Thursday, December 1, 2016

Why don't you just show us why it's better?

If you think you have a better economic modeling paradigm, just show that it's better. If you rely on arguments against other paradigms, you'll end up looking foolish when you have to make the same "mistakes" as those paradigms.

Pedro SerĂ´dio points us to a paper [pdf] that trashes DSGE in favor of agent based modeling. It says a couple things about DSGE having unrealistic assumptions (these are just two examples):



However a few pages later, we get this:


I fail to see the difference between "not the most descriptively accurate" and "unrealistic". Here's the definition of realistic:
re·al·is·tic
adjective
2. representing familiar things in a way that is accurate or true to life
So a good definition of unrealistic is "not representing things in a way that is accurate", so in order for ABMs to deliver meaningful statistical implications, research must often employ unrealistic assumptions.

This not to say I think ABMs are misguided. It's a tool, and tools can be used effectively or not. Basically, my view can be boiled down to a few points. First, I've used an agent based model here (an incredibly simple agent, but an agent nonetheless). But the major issue with complex agents -- basically the one above -- is the same issue with any complex economic model: too many parameters. My agent model was simple. Check out this agent, though:


... and the authors called it "stylized" (and "Mark 0") meaning it's only going to get more complex.

Second, I think it is an interesting question as to whether information equilibrium relationships arise out of agent based models. Some preliminary results say yes, they can. Note that this agent model is also pretty simple (a few lines of Mathematica code).

And finally, there are some basic reasons to expect that macroeconomics is either independent of the details of its agent substrate (macro is a theory restricted a low-dimensional subspace of the million-dimensional agent space) or completely intractable (macro is a theory that requires most of the million-dimensional agent space). I wrote about this more here (with a Socratic dialog).

Instead of arguing that mainstream paradigm X sucks (e.g. X = DSGE), you should just show us that your new paradigm Y works. Show us that Y makes successful predictions. Show us that Y describes the data well (like this). Show us that Y is useful. This is how science works.

No comments:

Post a Comment

Comments are welcome. Please see the Moderation and comment policy.

Also, try to avoid the use of dollar signs as they interfere with my setup of mathjax. I left it set up that way because I think this is funny for an economics blog. You can use € or £ instead.