Wednesday, January 28, 2015

Unemployment insurance and unemployment

"The Impact of Unemployment Benefit Extensions on Employment: The 2014 Employment Miracle" by Marcus Hagedorn, Iourii Manovskii and Kurt Mitman is making waves. NBER working paper here. Kurt Mitman's webpage has an ungated version of the paper, and a summary of some of the controversy. It's part of a pair, with "Unemployment Benefits and Unemployment in the Great
Recession: The Role of Macro Effects" also including Fatih Karahan.

A critical review by Mike Konczal at the Roosevelt Institute blog, and a more positive review by Patrick Brennan at National Review Online are both interesting. Both are thoughtful reviews that get at facts and methods. Maybe the tone of the economics blogoshpere is improving too. Bob Hall's comments and response on the earlier paper are also worth reading. This is a bit deja-vu from the observation that North Carolina experienced a large drop in unemployment when it cut benefits. My post here, WSJ coverage, and I think there are some papers which google isn't finding fast enough at the moment.

The basic issue: I think it's widely accepted, if sometimes grudgingly, that unemployment insurance increases unemployment. If you pay for anything, you get more of it. People with unemployment insurance can hold out for better jobs, put off moving or other painful adjustments, and so on. The earlier paper points out that there are important general equilibrium effects as well. We should talk about how UI affects labor markets, not just job search.

Quick disclaimer. Let's not jump to "good" and "bad."  Searching too hard and taking awful jobs in the middle of a depression might not be optimal. Pareto-optimal risk sharing with moral hazard looks a lot like unemployment insurance.  Perhaps that disclaimer can settle down the tone of the debate.

But the question remains. How much?  How much does unemployment insurance increase unemployment? And the related macro question, just why did unemployment in the US suddenly drop coincident with sequester and the end of 99 week unemployment benefits?

Method is important. Too much media coverage starts and stops with "study finds unemployment insurance raises unemployment." And then the next day "study finds unemployment insurance crucial to stopping people from dying in gutters." If we focused on the facts, we'd all get along better.

In macro, we always are faced with the problem that interpreting time series, we never know what else changed. Sure, congress lowered unemployment benefits and the economy took off. But lots of other stuff happened. Maybe it's "despite" not "because."

This paper is part of a new breed trying to get around this problem by looking at cross-sectional evidence. Roughly, the evidence in this paper builds on the fact that Congress' action had different effects in different states.

Bob Hall described the strategy compactly:
They compare labor markets with arguably similar conditions apart from the UI benefits regime. In their work, the markets are defined as counties and the similarity arises because they focus on pairs of adjacent counties. The difference in the UI regimes arises because the two counties are in different states and UI benefits are set at the state level and often differ across state boundaries. The research uses a regression-discontinuity design, where the discontinuity is the state boundary and the window is the area of the two adjacent counties....
Table 3 contains the basic number, which the authors digest as
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.
(Small complaint: economists should not write that jobs "were created," especially economists writing in the search, match and labor-supply tradition, to say nothing of passive voice and strong causal inferences.) I tried to digest the fact a bit more, but stopped here:
Column (1) of Table 3 contains the results of the estimation of the effect of unemployment benefit duration on employment using the baseline specification in Equation (6).

If commenters can vocalize the actual fact in words, fixed effects, controls and all, I'd be grateful.

Bob Hall echoes standard but important complaints.
The issues that arise in evaluating the paper are those for any regression-discontinuity research design: (1) Are there any other sources of discontinuous changes at the designated discontinuity points that might be correlated with the one of interest? (2) Is the window small enough to avoid contamination from differences that do not occur at the discontinuity point but rather elsewhere in the window?
In words, is there something else about state policies that changed at the same time in the "generous" vs. "stingy" states? And are counties really small enough to capture only the border effects?

The deeper issue in evaluating this paper, I think, comes from blowing the county results up to the aggregate, as Bob but it
The authors conclude that, absent the increase in UI benefits, unemployment in 2010 would have been about 3 percentage points lower.
The jump back from micro to macro isn't so easy either.  For example, suppose the expansion came from selling more goods from expanding states to contracting states. Then you'd see a micro effect but no macro effect. I don't think that's the case, but I have been skeptical about other papers jumps from micro to macro. For example, if the Federal government spends a trillion dollars in the desert, and a bunch of businesses move to sell donuts to the construction workers, you get a nice stimulus. That doesn't mean stimulus works for the economy as a whole.

This is a small nitpick. The basic fact is interesting, and I think a lot harder to dismiss.

It's interesting that so much of the pushback, both from Bob and from Mike Konczal's critical review comes down to theory, not the fact.

Update: Wednesday's Wall Street Journal covers the paper. The WSJ spends more time on the macro question, the claim that unemployment insurance actually boosts the economy via stimulus. 


  1. People should save so that losing job does not immediately make them unable to pay for necessities. Why not create unemployment insurance accounts that people can contribute to from pre-tax income and from which withdrawals can be made upon losing a job or at age 60? To limit the loss of tax revenues you could disallow contributions to accounts that already have $100K or more.

    1. Quant: I have been advocating precisely that for years.

      Ever since I served on a Federal jury with a fellow juror who confided to us how he optimally exploited NY's UI program. Basically, he worked for H&R Block for a few months a year during tax season, then got laid off afterwards and immediately claimed UI for the rest of year. Nice guy, clever parasite.

      p.s. JC: this website has the most difficult CAPTCHA I have ever encountered...

  2. Dean Baker notes several data quality problems in the analysis. It's difficult to discount a partisan appeal if the robustness of the results are not addressed up front in the face of some volume of contradictory evidence.

  3. In an economy with ~320 million people, cause and effect relationships are impossible to prove. However, this is indisputable:

    "If you pay for anything, you get more of it."

    Anyone that disputes that has an agenda.

  4. Seems to me that you did not understand the math, and that's the basis for your review, aside from the observation that the sample size is not representative of the whole. For a more substantive review, check out this comment by me: which one of the co-authors responded to, and my rebuttal.

  5. As I look at this (being a conservative and a business expert) it becomes clear that economists sit in their offices playing with models and statistics, most of them have no concept of or connection to the realities of business or the labor market or just about anything else.

    The use of math does not mean economics is a science, despite what economists might think.

    The claims for unemployment in 2010 are ludicrous to anyone who knows anything about the real operations of businesses.

    1.8 million jobs created by the end of extended unemployment? No undergrad business major would ever say anything that ridiculous.

    1. Simply stating that the arithmetic is "ludicrous" doesn't disprove anything presented in the post.

      Although you can make the case that math can only find the relationship between variables and not predict the future relationship, simply posting that you're a "conservative and business expert"(if you even are) is in now way more credible then the material presented.

      If you have a real argument, other than crude misrepresentation of economists, then present it for debate. Don't come on this blog and just throw around arrogant statements.

  6. The length of unemployment insurance ought to be tied to the state of the labor market, possibly state by state, which means that it should be market determined, not politically determined.

  7. So it appears the jobs existed but were not filled because people had unemployment insurance benefits and preferred to stay unemployed. If the positions existed, they were most likely advertised. Why not look at the number of advertised jobs and its time evolution. If the paper's authors are right, the number of advertised jobs should stay constant and then it should start dropping as the positions are being filled. The drop should begin at the time when benefits were stopped. Otherwise, the paper most likely demonstrates correlation but not causality.

  8. Hagedorn et al use the employment data from the household survey. Why didn't they use the establishment survey? UI benefits are tied to the state where you work - not where you live. The establishment survey would allow apples to apples comparisons, while household does not.

  9. It is true that any benefits "create" a group of people who just want to exploit them. Two important questions are 1. How many and 2. Would the cost to society be greater if we force them to rely on something else to survive.
    Second question is not important in relation to this article.
    If I remember correctly, between 2008 and 2010 relation between unemployment and new jobs was around 10 unemployed per job. For conclusion in the article to be correct, everyone who lost their job would have to prefer to sit on benefits instead of finding new job ASAP. I find that impossible to believe.


Comments are welcome. Keep it short, polite, and on topic.

Thanks to a few abusers I am now moderating comments. I welcome thoughtful disagreement. I will block comments with insulting or abusive language. I'm also blocking totally inane comments. Try to make some sense. I am much more likely to allow critical comments if you have the honesty and courage to use your real name.