New Research

OK, I'm far enough along on two "back burner" research projects that I think it's worth talking about them here.

The first is a paper that I'll be presenting at the Southern Economic Association conference in D.C. this November. It addresses the issue of "skills mismatch" in the United States, or more accurately - it computes an index of skills mismatch suggested by Petrongolo and Pissarides (2001) that I haven't seen estimated empirically anywhere and was curious about.

There are lots of "matching functions" out there that try to describe frictions involved with matching one party to another in any of a variety of transactions. The matching I'm concerned with is job matching. A job matching function is usually of the form: M = z(U, V). The number of matches is a function of the number of unemployed workers, the number of job vacancies, and a matching technology z. I won't get into the weeds here, but there is a variation on this model called the "ball-urn model" that looks like this: M = V(1-e^(U/V)). Petrongolo and Pissarides (2001) mention a variation of the ball-urn model that includes "K" - an index of skills mismatch: M = V(1-e^(KU/V)). With a little math, you can solve for K (although they don't). K is the percent of unemployed workers who are qualified for a job vacancy. As K goes up, the second equation I presented converges to the first equation. Anyway, its a nifty, easy little index they suggest that even this public policy grad student can understand - so I thought, why not calculate it for the U.S., and see what I can say about (1.) whether this is even a valuable index, and (2.) trends in skills mismatch in the U.S.

I use the Job Opening and Labor Turnover Survey (JOLTS) data produced by the Bureau of Labor Statistics, along with unemployment figures. The JOLTS data are relatively new - going back only to 2000 - but it is produced monthly, so there's a fair amount of data points. My calculations for K are below:
Conclusions so far? K is clearly very cyclical - it goes up in the summer and down in the winter. Not sure why this could be yet - I don't know much about the seasonality of unemployment. But basically that says that unemployed workers are more qualified for summer jobs than they are for other jobs. I guess that's believable... What is kind of neat is that K is not responsive to the U/V ratio (the red line above). The U/V ratio (also known as the Beveridge Curve) is an index of labor market tightness. When the ratio is high, it means that there are a lot of workers chasing very few jobs. If it is low, it means that there are a lot of jobs out there available for workers. U/V is also the denominator of K when you solve the matching function I presented above. So you would expect that when U/V increases dramatically (as it does in the first few months of the graph), K will decrease. Maybe you could convince me there's a slight drop at that period in time, but not really. K stays very consistent. This is probably a good thing. I don't think many people would be covinced by the idea that skills shortages in the US economy swing dramatically over time. If there are too many workers chasing too few jobs or too few workers chasing too many jobs, you would think the skills ratios among workers and jobs should stay pretty constant.

Next on the list:

(1.) measuring a standard Cobb-Douglas matching function (M = b0 + b1ln(U) + b2ln(V)), allowing b0 to vary over time. b0 is equivalent to the matching technology "z", that I mentioned above. It's an index of general frictions in matching. I'll then chart K against z to determine whether K is really picking up skills mismatch, or whether its just "the share of workers who have a tough time matching to jobs."

(2.) The American Community Survey collects information on the time it takes to travel to work, which could be a proxy for spatial mismatch - one of the most common types of mismatch discussed in the literature. Its an annual survey, unfortunately, so I won't get nearly as many data points - but I want to compare changes in spatial mismatch to K to make sure K isn't picking up those changes accidentally.

(3.) I'm going to reestimate K after differentiating between full time and part time job seekers. In effect, the equation will look like this: M = V(1-e^((-KFT + PT)/V)). This version of K assumes that all part-time job seekers are qualified for whatever jobs they may seek, but some full time job seekers may not be. High paid consultants aside, this seems like a reasonable assumption... and it may get some different dynamics for K.

OK, I need to get a few things ready for work now, so I won't talk about my second research project right now - but here's a teaser title for you:

"Unemployment in the Upper Tidewater: A Job Flows Explanation"... using Quarterly Workforce Indicators data from Census.