aspect of ABM is that we can use it to model economies as complex adaptive systems (CAS) and to use the models to observe the patterns that emerge given different sets of parameters. This allows the researcher a greater ability to understand the emergent patters and the macro-phenomena, rather than focusing solely on the atomistic level of the individual. So what is a CAS? To quote John Holland (from the wikipedia site,

"A Complex Adaptive System (CAS) is a dynamic network of many agents which may represent cells, species, individuals, firms, nations) acting in parallel, constantly acting and reacting to what the other agents are doing. The control of a CAS tends to be highly dispersed and decentralized. If there is to be any coherent behavior in the system, it has to arise from competition and cooperation among the agents themselves. The overall behavior of the system is the result of a huge number of decisions made every moment by many individual agents."

As can be seen the ideas of cooperation and competition of agents in order to produce possibly unforeseen or unintended macro outcomes fits well into the economic paradigm. For those interested in a good conversational introduction to complexity theory, focusing on the individuals involved, I'd suggest M. Mitchell Waldrop's Complexity. Alternatively another good, and more up to date book is Scott E. Page's Complex Adaptive Systems.

What do practitioners of ABM do? Well they set up computer programs with 'agents' (little programs or automata) interacting in a 'landscape'. A recent and famous example is that of Josh Epstein and Robert Axtell's 'Sugarscape' where they replicated unequal distributions of wealth, seasonal migrations, pollution, sexual reproduction, combat, transmission of diseases and nascent forms of 'culture'. Eric Beinhocker offers a good summary of sugarscape in his book The Origin of Wealth (pp. 80-97).

Let's take a brief look at Sugarscape. We have:

1. a notion of physical space (NSEW)
2. a source of energy (or base resource) with heterogeneous distributions across the landscape
3. the terrain is differentiated: hills, mountains, valleys, etc.

What do the agents do? They:

1. look for sugar
2. move to sugar
   
3. eat sugar

Agents must move around and eat sugar, or they die. Later they were programmed to do more complex things, but just think about these for the moment. Agents also have 'metabolisms' which use up sugar at specific rates. They also have different levels of 'vison' (how far on the lattice they can see). These two variables, 'metabolism' and 'vision' constituted the characteristics that were heterogeneously distributed in the population. Having good vision (seeing further) and good metabolism ('burning' sugar slowly) would constitute evolutionary fitness in most landscapes. But what happens if you are born with great 'genes' but there are no resources around you, i.e. you just happened to be unlucky enough to end up in a resourceless area? You die. The novelty of Sugarscape was that it replicated so many real world phenomena, from 'classed' societies and unequal distributions of wealth where 'luck' really helped, to trade and markets and several other interesting phenomena. I'd highly recommend looking at the Beinhocker book (as I have before). Other practise of ABM include uses of artificial neural networks or evolutionary computation (rooted in evolutionary game theory but don't confuse the two).

As I've outline above ABM can bring together behavioural criticisms of economics and the physics-based theories of complex systems. Both of these can and do provide relevant and worthwhile avenues of research for future economic and sociological research. Some time this week I'll assess the recent BPS article on behavioural economics and give my thoughts on the discussion between Pete Lunn and Tim Harford. I thought to preempt some of this discussion with a brief review of ABM prior to discussing the relevance of behavioural economics.