When Complicated is Better

Yesterday I presented an argument as to why we should take "toy" models seriously, but thinking about it again I realized I left out a significant part of the picture.  While we do use those simpler models I suggested for gravity and human interaction, a sharp observer would point out that the more complicated models are without a doubt better, and since economics is very serious business, perhaps we should consider going through the trouble of using the complicated models we have.

If you said that, you would be wrong though.  Our more complicated models of physics and human interaction are better not because they are more complicated, but in spite of their complicated nature.  They are better because they are more accurate.  Looking however at economics, we largely have two camps: Complicated modelers, often found in the Op-Ed pages of the WSJ, and simple modelers, like Krugman.  Lets compare their track records:

WSJ folks (AKA freshwater school):

• Don't believe in unemployment (yup, more on that in another post, but really, they don't believe in it)
• Have been screaming about inflation "just around the corner" for the last 4 years
• Have been very concerned about the US Government losing its ability to borrow at low rates, any moment now, for the last 4 years
• Were not worried about the housing bubble

Krugman and friends:

• In early 2009 predicted that the stimulus would be enough to bring unemployment down into the 8-9% range, but not enough to "fix" the economy
• Predicted continued low interest rates for government borrowing costs
• Predicted continued low inflation
• Said "OH HOLY S***!" when they noticed the housing bubble
• Think that, just maybe, unemployment might be a problem

First rule of science:  If your model makes bad predictions, it's WRONG.  Microfoundations models (the complicated ones) are pretty darn cool, and they're internally consistant, and interesting to study.  They also happen to make terrible predictions about the economy.  For now, I say we stick with the models that make good predictions.