Bryan Caplan  

I Win My Long-Term Unemployment Bet with Tyler

Altruism, No Regard for Some O... My Talk at the University of T...
Barely two years ago, Tyler Cowen and I agreed to the following bet:
...U.S. unemployment will never fall below 5% during the next twenty years.  If the rate falls below 5% before September 1, 2033, he immediately owes me $10.  Otherwise, I owe him $1 on September 1, 2033.
Unemployment has been stalled at 5.0% for months.  But this January, the rate slipped to 4.9%.  I have therefore won the bet with almost 18 years to spare.

Per my Bettor's Oath, I commend Tyler for having the courage to publicly bet on his pessimism.  (Unlike some people).  But it's not out of line to observe that his extreme confidence was extremely out of place.  Where did he go wrong?  My port mortem: Like most news junkies, Tyler refuses to admit the near-insignificance of current events.  He loves the idea that the latest headlines reveal the latest deep truths.  I, in contrast, take the far view.  I form my views on the basis of long-run basic facts, and regard news as near-noise. 

The far view that guided my bet is rooted in the following graph for U.S. unemployment rate from 1948-2016.

There's a strong pattern: While unemployment spikes fairly unpredictably, it reliably falls by about 1 percentage-point per year after it peaks.  So while I wasn't confident about how high unemployment would go, I strongly expected it to recede at about its usual rate.  Since it peaked at 10% around January 2010, I expected it to get back to 5% by January 2015.  In the end, it took six years instead of five, but well within historic bounds.

My deeper incredulity, to be sure, was Tyler's extreme (10:1!) confidence that unemployment had entered a durably high natural level.  This has happened in the past, but very rarely for a full two decades.  Furthermore, there were no dramatic warning signs the last time we entered the era of high unemployment, just as there were no dramatic advance warning signs we were exiting that era.  Here, as elsewhere, claims to gnosis do not impress me.  

Did I get lucky?  I'm happy to restart the bet the next time a recession hits.

COMMENTS (13 to date)
john hare writes:

Actually I wouldn't bet with you on a sure thing. Letting you hold my money is unacceptable. Tying up my money for years instead of working it is also unacceptable.

Micke writes:

If I read the graph correctly, unemployment was above 5% between 1974 and 1997. (Maybe it touched 5% in 1988, in which case the question is if the same bet but for 4.8% instead of 5% should be treated wildly differently.)

I would much rather have had your than Tyler's side on this bet, especially at those odds, but it doesn't seem like such a slam dunk to me. I wouldn't have been very surprised if unemployment had had lows 0f 5.5%-6% 3-4 times in the period but never going all the way below 5.0%.

dWj (Dean) writes:

Continuing on Micke's point, I can imagine having some confidence in 1972 that rates will get back below 3.8% on more or less that basis; that bet you would have lost.

I assume though, that you wouldn't have taken the bet with the odds reversed -- is that right? There's a much greater than 10% chance per business cycle of something of a regime change, so that 10:1 odds is a bit too confident to be of this sort of thing.

Jack PQ writes:

What about the employment level? It has remained low. Prof Cowen might have expected the unemployment level to absorb more of this, but instead it is the out-of-workforce numbers that are increasing much more.

pongogogo writes:

The better conclusion here is to state unemployment rates are volatile and we can't make long term predictions of where they will head. Additionally, if you have to make a prediction then look for a bet with an asymmetric payoff. Otherwise you're just left making predictions and hoping to be lucky, just like Tyler was. I'm not sure that pattern matching from historical data is a valid approach at all.

Daniel D writes:

@Jack PQ

Employment levels are influenced by changes in population structure (% age 65+) and employment levels are rising, not falling. What you really want to do is look at the employment to population ratio in age group 25-54 which shows the emplymwnt rate rising much faster or else look at the u-6 unemployment rate which includes marginally attached and discouraged workers and which continues to fall.

ThomasH writes:

I would think it would be more fun to bet on a parameter estimate of a model. It's not clear who learned what from the exercise.

Ricardo writes:

Don't you need to wait for the revisions next month and the month after?

MKB writes:

Curious as to why the bet was based on headline unemployment, as opposed to, say, rate of labor force participation, or non-farm LFP?

pgbh writes:

Tyler doesn't seem that "extremely" confident to me. If he were, he probably would've bet more than $10.

john hare: If you're "sure" to win, you don't need to hold money while you wait for the bet to resolve.

john hare writes:


Bryan has posted his method a few times. You make a bet with him and he holds the money until the bet resolves. There are many reasons to not let other people hold my money. The sureness of the outcome is a different issue.

Nathanael writes:

Thing about this is that Tyler picked the 5% number out of a hat; if unemployment promptly goes up again and stays above 5% for another 18 years, you will have won the bet, but he will have won the argument.

Kevin writes:

With workplace participation rates in flux I think this is a poor metric. And does the bet start over if the numbers are revised?

I would have bet on something that better represents overall employment because if enough people quit looking for work or get on disability you win. Thats does not support your position very well.

I would also have specified that the rate needed to maintain for some period. The randomness of events means that unemployment data is an estimate of a true value and so bounces around a bit.

Comments for this entry have been closed
Return to top