Why macro forecasting is difficult
At first glance, this post may seem a rather pessimistic take. When people ask me what’s going to happen to the economy, they do not wish to be told that it’s hard to forecast macro variables. Nonetheless, I see this as a hopeful post. Writing it has actually made me more optimistic about forecasting.
Before explaining my theory, let me review two analogous but clearly different theories, the Efficient Markets Hypothesis (EMH) and the Lucas Critique:
The EMH says that’s it’s hard to predict asset prices, because current asset prices already reflect the expected impact of publicly available information. Thus knowing that Tesla car sales are rising fast and that governments are pushing green energy doesn’t help me predict the rate of return from investing in Tesla stock. The market has already priced in this information.
The Lucas Critique says that when policymakers try to take advantage of the historical relationship between the policy instrument and a policy goal variable, the relationship will shift, and become unstable. Thus if you notice that there is a positive relationship between the money supply and employment levels under a gold standard, and then artificially increase the money supply in order to create jobs, the relationship will tend to break down. Workers will begin demanding higher wages in anticipation of higher future inflation.
Neither of these theories precludes the ability of me or anyone else to forecast macro variables. I’m not a policymaker, and thus the Lucas Critique does not apply to me. And the EMH doesn’t preclude the possibility of being able to predict rising inflation or recession in 2023, as those forecasts might already be embedded in asset prices. Nonetheless, these two well-known theories are somewhat analogous to the hypothesis that I’m about to offer, which is built on three assumptions:
1. Much of what we are asked to predict represents policy failures. Not all predictions; it is certainly possible to predict a healthy economy. But the predictions that people value most are policy failures, such as a surge in inflation or the timing of the next deep recession.
2. We often forecast by looking at past patterns in the data. We say, “The last time X happened, the economy experienced Y.” Importantly, “X” is almost always public information.
3. Policymakers are generally trying to prevent policy failures, and rely on public information.
Each time a major airliner crashes, investigators retrieve the black box and try to figure out the cause. If a component has failed, they may ask airlines to replace that component with something more reliable. If it was pilot error, they may inform pilots of what went wrong and how to respond to the situation more effectively next time. As a result, it’s really hard to predict what will cause the next major airplane crash.
Much of macro forecasting consists of little more than economists observing something like: “In the past, I notice that macro shock X was often followed by policy failure Y.” If policymakers never learned from their mistakes, then this would be a useful method of forecasting the macroeconomy. But policymakers do learn from their mistakes. They don’t learn as quickly and as effectively as I would like, but they do learn. And that learning (combined with the subsequent adjustment in policymaking) makes macro forecasting much more difficult than otherwise. Indeed, this point holds even if policymakers learn the wrong lesson—say by overreacting where in the past they under-reacted. Any adjustment in policy based on learning makes forecasting much more difficult than otherwise.
In my view (and here’s the optimistic part of the post), this gives us two useful avenues for forecasting.
1. Not all bad outcomes reflect future policy mistakes. Some bad outcomes might end up being a lesser of evils, given previous policy mistakes that had already occurred. For instance, as the Great Inflation was getting underway (due to excessive monetary stimulus), the Fed briefly adopted a tight money policy during late 1966 and early 1967, which slowed NGDP growth to about 5%. Fearing a recession, they then backed off from that policy and NGDP growth surged and averaged over 10% over the next 14 years. In retrospect, they should have continued with the monetary restraint (say 5% NGDP growth) even if it resulted in a mild recession during 1967. The alternative (the Great Inflation) was much worse.
Today, the Fed needs to slow NGDP growth down to no more than 4%, perhaps a bit less. Doing so increases the risk of recession, but it is still worth doing. That fact is what allows so many people today to confidently forecast a recession, whereas it is much harder to forecast recessions during periods when the economy is in equilibrium with low inflation and high employment, and any recession would represent a policy error. Thus bad outcomes can be forecast when they represent optimal policy—the lesser of evils in addressing an already bad situation.
2. Another way of forecasting bad outcomes is to look for evidence that policymakers have not learned the right lessons. In 2020 and 2021, Bob Hetzel looked at the rhetoric coming out of the Jay Powell Fed and noticed disturbing parallels with policy that produced the Great Inflation. The Fed did learn some useful lessons from the mistakes made during the Great Recession of 2007-09, but overreacted because it ignored the lessons of the 1960s and 1970s.
To summarize, any attempt to forecast bad macro outcomes involves a combination of two types of analysis. First, ascertaining when bad outcomes are almost inevitable, because they represent the lesser of evils (often due to previous policy mistakes.) Second, trying to figure out what sort of mistakes a given set of policymakers is likely to make.
But we also shouldn’t ignore the pessimistic side of this analysis. History almost never plays out in the same way twice as policymakers are always learning from past mistakes, even where they learn the wrong lessons or only a portion of the true story. As we try to forecast the timing of bad outcomes for the economy, Jay Powell is trying to make us fail. And he has very powerful tools at his disposal.
No amount of progress in the science of macroeconomics can solve this problem, because it’s essentially an arms race between forecasters and the Fed.