The Lucas Critique and the hard-money consensus

The Lucas Critique is simple, and it is correct. If you have a model of the economy that works pretty well, and you try to use that model to predict the effects of a new policy, the policy may change people's behavior so that your model no longer works pretty well, thus leading (among other things) to the policy failing to have its intended effect.

The Lucas Critique was applied by Lucas to invalidate many of the "Phillips Curve" models of the 1970s. The idea was that if central banks cause inflation in an attempt to pump up growth, people will start expecting higher inflation in general, and the inflation-growth relationship that held in the past would change. That seems to have been borne out by the events of the 70s. And so people started to think that the Lucas Critique was of great practical importance.

The Lucas Critique still figures prominently in debates between macroeconomists today. Here is Charles Plosser, one of the founders of "Real Business Cycle" theory, criticizing the New Keynesian DSGE models that are popular nowadays:

In my view, the current rules of the game of New Keynesian DSGE models run afoul of the Lucas critique... 

I have always been uncomfortable with the New Keynesian model’s assumption that wage and price setters have market power but, at the same time, are unable or unwilling to change prices in response to anticipated and systematic shifts in monetary policy. This suggests that the deep structure of nominal frictions in New Keynesian DSGE models should do more than measure the length of time that firms and households wait for a chance to reset their prices and wages... 

When the real and nominal frictions of New Keynesian models do not reflect the incentives faced by economic actors in actual economies, these models violate the Lucas critique’s policy invariance dictum, and thus, the policy advice these models offer must be interpreted with caution... 

During the 1980s and 1990s, it was quite common to hear in workshops and seminars the criticism that a model didn’t satisfy the Lucas critique. I thought this was often a cheap shot because almost no model satisfactorily dealt with the issue. And during a period when the policy regime was apparently fairly stable — which many argued it mostly was during those years — the failure to satisfy the Lucas critique seemed somewhat less troublesome. However, in my view, throughout the crisis of the last few years and its aftermath, the Lucas critique has become decidedly more relevant.

First, let me say that I agree with Plosser: Calvo pricing, which is one feature of New Keynesian models about which Plosser is complaining, seems to me not to satisfy the Lucas Critique. (As Plosser points out elsewhere, it also looks to be simply false.) Models in which firms choose when to change their prices are much more desirable. Of course, people are working on these. They are really hard to do, since the decision to change prices can depend on all sorts of weird, hard-to-aggregate stuff, like coordination with other price-setters. This kind of realistic behavior is very hard to shoehorn into the kludgey modeling framework of DSGE, which is why the New Keynesians have been forced to adopt Calvo Pricing as a modeling convenience.

But I digress. What I really want to talk about is the question of how a model satisfies the Lucas Critique.

Plosser says that "almost no model [has] satisfactorily dealt with the [Lucas Critique]." This implies that some of them do. How do we identify the few that do?

In a science, the way you establish the correctness or incorrectness of a proposition is by looking at some sort of real-world evidence. What kind of real-world evidence would allow macroeconomists to know that one of their models is policy-invariant? Well, you could have a central bank try out some policies as an experiment, and see if the model's predictions held up. Or you could call up Blizzard Entertainment and get them to flood Diablo III with virtual gold, or something like that.

Alternatively, you could understand the behavior of individual consumers and firms really, really well, and then find some way to aggregate them that is robust in agent-based simulations. In other words, we could get real microfoundations. This is extremely hard to do, of course.

But this is not actually what macroeconomists do when the subject of the Lucas Critique is brought up. Instead of looking at evidence, what they do is make a judgment call. If all the pieces of a model sort of intuitively seem like things that wouldn't change under different policy regimes, then people nod their heads and say "OK, that seems like it satisfies the Lucas Critique", and they think no more of it. This basically happened with RBC models. "Technology shocks" sound like something that people don't control, and that therefore couldn't change if policy changed. And people assumed that the other features of the model (costless price changes, for example) would hold up under different policy regimes as well. So RBC was thought to have satisfied the Lucas Critique.

But if someone says "Hey, these shocks don't seem structural," or "Hey, I think agents in this model would change their actions in response to policy, don't you?", then there is no consensus, and a vocal group of dissenters continues to say that a model "fails the Lucas Critique". This is what has happened with New Keynesian models. To see how this works, check out this 2008 paper by Chari, Kehoe, and McGrattan of the Minneapolis Fed. They discuss the Smets-Wouters model, widely considered to be the "best" of the New Keynesian models:

The Smets-Wouters model has seven exogenous random variables. We divide these into two groups. The potentially structural shocks group includes shocks to total factor productivity, investment-specific technology, and monetary policy. The dubiously structural shocks group includes shocks to wage markups, price markups, exogenous spending, and risk premia.

How do the authors decide which shocks are "potentially structural"? They don't say. The "dubiously structural shocks" are labeled "dubious" because of a mix of evidence and reasonable-sounding thought experiments that show how these shocks change, or might change, in response to changing economic conditions.

But of the "potentially structural" shocks they say nothing. They simply give technology and policy shocks a free pass. These parts of the model are thus judged to "satisfy" the Lucas Critique because no macroeconomists - or, at least, none who matter! - happen to be concerned about whether they satisfy the Lucas Critique.

In other words, the decision of whether a model satisfies the Lucas Critique is made not by evidence, but by the consensus judgment of macroeconomists.

This is just one more judgment call in macro. Which means one more place where personal and political bias can creep in. In the comments on my earlier post, someone wrote: "I think of the Lucas Critique as a gun that only fires left." What that means is that in practice, the Lucas Critique is generally brought up as an objection to models in which the central bank can stabilize output. Or in other words, consensus has a well-known hard-money bias.

Update: Steve Williamson offers some thoughts, especially this:

[W]hat's the big deal? Noah seems endlessly perturbed that economics is not like the natural sciences. There are no litmus tests that allow us to throw out bad theories so we can be done with them. But that makes economics fun. We have to be creative about using the available empirical evidence to reinforce our arguments. We have to be much more creative on the theoretical side than is the case in the natural sciences. We get to have interesting fights in public. Who could ask for more?

File this under "things I cannot possibly argue with"...