Why agent based modelling is for economics
In this article I give a brief sketch on why I’d choose agent based modelling as the main method for economic models. It gives you an idea of the EEWCO methodology and why neoclassical and keynesian models chose unappropriate methods for dealing with economic questions. The line of reasoning is as follows:
(1) First: The question
What is the question. My question is about money. What are the effects of coordinating our economy with the monetary system we have? What are the effects of alternative arrangements.
This said, it is clear, that I need to look at the economy as a whole.
(2) Scope of model elements
From current existing background knowledge I can now gather elements I want to include in the model: E.g. banks, one bank going bankrupt the others follow or not, a market where a good is exchanged for money, people saving in money.
The takeaway here is, that these elements relate to heterogenous agents: One buys, the other one sells. One bank going bankrupt, others not, etc.
(3) The link to the real world: observations
As I want insights as to how our economy is functioning, the model has to be related to the real world. The link to be used in science is observations. Let’s have a look what kind of observations are available:
- Experiments: Hardly any experiments can be made on societal and the state level for scientific purpose.
- Empirics: Data from past developments. As different countries go along different development paths, some information can be gained by looking at national account statistics. But of course these data do not tell us what would have been, when different politics would have been applied. So empirics is not a substitute for the lack of experiments.
- There is a third source: We know about the rules we have put in place to guide our economic activities. We know about the processes going on in diverse institutions. And we have observations of every day economic activities. I call this kind of observations process structure knowledge.
This means: To develop an elaborated model we must rely on process structure knowledge. Just build models in an arbitrary way and afterwards test them against empirical data or experiments will lead to poor theory as the available observations in that case are poor. We need to build a model of the structural elements we know. And from that we can derive the possible dynamics.
(4) The method to chose
What method is suited to build structural models of heterogenous agents and derive possible dynamics? A simulation approach. system dynamics (sd) might come to your mind. But sd starts from an aggregate perspective and is thus less suited to deal with heterogenity. So we have agent based modeling (abm) to model a virtual economy.
Now let’s do a first test. Let us implement a simple market transaction. Person A has 50€ and goes to a fruit market. Person B is at the fruit market and offers apples. 3€ per kg. A decides to buy 2kg apples and they exchange good for money.
Thinking in abm terms you will find, that including this process in a model is straightforward in principle. It will be quite a bit of work, but it works. Just ad the needed elements.
Now try to expand the neoklassical model with this process. Or the keynesian model. It won’t work. From the above reasoning this means: These models apply a method that is not suited for economic analysis.