Agent Based Models
I have some interest in agent based modeling. Most of the narratives of economic logic are agent based, yet very few of the models are. I find this to be one of the major annoyances (I won’t go so far as to call it a shortcoming) of economics as I have found in my courses and research. It may very well be that agent-based modeling is a waste of time (but then, you won’t know until you find out, at which point it will be ruled out, which wasn’t a waste of time!), but it is at least very intriguing.
I believe Rubinstein has an issue with these models (he refers to evolutionary models, which I believe is a reference to the simulated nature of the agent-based models). He says “In contrast, evolutionary models treat agents as automata, merely responding to changing environments, without deliberating about their decisions” in the opening of his book Modeling Bounded Rationality. This is an intriguing shortcoming, since a major purpose of going the evolutionary route is to get an ideas of the limitations of deliberation. Unfortunately, I do not know enough about either method to comment more deeply.
Anyway, this came up due to Gabriel’s thoughts on the subject:
Agent-Based Models
The topic of agent-based models came up in the comments to one of Robert V.‘s posts, A Reader Of Mirowski And Georgescu-Roegen?, and since commenting is cheap, here’s my comment.
There are excelent ABM papers out there, one of which is Gintis’ The Dynamics of General Equilibrium, a paper I read a while ago and I will certainly revisit in the future.
The problem with ABMs and why they’re not more popular with the profession has, I think, nothing to do with them being radically different than current models and numerical tools (they’re not really) and nothing to do with them overturning current wisdom (sometimes they do, sometimes they don’t). — The problem is that this is simulation and I, for one, am not sure what I’m supposed to do with simulations.
Simulations are repeatable (if the random number generator is explicitly seeded) so this passes basic rigor, so to speak. — The problem is that simulations don’t yield analytical results.
When economists think of proving something, they think of proving a theorem and/or rejecting some empirical hypothesis. Simulations don’t fit that frame of mind.
A simulation can act as a counterexample to some claim, i.e. that some process converges, or something like that, but it can’t be a counterexample to analytical results without stating the problem differently. (There’s a ABM-based “proof” out there of a “competitive” market converging to monopoly pricing, I forget by who, but that’s just improper use of terms.)
What you get from simulations/ABMs/numerical examples is that for some seed, some parameter values and some initial conditions something interesting happens (and whether that’s sensitive or not, by some definition of “sensitive”). — It’s not obvious what we should do with that, I think. Hence the unpopularity.
