Thomas Sargent, a famous economist, is one of the main pioneers and proponents of Dynamic Stochastic General Equilibrium models - the kind of models that pretty much all macroeconomists use these days (to see how one of these works, watch this). These models have been criticized a lot recently (see here, here, and here for my own criticisms). But Sargent is a believer, and he's speaking up to defend the modeling approach he helped invent:

Here we see Sargent employ the "Well, you're just a dummy" defense against critics who say there's too much math in macro. Sadly, the interviewer pretty much sets him up for this cheap shot, by caricaturing the DSGE critics. Rolnick pretends that DSGE's critics' main problem is with the use of math, rather than which kinds of math and how that math is used. Thus, Sargent gets a free pass.Rolnick: ...Examples of...criticisms are that modern macroeconomics makes too much use of sophisticated mathematics to model people and markets; that it incorrectly relies on the assumption that asset markets are efficient in the sense that asset prices aggregate information of all individuals; that the faith in good outcomes always emerging from competitive markets is misplaced; that the assumption of “rational expectations” is wrongheaded because it attributes too much knowledge and forecasting ability to people; that the modern macro mainstay “real business cycle model” is deficient because it ignores so many frictions and imperfections and is useless as a guide to policy for dealing with financial crises; that modern macroeconomics has either assumed away or shortchanged the analysis of unemployment; that the recent financial crisis took modern macro by surprise; and that macroeconomics should be based less on formal decision theory and more on the findings of “behavioral economics.” Shouldn’t these be taken seriously?Sargent:Sorry, Art, but aside from the foolish and intellectually lazy remark about mathematics, all of the criticisms that you have listed reflect either woeful ignorance or intentional disregard for what much of modern macroeconomics is about and what it has accomplished. That said, it is true that modern macroeconomics uses mathematics and statistics to understand behavior in situations where there is uncertainty about how the future will unfold from the past. But a rule of thumb is that the more dynamic, uncertain and ambiguous is the economic environment that you seek to model, the more you are going to have to roll up your sleeves, and learn [to] use some math. That’s life.

But there are problems with the ways DSGE theorists use math. Let me point out a couple of these.

Problem 1: DSGE math is unwieldy as heck. The math used by macroeconomists has a reputation for being really hard. In a way, it is, but not in the way that, say, the proof of Fermat's Last Theorem is hard. Take it from me: DSGE math is stuff that most physics undergrads could understand with little difficulty. It is not string theory. It is about as "elegant" as Thomas Sargent's haircut. What DSGE math is, is extremely tedious to use. To calculate the predictions of even the simplest DSGE models requires computing power that wasn't even commonly available until the 1990s.

This is bad because it biases DSGE analysis toward oversimplification. The only DSGE models that can be (relatively) easily computed are the simplest ones - the ones that assume government is useless, that businesspeople and consumers always make the right decisions, that everyone has full information, that all contracts are enforceable, that prices are fully flexible, and that the entire economy is composed of just one aggregate "representative agent". Of course, such a simple model can't predict anything useful. But just try to add realistic stuff - heterogeneous agents, imperfect information, learning processes, incomplete contract enforcement, etc. etc. - and you'll quickly find yourself bogged down for literally years in computer modeling hell while your colleagues pump out simple models that explain nothing but appear to contain math just as "hard" as yours.

(And of course, the simple, quickly computable models just happen to be the models that assume that government has no role in the economy, that businessmen are all-wise, and that the economy runs perfectly with no interference. Surely this fact did not escape the notice of the eminent economists who insist that we use DSGE math for every model...)

Problem 2: DSGE math is often irrelevant to DSGE models. Read a DSGE paper, and you'll read a couple pages of assumptions (stated in as stiffly formal and obtuse a manner as possible), and then about 30 pages of technical details. Much of those 30 pages is a tiny variant on the same undergrad-level math used in every other DSGE model, but you end up reading it anyway looking for the one line that's different. Then you read how the researcher computed the damn thing (or, rather, how his Scandinavian co-author computed it), and you trust he did it right, and you look at the results and you realize - Hey! We could predict it would look like this just from reading that one page of assumptions! All those pages of unwieldy formalistic math served little purpose other than to distract you from the unrealistic silliness of the assumptions (and maybe to make the field of macro opaque to outsiders who might try to inject some common sense into the discussion).

(Now, after reading all this, don't you want to take a look at a DSGE paper and see what I mean? I know you do! Check out "Nominal Rigidities and the Dynamic Effects of a Shock to Monetary Policy", by Larry Christiano, Martin Eichenbaum, and Charles Evans.)

Taken together, these problems make it hard to make complex, realistic DSGE models, but easy to cover up the preposterousness of simple, unrealistic DSGE models. As one would expect, this means that the world of modern macro is populated with a bazillion models, one of which can be picked to explain any given phenomenon. In short, we have cargo cult science.

And in fact, Thomas Sargent seems pretty happy with this!!

So instead of insisting that one model explain different phenomena, Sargent et. al. take two contradictory models, each of which explains only one phenomenon, and then use judgment and intuition to patch them together into a policy recommendation. That's like if an engineer said "Well, this physics theory predicts this laser will be red, and this other theory says it'll be blue, so I'm gonna go ahead and say it's gonna be...um...green!"Sargent:I like to think about two polar models of bank crises and what government lender-of-last-resort and deposit insurance do to arrest them or promote them...I call them polar models because in the Diamond-Dybvig and Bryant model, deposit insurance is purely a good thing, while in the Kareken and Wallace model, it is purely bad. These differences occur because of what the two models include and what they omit...Both models leave something out...an important theme of research for macroeconomics...has been about how to strike a good balance.

And economists wonder why nobody trusts them.

## No comments:

## Post a Comment