“Potential” output forecasts are actually worthless

Towards the end of 2000, at the peak of America’s irrational exuberance, staff economists at the Federal Reserve Board presented the following forecast:

Structural productivity is estimated to have risen 3.2 percent in 1999 and 3.5 percent this year and then to accelerate to 3.6 percent in 2001 and 3.7 percent in 2002. Accordingly, with the contribution of labor input assumed to be advancing at a steady rate, our projections for the rate of expansion in potential output move up from 4.3 percent last year and 4.6 percent this year to 4.7 percent next year and 4.8 percent in 2002.

A few months later, the boffins at the Congressional Budget Office released their forecasts for 2002-2011. They were only slightly more conservative than the Fed, expecting the “potential output” of the private sector to grow at an annual average rate of 3.8 per cent and for the overall economy to expand at a yearly pace of 3.3 per cent:

There are only two ways to interpret these forecasts. Either economists were ridiculously optimistic at the beginning of the millennium or the US economy spent the first decade of the 2000s operating far below its “potential”.

Standard theory* suggests the first interpretation is closer to the truth: technocrats at the central bank and the budget office were deluded into believing the anomaly of the late 1990s would continue indefinitely.

Their error seemed plausible at the time because both institutions had spent the second half of the 1990s consistently under-estimating the economy’s ability to grow.

Back in 1996, for example, the CBO expected “potential output” would only grow at a yearly average rate of “about 2.1 percent” from 1997-2006. That was about a percentage point slower than what actually happened, and about two percentage points slower than the average growth rate in 1997-2000. Similarly, the Fed’s staff economists thought potential output had a “2 percent trend” back in 1996.

By the time they’d finally corrected their earlier errors the world had changed again and they were left looking as foolish as people buying the S&P at a Shiller earnings multiple in the 40s.

The lesson here isn’t necessarily that the CBO or the Fed are uniquely incompetent, but that long-term forecasts are hard.

It’s basically impossible to tell the difference between a sustained but ultimately temporary deviation from a trend and a change in the underlying trend. Like people everywhere, economists try to learn from their mistakes. Often, like the rest of us, they over-correct.

Forecasters’ responses to the financial crisis fit this pattern, just in the other direction. Consider the following chart from a fascinating new paper by Olivier Coibion, Yuriy Gorodnichenko, and Mauricio Ulate of how CBO forecasts of “potential” output have evolved over time:

Despite what some now soberly say was an inevitable slowdown thanks to population aging, the yawning gap between the black line and the red line shows that the weakness of the recovery was wholly unexpected. Years of slow growth and relatively stable inflation gradually pushed forecasters into concluding that we now live in the best of all possible worlds and this is more or less as good as it gets.

Coibion et al think official forecasters — the CBO is not alone — are making the same mistake as in the 1990s. Moreover, they think this is an inherent feature of “potential” output forecasts (emphasis in original):

Estimates of potential GDP are failing to adequately distinguish between permanent and transitory shocks. In this respect, estimates of potential GDP are sensitive to cyclical fluctuations in GDP originating from demand shocks…Our estimates imply that U.S. output remains almost 10 percentage points below potential output, leaving ample room for policymakers to close the gap through demand-side policies if they so chose to.

Remember that “potential” output is impossible to measure in real time. To forecast it, you have to be able to predict the growth of:

How many hours people work (harder than you’d think)
Capital investment (quite tricky)
Underlying or “total factor” productivity (good luck with that)

You might think you can predict working hours by forecasting the size of the labour force using demographics, but historically that hasn’t worked well. After all, the demographic projections made in 2007 were pretty close to what America actually experienced, but the labour force forecasts were way off. Meanwhile, in Japan, the declining number of people aged 15-64 has been more than offset by a surge in employment rates. Demographics may matter, but other things matter more.

Predicting capital investment is tricky since business spending is extremely cyclical and tends to be driven by changes in stock prices more than anything else. Besides, not all new investment is equal. Methodological limitations mean the statisticians treat shopping malls in the desert as equally helpful to the economy’s productive capacity as robotics factories and cell towers.

Then there is TFP, which is another important variable that can’t be measured directly. It can only be estimated by stripping the estimated impact of capital investment from the economy’s average output per hour. This underlying productivity could be affected by all sorts of things, from superior managerial practices to technological innovations. But nobody really knows.

So two of the three things you need to predict “potential” output are heavily affected by cyclical conditions and the third one is a residual of a residual with mysterious properties. No wonder Coibion et al are so sceptical of mainstream estimates:

The fact that most of the output declines observed since the Great Recession are now attributed to declines in potential GDP implies little other than that these declines have been persistent since estimates of potential GDP fail to adequately distinguish between the underlying sources of changes in GDP.

Their solution is to calibrate an estimate of “potential” that doesn’t respond to changes in government spending or monetary policy, only the oil supply, TFP, and tax rates. (The method was originally invented by Olivier Blanchard and Danny Quah in the 1980s.) The result looks like this:

This looks quite a bit different from the CBO’s estimates. Intriguingly, there does seem to have been a significant shift in the economy’s “potential” between 2011 and 2013, although it’s not clear why.

Since 2013, however, this alternative approach suggests the longer-term outlook hasn’t gotten any worse. On the other hand, the gap between “potential” output and actual output has widened into a chasm. In other words, conventional macroeconomic methods imply plenty of room for rapid growth without inflation. The recent rise in the labour force participation rate for Americans aged 25-54, after years in which it fell and pessimists assured us it was destined to keep falling, corroborates that hypothesis.

Still, Coibion et al are quick to point out that their result shouldn’t be treated as an exact answer even if it’s a lot more plausible than what’s peddled by official agencies:

The absence of clear ways to precisely estimate potential output in real-time suggests that the practice of relying on “judgement” by professional economists should not be discontinued anytime soon.

*Economies are supposed to grow in line with their “potential” over time even if they sometimes overshoot or undershoot over brief periods. According to one common theory, the rate of consumer price inflation should speed up or slow down if real output persists in being either above or below its “potential”.

That theory doesn’t make much sense (related) but it affects government budgeting and monetary policy. Of course, even if the theory were true, you couldn’t do anything useful with it because the key variable is impossible to measure in real time.

This means the unlucky souls tasked with the job of creating “potential” output estimates have to reverse-engineer their results by looking at inflation. If inflation is stable, gross domestic product is close to “potential”. If inflation is accelerating, then GDP must be above “potential”. And if inflation is slowing, then GDP must be below “potential”. (Sometimes there is a variation where instead of the change in the rate of inflation the metric is the gap between current inflation and the central bank’s target, but you get the idea.)

Inflation has been remarkably stable around the Fed’s implicit and later explicit target of 2 per cent per year for more than two decades, which implies the US economy has generally been pretty close to its “potential” over that entire time span, if you believe in that sort of thing.