In a recent op-ed, Paul Krugman makes the case for simple models in macroeconomics. He wants
Something like IS-LM, which is the simplest model you can write down of how interest rates and output are jointly determined, and is how most practicing macroeconomists actually think about short-run economic fluctuations. It’s also how they talk about macroeconomics to each other. But it’s not what they put in their papers, because the journals demand that your model have “microfoundations.”
I think Krugman is a little too quick to dismiss the need for microfoundations in rigorous macro models. But he is right that simple models are often sufficient. In this article, I discuss the importance of microfoundations. In my next article, I will make the case for a model even simpler than IS-LM: one based on the equation of exchange.
The term “microfoundations” means microeconomic decision-making forms the foundation of the model. For example, it is not enough to say there is a correlation between inflation and unemployment over short periods of time. We need to know what decisions produce that correlation. We need to state precisely how the choices of workers, business owners, capital owners, land owners, policy makers, and other economic decision-makers interact to produce the observed result. What are the agents’ goals? What information do they have? Why do they do what they do? These are the questions a model builder must grapple with when starting from microfoundations.
There are at least two good reasons to insist on microfoundations in rigorous macroeconomic models. First, it is impossible to assess welfare consequences without microfoundations. Suppose a model predicts that, following an unexpected positive monetary shock, real GDP grows faster temporarily. Are the agents in the model better or worse off? Without knowing what those agents are trying to do, what constraints they face, and how they perform in normal times, we cannot know whether the shock improves their lot.
Perhaps the shock fools them into temporarily overproducing. They work more than they would if they were fully informed, forgoing valuable leisure time. They run their machines, and incur the costs of wear and tear, more intensively. And they regret their decisions when they realize they have been duped. If they do well in normal times, the monetary shock makes them worse off. If they tend to systematically underproduce in normal times, the monetary shock might make them better off. Without thinking about the microfoundations, we just observe the change in aggregate variables like output when what we really care about is the welfare consequences for individuals.
Second, we cannot be confident that policies will have the desired effect without microfoundations. For example, suppose you are teaching a course in macroeconomics — a course you have taught for several years. You observe that, in the past, students have done really well on exams. Indeed, even though you maintain high standards and write difficult exams, everyone always gets an A. You might think to yourself: why waste all of those classes proctoring exams if everyone is going to get an A? You could cut the exams and cover more material. Then your bright students would learn even more.
Not so fast! Your reasoning is subject to the Lucas critique. We make decisions subject to constraints. Policies change those constraints and hence might cause the relevant decision-makers to make other choices. Your students are choosing how hard to study given that they will be evaluated on the exams. If they study hard, they are more likely to do well on the exams. If they don’t, they probably won’t. Eliminating the exams thus removes part of the incentive for your students to study hard. Hence, rather than learning more, your students might learn less. In requiring one to consider what the agents are trying to do, what constraints they face, and how they perform in normal times, microfoundations similarly help policy makers avoid unintended consequences.
Let me be clear: I am not opposed to simple models. A simple model often suffices. Nonetheless, one must think about microfoundations, even if those foundations are not explicit in the model. Otherwise, one cannot know how individuals will respond to new policies nor whether those policies are likely to make them better or worse off.