Unemployment, aggregate demand, and search/matching

For what it's worth, I thought I would lay out my own views, after reading John Quiggin and Noah Smith. This is what I think:

1. Fluctuations in the unemployment rate from year to year are mostly caused by fluctuations in aggregate demand. Very standard, boring, demand-side monetarist/keynesian view of the business cycle, that predates both monetarism and keynesianism, that most macroeconomists generally accept. An ideal monetary policy that eliminated those fluctuations in aggregate demand could, if it were possible, greatly reduce those fluctuations in unemployment associated with the business cycle.

2. Eliminating those fluctuations in unemployment caused by fluctuations in aggregate demand would also reduce the average level of unemployment. Booms and recessions are not fully symmetric. There is a zero lower bound on the number of people unemployed (the unemployment rate can't go negative), so recessions increase unemployment more than booms reduce unemployment. The economy is not linear, so Jensen's inequality says that reducing the variance will also reduce the mean.

3. But aggregate demand, and fluctuations in unemployment caused by fluctuations in aggregate demand and other shocks, can't be the whole story. We observe some unemployment even in economies where the unemployment rate is fairly stable over time. And attempts to reduce the unemployment rate to anywhere near zero do seem eventually to lead to accelerating inflation. And we observe some unemployment even in booms. There must be some other cause too. What else could cause unemployment, even in an economy where there are no aggregate demand shocks or other aggregate shocks?

4. Search theory is one possibility. But "search" theory should really be called "search/matching" theory. Because without heterogeneity of workers and jobs the search problem would be trivially easy. "You want a job?" "Yes. You want a worker?" "Yes." "Done!". And "search/matching" theory should really be called "wait/matching" theory. Because even if both sides of the labour market have perfect information about the other side -- about who's looking for a job and who's looking for a worker -- there might not be any suitable matches on the market right now, and one or other side might choose to wait until a better match appears on the market.

Kijiji tells me almost for free what used cars are for sale right now, but I might choose to wait, and keep watching Kijiji, until exactly the right Mazda MX6, with the V6 engine and manual transmission, and no rust or stupid body kits, and near to Ottawa etc. etc. appears on Kijiji. And the seller might also choose to wait until exactly the right buyer who wants exactly that sort of Mazda MX6 comes along who will offer a better price than a buyer who really wanted something a bit different. For people who really care about the particular characteristics of a car, the used car market is a very thin market. At any point in time, the number of buyers, or the number of sellers, may be precisely zero.

If anything, the fact that Kijiji is almost free to use encourages both sides to wait longer than they otherwise would. If Kijiji charged me a fee per view, I might be less picky about waiting for the best car. If Kijiji charged sellers a fee per day listed, they might be less picky about waiting for a buyer who really liked their particular car and would pay the higher price. More information, and cheaper information, might increase the unemployment rate and vacancy rate in the market for used cars. It could increase the quality of matches, rather than the speed of matches. It is like an outward shift in the Production Possibility Frontier for matches. Whether we spend it on quality, or speed, or a bit of both, depends on our preferences, and on what happens to the slope of that PPF.

If the only girl you will ever meet is the girl next door, and if you are the only boy she will ever meet, the two of you will either get married as soon as you want, or else exit the marriage market altogether. Full employment, or else out of the labour force altogether, with no unemployment in between. Neither of you will spend any time unmarried and searching for a good match.

Improved search technology need not reduce the unemployment rate, in a matching model. Because it increases the marginal benefits of waiting for a better match. The substitution effect can be bigger or smaller than the income effect.

5. But there are lots of other possibilities too, besides matching theory. Efficiency wage theories, monopoly union theories, etc. Probably no model is the whole truth.

Comments

[marris]

awesome post!

[Jacques René Giguère]

Nick: as a measured variable, Unemployment rate can't go smaller than zero.
But U is equal to (Participation ratio-Employment ratio)/Participation ratio. If people have a Participation ratio objective or expectations or some sociologically anchored rate, then a high employment ratio could lead to a perceived "negative" Unemployment rate. We usually call it an "overheating economy" and keeping Unemployment from going "negative" is a more-or-less-official goal of a lot of policy planners.They call it "possibility of upward wage pressures due to too tight a labor market" or something.

[Majromax]

The improved search possibilities may also amplify the effects of aggregate demand fluctuations. Employment stabilization from point #4 may be partially offset by destabilization from #2, with not-yet-eliminated AD fluctuations.

If a firm's demand is variable on a medium-term timescale but it takes a while to find a worker, then firms may have steadier and higher-on-average employment than a frictionless alternative.

If a firm experiences a demand increase that may be temporary, it costs little or nothing to begin the search for an employee under even "older" methodologies. While it might take some time to find the employee and complete the hire, if the demand increase proves transitory the hiring process can end at any point. This calculus is the same regardless of whether it's a long or short wait to find an employee.

If the firm instead experiences a demand decrease that may be temporary, however, the firm's actions differ depending on wait times. If it is easy and fast to hire a new worker, then it makes sense for the firm to lay off employees to meet the temporary demand drop: as soon as demand picks up again production can again return to normal. If, however, it is slow and painful to hire a new worker, then it makes sense for the firm to keep its employees on staff until the drop proves permanent -- slack capacity acts as insurance.

In an odd sort of way, the matching costs make firms act more like households. By making it difficult to react quickly to demand, firms have to make decisions based on expectations of permanent (long-term) income.

[Nick Rowe]

marris: thanks!

Jacques Rene: OK. But then we could ask the question as: "why does upward wage (and price) pressure start to appear even when the measured unemployment rate is above 0%?"

[Jason]

Really nice.

The issue I have with micro explanations though is that the recoveries from recessions in the US measured with unemployment have remarkable regularity:

http://informationtransfereconomics.blogspot.com/2014/07/remarkable-recovery-regularity-and.html

I find it hard to see how the internet, government intervention, monetary policy, etc can conspire in different combinations to give approximately the same rate of decline in unemployment after each recession over the past 70 years (approximately).

[Min]

As you present it, search/matching theory has an obvious flaw (to me, anyway). Namely that it assumes a fixed number of jobs. But if I, as a potential employer, know the characteristics of all potential workers, and vice versa, then new opportunities for jobs will arise from that knowledge. Hey! Let's get together and . . . . That sort of thing.

[Nick Rowe]

Jason: that's interesting. But search/matching theories etc are more like theories of the stable level of unemployment than theories about the speed of adjustment to that level following a recession.

Min: They don't normally assume a fixed number of jobs. The number of jobs can depend on real wages and the expected ease of hiring a worker. But they normally do assume an exogenous distribution of types of jobs, I think, to keep it simple. I don't think relaxing that assumption would totally change the results, unless you could costlessly change the type of job instantly, and hire two qualified bartenders instead of one brain surgeon.

[JohnQuiggin]

Thanks for some good discussion. I anticipated the "Mazda MX6" point in my post. It's certainly possible that improved search technology could lead to longer searches but if so we would expect to see much better matches, as in your example. That is, higher wages, greater productivity, and lower turnover. The last of these seems to be happening, but the first two are not.

[Min]

Thanks, Nick. :)

[Jacques René Giguère]

Nick: because it is a simplification for mere educationnal purposes.Reality is neither a linear nor a purely step function. Multiply the difference between PR and ER by some function(Econometricians,the Queen wants you!) and you'll get what you really really want (cue the Spice Girls).

[Nick Rowe]

John: thanks. Good point about productivity and real wages. Getting better matches ought to have increased those.

[Jason]

Nick: You're right of course. However it does seem unlikely to me that the theory that describes the level has nothing to do with the theory that describes the adjustment process (although one can't rule out the possibility that the fluctuations around the equilibrium are described by a different theory than the equilibrium itself -- I've seen that before).

The matching theory also seems to just depend on the numbers of job seekers and vacancies:

http://informationtransfereconomics.blogspot.com/2014/03/information-transfer-and-cobb-douglas.html

[Frances Woolley]

Interesting post

"If anything, the fact that Kijiji is almost free to use encourages both sides to wait longer than they otherwise would."

I suspect that's right, but I'm wondering if there aren't other effects of Kijiji as well - by lowering the transaction costs of buying a used car, it might persuade people to switch from buying a new car once every 12 or 14 years and holding onto it until it dies to buying used cars. In other words there might be more churning in the labour market as well. If, over time, people discover that the match isn't quite as good as they thought it was, it's easier to chuck it in and search for a new one.

[Nick Rowe]

Frances: Hmmm. That makes an interesting counterpoint to John Quiggin's comment above, about turnover. The initial match will be better, which reduces turnover (John's point). But finding a new match will be easier, which increases turnover (your point).

Cars age over time, so what was once a perfect match can become an imperfect match, if different people have different preferences over aging cars. Which suggests your effect should dominate for cars.

[Ben Kloester]

Your Mazda MX6 example is useful for illustrating how improved search/matching tech increases the marginal benefits of waiting for a better match, but the opportunity cost of waiting to buy a new car is a lot less than the opportunity cost of waiting to accept a job if you are unemployed.

My instinct is that the marginal increase in benefit seen from better search/match is an order of magnitude less than the opportunity cost of not working, so I'd be surprised if your posited effect of longer search/wait times ends up being strong, even with serious improvements in matching.

Not to mention that the search/match tech is also putting downward pressure on matching times in all the cases where the (hetergenous) market (for that one specific job/whatever) is not thin. Which time effect is stronger will depend on the 'pernickity-ness' of job-seekers and the efficiency gains from better search/match.

[Luis Enrique]

Search and Matching research group: http://sam.univ-lemans.fr/

[Dan]

This is a great post. More topics should be laid out like this.
Thank you
dan

[Jeff]

@Jacques René (&Nick): Maybe there is some "inside" reason for economists to allow that PR could be smaller than ER in the same formula, but from the outside for reasoning purposes that is a fallacy of equivocation somewhere (on "population", "ratio", "participation", and/or "employed" etc.). Because the only way that could happen is either you are counting some employed persons as "not participating", or you are using a different population count in the denominator between the two terms, or both. Both are absurd/wrong. Explanation for a layman?

[Nick Rowe]

Ben: I'm currently trying to think up a simple model/framework to help us get our intuitions clearer on that point, so we can at least see what it depends on.

Luis Enrigue; good find! Yep, the idea that it is search *and matching* seems to have taken hold.

Dan: thanks!

Jeff: economists sometimes say things that sound really weird, like "I estimate that full employment is 5% unemployment". It sounds like they are saying "5% = 0%". (They also say things like "output is currently above potential output", which sounds like they are saying that output is higher than the maximum possible level of output.) What they really mean when they say things like this is that 5% unemployment is as low as it can sustainably go using aggregate demand policies alone, and if you want it below 5% permanently you need some other policies. (And those other policies may or may not be good policies because, for example, a policy that forced the unemployed to work at cleaning litter for $1 per hour might reduce unemployment, but might not be a good policy.)

What they are doing is implicitly setting aside the other causes of unemployment (like search/matching) so they can talk about aggregate demand.

[Jacques René Giguère]

Jeff:to concur with Nick: I told explicitely about a PR target.
I also used this concept when studying QC economy vs ON. If in your calculation of UR, you replace the actuel PR with a target, you can get a perceived UR. So, I replace QC PR with ON and I get avery high UR forthrough the 60's instead of the rather low that was reported. It explained the political preoccupation about U in QC , even though it was officially low (about 6 in 1966) because the perceived rate was way above 10%. Today, perceived and official rates are equal,and are equal for both provinces. (And the Gazette had stopped whining that "separatists" had increased U, while in reality, whatever the political party, ER and PR increased faster in QC than ON over the last 50 years,
My spreadsheet was made available Stephen a couple of years back during another thread.
Hope I am clearer.

[Jacques René Giguère]

Corrections: "forthrough"; for QC through

"made available through Stephen"

[John]

It's very good that John has focused on this. At the same time, it seems to me that most of the discussion misses some fundamental things about the nature of matching: perfect vs. imperfect is a worthless characterization; acceptable vs. unacceptable gets more to the problem.

It's a well-known (among computer scientists like myself) but counterintuitive logical fact that matching decreases with the specificity of a search: a more specific search will yield fewer matches than a less specific search. This runs counter to the lay intuition that a more acceptable match will result from a more specific search, or that less acceptable matches will result from less specific searches.

Where job search/matching is concerned, and especially for more technical jobs, overspecified job postings are the norm, so much so that the industry is wide open to abuse (e.g., Google "purple squirrel"). And whether employers know better or not is largely moot where online search/match is concerned, because the software does what it does.

And what it generally does not do might help with the problem: it doesn't do general to/from specific matching. If job search/match algorithms would do more to broaden their matching to near matches and/or reasonable but "imperfect" matches, then imperfect but acceptable may be less mutually exclusive than they are with simple matching algorithms that run with the overspecificity they're given and as often as not come up empty.

Overall, it's not safe to assume that matching has improved; the truth might be just the opposite. And that makes perfect logical sense, albeit counterintuitively.

And by the way, online (and other) job search/matching is generally free to candidates, but generally not free to employers.

[Jeff]

Nick, Jacques René, thanks. I can understand changing the baseline to measure something other than "actual" or "absolute" (e.g. "perceived" or "relative to target"). I guess I was thrown by the usage "perceived 'negative' Unemployment rate" where the meaning was actually "negative perceived-Unemployment-rate". I.e. we're not talking about U itself any more but some variant pU or something. Thanks for taking the time to clarify for us lay people.