Felix Salmon on Sebastian Thrun and Udacity
CBO Releases Its Economic and Budget Outlook

Hysteresis and the American Unemployment Problem

Continuing to work on my Brookings Paper with Larry Summers: "Fiscal Policy in a Depressed Economy".

Here is an early sketch of what we might say about labor-side hysteresis:


Since the seasonally-adjusted U.S. unemployment rate reached its peak in the second half of 2009, the measured unemployment rate has been on a decline: fro a peak of 10.0% to a value of 8.5%, seasonally-adjusted, in December 2011.

However, this decline in the measured unemployment rate does not mean that a larger proportion of the American adult population is at work today than at the unemployment rate’s peak. In fact, the employment-to-population ratio in December 2011 was at exactly the same value that it had been at during the late-2009 unemployment rate peak. The unemployment rate today is lower today because the labor force participation rate is lower, not because the employment-to-population ratio is higher.


It would be expected that the labor force participation rate would be lower when the unemployment rate is higher: to be in the labor force requires that you have a job or be searching for a job, and when jobs are scarce—as is the case when the unemployment rate is high—taking the steps and making the effort to be in the labor force will flunk the benefit-cost test for more people.

What is slightly more surprising is the pattern over the past two years. Jobs right now, at least as measured by the employment-to-population ratio, are no more difficult to get than they were two years ago. But two years ago the labor force participation rate stood at 65%. Now it stands at 64%. 2.4 million American adults who would have been in the labor force two years ago are not in the labor force today, even though jobs appear to be no more scarce today than they were two years ago.

Over the post-World War II period in the United States, on average a one percentage point decline in the employment-to-population ratio has been associated with an 0.27 percentage point decline in the labor-force participation rate—and that decline in the labor-force participation rate is then made up when the employment-to-population ratio recovers. And, indeed, starting in January 2008, the recession-period change in the labor-force participation rate tracked what one would have expected from the historical pattern and the behavior of the employment-to-population ratio closely. But since the employment-to-population ratio finished its decline and began its flatline in late 2009, the behavior of the labor-force participation rate has been anomalous. Rather than stabilizing, it has continued to decline. It appears likely that persistent high unemployment in the United States is generating labor-force participation dynamics that have not previously been seen, at least not in the post-World War II period.


The obvious hypothesis is that the considerations raised by Blanchard and Summers (1986) in their “Hysteresis and the European Unemployment Problem” for Europe in the 1980s are starting to apply to the United States today. Persistent high transitory cyclical unemployment is transforming itself into permanent structural unemployment as the labor market recovery continues to delay its appearance.

How large should we mark down our estimate of the long-run economic growth path of the United States from the employment reports of the past two years? Two years ago, after all, the recession was over. The employment-to-population ratio was where it is. The labor force participation rate was a full 100 basis points higher.

If the unemployment rate follows a stationary stochastic process without much memory, then it by itself encodes the state of the business cycle and thus the gap between current and potential GDP. Under this shaky but not implausible assumption, the one percentage-point decline in labor-force participation over the past two years indicates that the long-run average employment-to-population ratio in the U.S. in the future will be one percentage point lower than would have been forecast two years ago under the assumption of a rapid labor-market recovery. Then the forecast would have been for a long-run employment-to-population ratio forecast of 62.5%. Now the forecast is for a long-run employment-to-population ratio forecast of 61.5%—a reduction in expected future long-run labor input of 1/60.

It is possible to argue that, with a labor share of .6, such a reduction in long-term labor supply carries with it a 1% reduction in potential output. It seems to us more reasonable to argue that the decline in labor supply is primarily a decline in raw unskilled and semi-skilled labor supply and not a decline in human capital supply, and that the decline in labor-force participation over the past two years as the employment-to-population ratio has flatlined has reduced potential output by 0.5%.

If so, then the experience of the past two years provides enough information to produce a one-episode estimate of the labor-side hysteresis parameter η needed for the simple analytical framework. Two years during which real GDP has stayed flat at 7% below our pre-2007 estimates of potential output have managed to push potential output down by 0.5%: that suggests a value for η of 0.5/(7 x 2) = 0.035.

In the context of equation (11):[1]


and with a value of 2.7%/year for the long-term growth rate g of the American economy, such a value for η looms very large in the social-welfare cost benefit analysis indeed. With a real social rate of time discount rd of 5%/year, η of 0.035 produces a present value of gross benefits from expansionary fiscal policy at the margin 2.5 times as large as simple multiplier calculations focused on current output. With a rate of time discount of 4%/year, it is not 2.5 but 3.7 times as large. And with a social rate of time discount of 3%/year, it is not 2.5 or 3.7 but rather 12.7 times as large.

The social discount rate of 4%/year also allows us to do a calculation of the cost of each extra month’s delay in the coming of a recovery proper to the U.S. labor market. Inserting an extra month with the output and employment gap at its current level costs the American economy roughly $100 billion in foregone immediate output. And, at a social real discount rate of 4%, if the reduction in labor-force attachment is indeed permanent, it also costs the American economy $270 billion in the present value of reduced future potential output.

What appears to be the slow transformation of cyclical into structural unemployment in America today is an order of magnitude more damaging to our prosperity than is the simple fact that we are mired in the Lesser Depression.

[1] Where:

  • ΔW is the (marginal) change in social welfare induced by a (marginal) change ΔG in this-period government purchases

  • rd is the real social rate of time discount

  • g is the real growth rate of the economy

  • η is the labor-side hysteresis parameter--the share of today's boost to employment/output that is permanent because deep recessions cast shadows and reduce aggregate supply by increasing failures in the labor market

  • δ is the deadweight loss from raising a dollar of tax money in the future

  • m is the multiplier, net of any central bank neutralization of fiscal stimulus

  • τ is the current tax rate

  • ρ is the (if positive) default premium interest rate that the government must pay over and above the social rate of time discount, or (if negative) the value to investors of having safe government paper backstopped by the taxing power to invest in).