In search of search theory
This is going to be a long and wonkish post, so I’ll just give the dot-point summary here, and let those interested read on below the fold, for the explanations and qualifications.
* The dominant model of unemployment, in academic macroeconomics at least, is based on the idea that unemployment can best be modelled in terms of workers searching for jobs, and remaining unemployed until they find a good match with an employer
* The efficiency of job search and matching has been massively increased by the Internet, so, if unemployment is mainly explained by search, it should have fallen steadily over the past 20 years.
* Obviously, this hasn’t happened, but economists seem to have ignored this fact or at least not worried too much about it
* The fact that search models are more popular than ever is yet more evidence that academic macroeconomics is in a bad way
In search of search theory
Following the most recent Internet dust-up over the state of macroeconomic theory, I’ve been thinking a bit more about search models of unemployment. I first ran across these models when I was a student in the 1970s, a period of very high unemployment.
The basics of search models are simple and seemed, at least in the 1970s, reasonably realistic. Workers can look for jobs in various ways, all more or less time-consuming: looking at help wanted ads (which used to come out twice weekly in Oz), cold-calling potential employers, and asking friends and relatives to look out for openings. Since workers aren’t fully informed about the labour market, it doesn’t (at least in a good market) make sense to accept the first offer that comes along. Rather, it’s worth looking until you have a good idea what wage the market is willing to pay, then taking a job at that wage.
What was new about the search theories was the idea that this process could explain not only the inevitable frictional unemployment, but also cyclical unemployment associated with recessions. Various different ideas come into play here. First, the models imply that if there is a sudden but temporary shock leading to large-scale job losses, it will take some time to return to full employment (or ‘normal’ rates of unemployment). Second, economic shocks may create more uncertainty about market wages, leading workers to search longer.
The Internet changes all of this. It’s possible to find all the publicly advertised jobs in any given field, anywhere in the world, in a matter of seconds. With a little more effort, it’s possible to get lots of information about firms that may be hiring, even if they haven’t advertised vacancies. And, much more than in the past, its possible to get lots of information about potential employers, to assess whether the jobs they have on offer are in fact likely to be good ones.
With such a massive improvement in the efficiency of search we’d expect to see two things:
(i) shorter time spent searching, and therefore lower unemployment; and
(ii) better matches between workers and jobs, which should increase productivity and wages, and reduce subsequent quits and fires.
Both predictions were made in the early years of the Internet and, at least until 2008, the general view was that they were proving correct, though more slowly than had been expected.[^1]
But experience since 2008 has been completely the opposite of what a search model would predict. Unemployment rates have risen and employment rates have fallen. Worse still, the duration of unemployment has increased greatly. And there’s no evidence that this has been offset by an improvement in matching. Rather, lots of people have been forced to accept jobs that make little use of their education and experience, suffering wage losses as a result.
What has been the response of academic macroeconomists? As far as I can tell, almost nil. I was prompted to write this post by the debate over Kartik Arthreya’s Big Ideas in Macroeconomics. Responding to my observation that the word ‘unemployment doesn’t even appear in the index, Steven Williamson pointed out that Athreya spends several pages on the general properties of search models [^2], presenting them as the primary basis of unemployment theory. And Athreya seems to be on the money here. The Economics Nobel (yes, yes, Bank of Sweden etc, I know) is a lagging indicator, but the 2010 award to search theorists Diamond, Mortensen and Pissarides certainly does not indicate a view that such models are fundamentally flawed.
Athreya acknowledges the problem, sort of, by saying that “search is not really about searching” and observing that, if it were, the Internet should have reduced search costs. But if search isn’t about searching, what is it about? Athreya doesn’t say, and his brief discussion of housing markets doesn’t get us far – these are much more location-specific than job markets, and the Internet hasn’t changed the process all that much.
If search models aren’t the right way to think about unemployment, what is the right way? The simple answer is that unemployment is primarily a problem of macroeconomics not of labor markets. If aggregate demand is far below the productive capacity of the economy, workers will be unemployment and capital will be idle.
But there is still a puzzle here, one that search models were designed to solve. Why doesn’t competition between unemployed and employed workers work quickly to reduce wages to the point where demand equals supply and where there is no involuntary unemployment [^3]? The problem seems not be, as search models assume, that employers and potential workers don’t know about each other. Rather, it’s that employers can’t easily use the threat of new hires at lower wages to drive down the wages of existing workers (of course, this happens, but it’s clearly costly and risky in terms of worker morale). There’s quite a lot of literature looking at this, and I’ll try to post on it another time.
[^1]: Simple models of search can’t rule out cases where only one of these effects is seen. For example, the improvements in the quality of matches might be so great as to encourage people to spend even more time looking. But that doesn’t seem very plausible. It would imply a big increase in productivity and wages that hasn’t been observed. Here are a couple of references
Autor, D. H. (2001): \Wiring the Labor Market,” Journal of Economic Perspectives, 15, 25-40.
Krueger, A. B. (2000b): The Internet is lowering the cost of advertising and searching for jobs,” The New York Times, July 20.
[^2]: With his characteristic grace, Williamson also pointed out that another blogger had also observed the absence of standard macroeconomic topics in Athreya’s index and accused me of plagiarising this point for my own review. As any reviewer knows, the index is always the first place you look in a book, so its unsurprising that the oddity of Athreya’s jumped out at two of us.
[^3]: The same question applies to capital, though it seems to be asked far less often. Recessions and depressions are characterized by idle factories and farmland, empty offices and shuttered shops. Competition between the owners of these assets should drive the associated rents down to a level where supply equals demand. In fact, however, the adjustment process seems mostly to rely on scrapping or depreciating enough of the existing stock to remove the excess.