Thursday, February 20, 2014

Lars P. Syll — Macroeconomic challenges

It would be better to just face the truth — it is impossible to describe interaction and cooperation when there is essentially only one actor.
Macroeconomic challenges
Lars P. Syll | Professor, Malmo University

The assumption of representative agents is atomistic in the sense that atoms are indistinguishable elements of a homogenous system that is characterized by timeless invariant relationships that can be captured as mathematical functions — as it physics. However, social systems are not at all like physical systems, are they? 

So why assume atomism in the first place, when it is bound to fail. The only realistic answer I have seen is methodological convenience. Such models are tractable. Modeling complexity is hard.


Ryan Harris said...

That might be a social science way of looking at it. I can't think of any area of study that would use representative agents in the way that they are in economics. At an abstract statistical sense the real range and distribution of a data set are eliminated, in a logical sense you are creating a fallacy of false choice, or false dichotomy. In a mathematical sense you are compressing the data, almost a fourier transformation that is reasonable but is not realistic. I don't care what your area of study, it plain idiocy unless you have a reason for doing it and are very, very careful and constrained about claims/conclusions that you make from your findings, patterns or whatever you studied.

Tom Hickey said...

I can add justifying simplified static models for student use as long as it is explained that the model is highly stylized and should not be taken as representations of a complex system like the macro economy.

Even in micro, the stylized models don't work, as Radford explains in the post I put up today.

But there is no harm in using such models for student use or to illustrate something about the history of economics. Relying on them for policy formulation is either stupid, or intentionally biased.

Tom Hickey said...

BTW, social science other than econ is moving on to agent-based modeling, acknowledging complexity. Of course, some economists are, too, but the mainstream is resisting. See Agent-based computational economics and Agent-based computational macroeconomic models. ABM has problems, too, but at least it recognizes multiple agents with different strategies instead of being founded on preferences that characterize a "rational" atom.

Ryan Harris said...

There are good uses for agent based modelling, especially in trying to work out iterative patterns, but the conclusions and information we get from the model tell us about the arbitrary set of rules we picked for the agents we put in the model. If I create a sun agent, and then set the rules to be when the agent is happy, the sun rises, and when the agent is sad the sun falls, and we feed our observations into the model, it doesn't tell us a heck of alot, because there isn't really a sun god. We can learn a bit about the pattens in moods of the sun gods but we are stuck thinking about the real pattern couched using a sun god. The rules of the model were bunk to begin with. All economic agents are about as goofy and flat a character as the sun god.
Where the ABMs would seem to be most help for us is if we want to figure out what ifs that we all agree would never occur. We could say what would happen if everyone acted in this normal way but did it for a long time... burned CO2 until the world was baked or something. How would markets act? Something like that, maybe could help us understand something new? Plenty of good uses.

Brian Romanchuk said...

The distributional effects are a much bigger effect that they realise. Since all business sector income is utlimately distributed back to the household sector via wages or dividends, the household sector in these models would be indiffferent to wages or profits in the actual solution. This means that relative prices have no effect on the solution, and the Euler equation they use does not apply. Since the mainstream do not solve the models, they do not realise that they are not solving them correctly.