Monday, April 4, 2011

Waiting for the perfect job does not work

We usually think that one advantage of unemployment insurance is that it allows unemployed workers to take more time to find an appropriate job and not rush to the first available job for which they may not be qualified. As Acemoglu and Shimer have forcefully argued, the absence of unemployment insurance could lead to important mis-allocations for this reason, and thus losses of efficiency in the economy. This argument implies that workers with longer unemployment durations, after controlling for other characteristics, should be getting better jobs. Is this true?

Miki Kohara, Masaru Sasaki and Tomohiro Machikita test this for Japan using job tenure as a signal of job quality. It turns out the hypothesis above is easily rejected. Indeed, the longer the unemployment duration, the shorterthe subsequent employment duration. So it appears that other factors, like loss of skills and human capital during unemployment and stigma matter more than the search for the perfect job. Of course, it could also be due to some missing control variables, but the effect is so strong I doubt it can be overturned.

2 comments:

Anonymous said...

There is an alternative explanation of the data that fits more closely with the Acemoglu and Shimer hypothesis: If unemployment insurance is currently 'too low' people with longer unemployment duration become more desperate (as savings are exhausted etc.) and are more willing to take a 'bad' job.
If unemployment insurance were extended, then these workers may be able to wait for a 'good' job.

Perhaps a good field experiment is in order.

ilkka said...

"Controlling for other characteristics" is usually code for "controlling for the few characteristics we have convenient data for".

No doubt prolonged unemployment and low quality jobs are both highly correlated with the market value of the applicant. No administrative data has a good view of that, so there is missing variable bias that should not be ignored. Correlation does not imply causation, especially not when there is another very obvious explanation for the correlation.

We need a natural experiment or IV or something here.