Monday, July 5, 2010

Ramblings on Micro-foundations, Part I...

This is the first of many future posts on the topic of proper micro-foundations of macroeconomic models...

A major critique of the complex systems approach that I hear repeatedly from economists is that the models lack micro foundations.  This is patently FALSE.  What they lack, typically, but not always is what I term Lucas-style micro-foundations.  This type of micro-foundations can be very useful (particularly as a normative base-line in certain circumstances), but the idea that all macro models MUST have this style of micro-foundation has more to do with the model preferences of economists and is not (in my opinion) grounded in theory.  I would argue there are two types of macro behaviors:

Type I: Macro behaviors that are well described as simple aggregates of the behavior of micro-automata.  Here interaction effects between the agents might not be important.  Thus modeling interactions as impersonal exchanges that happen only through prices is likely quite reasonable.  In a sense if you correctly specify your model of micro behavior, then you can simply "scale" it up and you will have a description of the macro system behavior.

Type II (Emergent): Macro behaviors that are not well described as simple aggregates of the behavior of micro-automata.  In this case the macro behavior of the system is highly dependent on the interactions between agents (i.e. strategic interactions effects, network effects that might engender positive/negative feedback loops, cycles, etc).  If you were able to correctly specify a model of the micro behavior, you can't simply "scale" it up and achieve an accurate description of the macro system behavior.

Now most Lucas style micro-founded macro models (in my opinion) implicitly assume type I behavior (this would include DSGE models as a high profile case).  When I mention behaviors that are irreducibly macro, I meant that they are Type II (Emergent).  If your goal is to describe the properties/dynamics of the macro system, then in this case you can not achieve it by reducing your model to the micro level and then simply "scaling" it back up (a la Lucas style DSGE models).

Another related criticism of the complex systems approach is that the micro level behavior heuristics assumed in these models are ad hoc.  This is sometimes true and sometimes false.  If a scientist chooses to model micro behavior by assuming that agents follow some decision rule that is well documented in the sociology or social psychology litereature (such as a decision rule documented by Kahneman and Tversky, etc) then is this ad hoc?  I would say no, but I imagine others might disagree.  Clearly a scientist who has documented some type of emergent macro property would want to check to see if this property is simply an artifact of the assumed micro behavior.  If this is the case, then this model is clearly less useful from both a scientific and policy perspective than one in which the emergent behavior is robust to different assumptions on micro behavior.