Hi there, argmin readers! Today’s post is a live blog of Class 10 of my graduate seminar “Forecasting: A Critical Retrospective.” The syllabus and list of past posts are here.
All of my cybernetically inclined friends are into Friedrich Hayek, but my research and teaching keep bringing me back to John Maynard Keynes. These two gentlemen occupy the two poles of the dialectic of the American Experiment. Hayek famously introduced the notion of markets as distributed computers, where prices carry knowledge across the economy. Keynes, with his economic theory of central banking, set the stage for centralized computing of economic variables to govern those markets.
In Keynes’ paradigm-shifting 1936 work, The General Theory of Employment, Interest, and Money, he lays out an oxymoronic formula for central planning in capitalist societies.1 The basics of the Keynesian model are laid out in the appendix of Robert Evans’ paper from this week’s reading. The national economy has four key variables: the amount of investment firms make into the economy, the amount of savings firms accumulate in financial instruments or by paying down debt, the demand for money in the economy to facilitate purchases and sales, and the supply of money from the government. To change these variables, the government can enact various policies. For example, it can invest in infrastructure, raise taxes, increase the money supply, or change interest rates. Keynes argues that the government can dictate the economy’s output—and hence the general welfare of all citizens, who are players in the big macroeconomic game—by properly executing its policy apparatus.
This would set the stage for the subsequent 90 years of democratic capitalist monetary policy. The government has to make policy to ensure a well-run economy. To do this, it has to know the current state of national investment and savings. It also has to be able to forecast what these variables will be if no policy changes are enacted. This last requirement has driven the major investment in macroeconomic forecasting.
Forecasting in macroeconomics is thus primarily a tool of control. Here I mean control in the academic sense: the theory of dynamical systems with inputs and outputs and the design of subsystems to drive outputs to desired targets. Keynes casts the economy as a giant control problem, where the goal is to deftly change policy to ensure a particular state of economic output and employment. It should be no surprise that tools from control, notably the work of Rudolf Kalman on optimal filtering and control of linear systems with quadratic objectives, play a central role in macroeconomics.
Now, to filter and control, we need to make predictions. Keynes didn’t tell us how to generate those predictions. But his disciples, in what is often called “Keynesian” macroeconomic forecasting, write down structural equations of the economy, fit the parameters of these equations using varied means, and then make forecasts directly from the fitted models.
This program of prediction ran into several obstacles. First, it requires mathematical equations that predict all of the aspects of the economy needed to precisely determine optimal policies. Second, it requires a massive measurement system to pin down all the relevant factors in the model. Both were terribly daunting and required a great deal of expert judgment.
Economists want their methods to be “scientific,” since they influence decisions with major consequences. However, with so many variables, so little stationarity in economic conditions, and so much politics involved, building a truly objective and replicable system seems like a fool’s errand. Beatrice Cherrier details some of the typically arbitrary, political nature of macroeconomic sausage-making in this blog post. Evans details the amount of analytical flexibility and expert judgment forecasters necessarily employ in their predictive techniques.
Beyond these nuances of modeling and measurement, however, a fundamental problem of feedback control remains insurmountable. The models have a ton of parameters that are fit to historical data. Different policies yield different parameters. These parameters change when a policy changes. And you can’t predict what the parameters will be after a policy changes. So what on earth are we doing?
The critique in the previous paragraph was levied at macroeconomic forecasting by Robert Lucas in 1976. That macroeconomic forecasting is still an influential practice 50 years later certainly tells us something. Macroeconomic forecasters occupied positions of power and held a sense of civic duty. So they took Lucas’ critique as a challenge, not a reason to close up shop.
I’m not going to hash out the various attempts to build complex, nonparametric macroeconomic models that add ornate complexity while failing to dodge the fundamental problem. Theoretical critiques can carry only so much weight. The fact that the Great Recession was substantially caused by terrible financial policy and inadequate forecasting should have been the nail in the coffin. In 2003, Robert Lucas himself declared that macroeconomics had been a great success:
“My thesis in this lecture is that macroeconomics in this original sense has succeeded: Its central problem of depression prevention has been solved, for all practical purposes, and has in fact been solved for many decades.”
Oops.
Economists didn’t see a problem with the deep instability created by hyperfinancialization. Indeed, though Evans did his ethnographic research on macroeconomists a decade before the crash, his point rings true:
“...[E]conomic policy cannot be based on a quantitative calculus of costs and benefits and must, instead, rest on the considered judgement of a community of experts. Macroeconomic modellers may be those experts, but to expect anything more from them is to expect too much.”
So who should we trust? The Obama administration hoped economists could help get us out of the mess. But 8 years of attempted neoliberal patching of the American system ended in such broad dissatisfaction that… well, you know what happened. We’ve since had a decade of federal unrest as we try to unmoor ourselves from the expertise of economists. While I don’t believe the Biden and Trump administrations have found themselves a suitable alternative, I make no predictions about whose policy advice we’ll be deferring ot next.
The quote from last week’s post was from a rebuttal Keynes wrote to critics of this book.