Scott Sumner  

Stone Age Economics

PRINT
The Lone Collectivist... Some Californians versus the S...

Back in 2000, Robin Hanson wrote a paper entitled "Shall We Vote on Values, But Bet on Beliefs?", which offered a way forward for economics in the 21st century. Unfortunately his ideas were ignored (and even ridiculed) and today we are still stuck in 20th century economics. In the future, our current approach will be regarded as Stone Age economics.

I will illustrate the problem with modern economics by discussing the impact of extended unemployment insurance. Last spring, Paul Krugman suggested that the elimination of the emergency extended unemployment program in 2014 was not leading to more jobs, thus refuting the claims of conservative opponents of the program. In earlier posts he suggested that ending the extended UI program would mean less fiscal stimulus, and hence more unemployment. As we'll see, this prediction turned out to be as ill timed as his famous "test" of market monetarism comment, which occurred a year earlier. Nonetheless, at the time it looked like Krugman might be right, as the first quarter of 2014 was weak (perhaps due to bad weather.) In addition, Congress was still debating an extension, which would have applied retroactively to those still unemployed.

In numerous posts over at TheMoneyIllusion, I suggested that the 99-week extended unemployment insurance program had probably increased the unemployment rate by about 0.5%. That's perhaps 700,000 people, which is significant, but not the major cause of high unemployment during the recession. I always acknowledged that this was little more than a guesstimate.

Now Tyler Cowen directs us to a fairly rigorous academic study that uses a "difference in difference" approach and estimates that ending extended UI led to an additional 1.8 million jobs in 2014:

We measure the effect of unemployment benefit duration on employment. We exploit the variation induced by the decision of Congress in December 2013 not to reauthorize the unprecedented benefit extensions introduced during the Great Recession. Federal benefit extensions that ranged from 0 to 47 weeks across U.S. states at the beginning of December 2013 were abruptly cut to zero. To achieve identification we use the fact that this policy change was exogenous to cross-sectional differences across U.S. states and we exploit a policy discontinuity at state borders. We find that a 1% drop in benefit duration leads to a statistically significant increase of employment by 0.0161 log points. In levels, 1.8 million additional jobs were created in 2014 due to the benefit cut. Almost 1 million of these jobs were filled by workers from out of the labor force who would not have participated in the labor market had benefit extensions been reauthorized.
Obviously this study is far superior to my guesstimate. And in a sense it does support my side of the debate I'm having with Keynesians, who (falsely) accuse me of promoting the "lazy worker" theory of unemployment. So I should jump on this result, right? Especially since I can't find any flaw in their empirical work (although honestly I just skimmed the paper.)

In fact, I'm not being a good Bayesian, I'm not shifting my prior view that extended UI cost about 700,000 jobs, although I am widening the band around that estimate to include 1.8 million as a plausible estimate. I'll try to explain my stubbornness, and I want smarter, less biased people to tell me if I am wrong.

In 2012 the US created about 2.25 million jobs, and in 2013 it created about 2.35 million jobs. In late 2013, before it was known that extended UI would be repealed, most economists seemed to expect 2014 to be at least as good as the previous two years. The markets also seemed to expect continued growth, although unfortunately we lack good RGDP and NGDP futures markets. But my sense was that 2014 was likely to be similar to 2013, and it seemed to me that forecasters in academia and the asset markets both expected a similar result.

In fact, employment growth in 2014 was 2.95 million, a number quite likely to be revised higher in the next year or two. That's why I still think 700,000 is a decent ballpark guesstimate. I just don't find it plausible that job growth would have suddenly plunged to 1.15 million in 2014, if nothing had been done about extended UI. I saw nothing going on in terms of "shocks" that would have suddenly caused job growth to slow.

Here's where Robin Hanson comes in. Under his "futarchy" plan, prediction markets would decide matters of fact ("beliefs") that have implication for policy, and voters and policymakers would decide the values that get embedded in policy. In this case, two prediction markets would have been set up in late 2013, to predict jobs growth during 2014. One market would be conditional on Congress extending the emergency benefits, and the other market would be conditional on the program being ended.

If the markets agreed with me, the difference would have been about 700,000. If they agreed with Marcus Hagedorn, Iourii Manovskii, and Kurt Mitman, the difference would have been 1.8 million. If they agreed with Krugman it might have been negative (fewer jobs if the program ended.) Think of the market forecast as a sort of meta-study, which efficiently incorporates Krugman's arguments, my arguments, and the empirical work in the study mentioned above.

When I forecast the effect of X, this is exactly what I try to do. I try to guess what the market would guess. Thus I think eurozone QE is something, but not a game changer, because that seems to be what the markets expect, based on their responses to QE news stories.

There is a mountain of evidence that the "wisdom of crowds" can aggregate many disparate views much more effectively than any individual researcher. So why didn't economists follow Hanson's lead? Perhaps we are a bit arrogant, thinking we are smarter that the markets. Maybe we like the idea of economists being a sort of cult of "high priests" that hold the secrets of the temple. It's a flattering position to be in.

I'm not suggesting that this sort of difference in difference study has no value. Just the opposite. It can provide useful information to markets. So please don't take this post as a criticism of this very fine study. Rather I'm saying that when policymakers in Washington decide whether to extend UI, or raise the minimum wage, they should look to market forecasts of the effect of policy. But first someone needs to spend the extremely small amount of money required to set up these prediction markets. I'm trying to do my part.

Most contemporary economists would find Hanson's blog to be a bit "weird." Economists of the 22nd century will wonder why everyone wasn't a Hansonian back in 2015.


Comments and Sharing


CATEGORIES: Labor Market




COMMENTS (13 to date)
Michael Bishop writes:

I'm a big fan of moving in this direction... that said, a market forecast conditional on a policy change is not the same thing as an unbiased estimate of the causal effect of the policy. My first thought is that the more of the relationships that matter can be traded the more we can "control" for just as with a standard regression approach. I'll keep thinking about how we might do better than that.

"Economists of the 22nd century will wonder why everyone wasn't a Hansonian back in 2015."

This is highly unlikely. Either they will still be stuck in the stone age, or they will understand the underlying reasons well and won't wonder at all :-)

Derek writes:

Even 700k seems like a high figure when you look at the job growth of the previous years and then 2014. Expectations for growth in 2015 were higher than in the previous 3 years, so some of the extra job growth in 2014 is due to ramping up in advance of 2015 and, in the second half of 2014, generally sunnier economic skies in the US. Now, if you want to chalk up the better macro environment in the second half of 2014 to the end of UI insurance, well, okay, but otherwise, it seems like we're looking at well less than half a million.

Jason Smith writes:

UI extension ending causes a job openings spike??

http://research.stlouisfed.org/fred2/series/JTSJOL

LK Beland writes:

The result of HMM is highly dependent on the source of employment statistics.

Using an alternative database, Dean Baker found the opposite result:
http://www.cepr.net/index.php/blogs/cepr-blog/did-cutting-the-duration-of-unemployment-benefits-lead-to-faster-job-growth-in-2014

Alex writes:

The Hansonian view towards more betting makes plenty of sense. But. Other comments note. Policy expectations influence market participants' weights on outcomes. Too much endogeneity to see the truth.

And. Researchers cite the efficacy of odds-bets on sports/political markets as proof of wisdom of crowds. However those outcomes are (roughly) uncorrelated to macroeconomic outcomes and so (roughly) require no risk premium. I'm skeptical there won't continue to be debate about 'true' point estimates.

Revealed preference: Hanson would be a trader, not an academic, if he thought markets were such a purely intellectual endeavor.

Mr. Econotarian writes:

Any analysis of the change in employment situation will have to wrestle with the fact that 1.4 million jobs were created in Texas, while the rest of the country lost about 300K jobs from December 2007 to December 2014.

LK Beland writes:

Mr Econotarian

You get a different result if you look solely at private employment.

12-2007 to 12-2014
US: +2.2M
Texas: +1.1M
Us minus Texas: +1.1M

Government jobs are a big part of the "Texas miracle".

Also, Texas had a much smaller recession, in part because it did not have a massive housing crash. We can compare post-recession private job growth:

12-2009 to 12-2014
US: +11M
Texas: +1.5M
US minus Texas: +9.5M

In my opinion, this is a better assessment of the change in the employment situation. Texas did have better job growth, per capita, than the rest of the country. But not to the extent that the AEI talking point suggests. Not even close.

Michael Byrnes writes:

Mr. Econotarian wrote:

"Any analysis of the change in employment situation will have to wrestle with the fact that 1.4 million jobs were created in Texas, while the rest of the country lost about 300K jobs from December 2007 to December 2014."

This is an incredibly misleading talking point, for several reasons, but here is the key reason:

First, it implies that Texas is largely responsible for US job growth since 2007 (since Texas gained jobs and "US minus Texas" lost jobs). But this isn't supported - from the article, we know "US minus Texas" lost jobs, but we don't know from the article how Texas compared to the other 49 states.

Did EVERY non-Texas state lose jobs - in which case the articles premise is reasonable), or did some/many states gain jobs while the rest lost enough jobs to offset the gains? If you add up jobs created in all of the states that had job growth, Texas included, you'd get a number that would be well in excess of 100%. In fact that number could be really high - more than 200% or 300% or 400%. Would it make sense to say "the 28 (making this number up) states with positive job growth were responsible for (say) 198% of job gains in the US?". Such a statement would be sort of meaningless. Though perhaps it is true, the article did not state outright that Texas' job growth was higher than that of any other state. The article did not present any real evidence that Texas' job growth was impressive relative to other states.

As a general matter, percentages like this (either stated or implied, as done here) don't make a whole lot of sense when there are both positive and negative contributors. (Thought experiment: Instead of "US minus Texas" losing 300,000 jobs, imagine that "US minus Texas" actually gained 300,000 jobs, and all 50 states had positive job growth. Now this really would be a remarkable statement about Texas, becauese we would know for sure that Texas was a real outlier. As it stands, the article didn't provide any information about whether Texas was unique among states or just one of those that did well.)

Beyond that there are more minor issues like whether absolute job growth vs. job growth per capita is the more important measure, to what extent this is driven by demographics, etc..

Michael Byrnes writes:

Scott wrote:

"In 2012 the US created about 2.25 million jobs, and in 2013 it created about 2.35 million jobs. In late 2013, before it was known that extended UI would be repealed, most economists seemed to expect 2014 to be at least as good as the previous two years.

...

In fact, employment growth in 2014 was 2.95 million, a number quite likely to be revised higher in the next year or two. That's why I still think 700,000 is a decent ballpark guesstimate. I just don't find it plausible that job growth would have suddenly plunged to 1.15 million in 2014, if nothing had been done about extended UI."

I'm far from an expert, but this line of reasoning seems spot on.

I have one thought that may be relevant, though I am not sure whether you or the authors of this analysis have already taken it into account:

Extended UI had a maximum duration of 99 weeks. Once people exhausted their 99 weeks, they were done with extended UI even if the program as a whole continued. And the duration of extended UI in a state depended on that state's unemployment rate - in many (all?) states it had fallen to 73 weeks before the program ended.

So it wasn't a scenario where all unemployed people were on extended benefits and then come Jan 1 2014 they were all kicked off a cliff. There must have been some rate at which long-term unemployed were exhausting their extended UI before 2014 (a rate that was gradually increasing as the extended UI duration fell). Then in 2014, everyone on long-term UI lost it when the program expired. That's a much smaller number of people than all long-term unemployed, though.

I'm not sure whether or how that would affect the impact of extended UI expiration. Maybe a stronger effect for individuals having extended UI benfits versus not having them, but a lesser effect on the US unemployment rate from 2013 to 2014, since whatever the effect of losing extended UI benefits, it was already happening to a lot of people before 2014.

Scott Sumner writes:

Michael, Sorry, But you lost me somewhere.

Partial, Good point.

Derek, Actually, a "better macro environment" is certainly a prediction of the supply-side view of this policy reform. But I concede it might have been less than 700,000. We just don't know. On the other hand the study that showed 1.8 million used well respected methods. That doesn't mean I agree, just that it's hard to pin these things down.

I would add that if extended UI had become permanent (as in Europe) we'd have much higher unemployment, even in booms (just as in Europe.)

Jason, Yes, that would be the expectation.

LK, I would assume they used the payroll figures, as the household survey is not considered reliable. Am I wrong?

Alex, Sorry, I didn't follow that at all. Why would Hanson be a trader if his comparative advantage is academics?

MR, Econotarian and LK, A few comments on Texas:

1. Texas has been growing much faster than the US, and much faster than it's neighboring oil producers, for many decades. Probably for supply-side reasons, small government, weak zoning laws, etc.

2. But even a small government as a share of GDP will grow fast when the overall economy is growing fast, as in Texas.

Michael, I see your point, but I think you underestimate the "churn" in the unemployment data. There is very large turnover, so by 2014 we were no longer seeing people on UI who lost jobs in the 2009 recession. Yes people leave after the benefits run out, but new people enter all the time.

Mark Bahner writes:
Economists of the 22nd century will wonder why everyone wasn't a Hansonian back in 2015.

There will be no economists in the 22nd century. The world per-capita GDP in 2100 will exceed $10 million (in year 2000 dollars):

Long Bets #194, on per-capita GWP in the 21st century

The world per-capita GDP in the year 2000 was approximately $7,200. The world per-capita GDP (in year 2000 dollars) will exceed $13,000 in the year 2020, $31,000 in 2040, $130,000 in 2060, $1,000,000 in 2080, and $10,000,000 in 2100.

P.S. If anyone knows Robert Lucas Jr., please let him know that this bet is especially open to him. ;-) (No, seriously. I probably shouldn't have put a wink there. I'm not kidding. The economics profession is almost universally ignorant of the likely effects of artificial intelligence on economic growth. A public bet with him would put the issue much more into the public spotlight.)

Larry writes:

Why do the Feds hate prediction markets?

Comments for this entry have been closed
Return to top