Why Peter Diamond Is Extremely Well Qualified to Be a Governor of the Federal Reserve
Peter Diamond on the current Beveridge Curve and related issues:
http://econ-www.mit.edu/files/6574: The Beveridge Curve shows the pattern of vacancies and unemployment over time. In economically good times we expect lots of vacancies and low unemployment, with bad times showing fewer vacancies and more unemployment. In the course of a business cycle, a movement from good times to bad and back again, we expect to see a loop around a curve, as shows up in a differential equation setting of a basic search model.
Figure 4 shows the empirical Beveridge curve for the US for the decade up to August 2010, with the open circles, connected by lines, for 2008-2010. Until the last 12 months in the figure, you can see the expected pattern of a recession, as vacancies shrink and unemployment rises, moving southeast roughly along a curve. Since then, we have had a rise in vacancies without a fall in unemployment.35 With rising vacancies and stable high unemployment, we are hearing claims that the US has just had a leap in structural unemployment – that the economy may have a long-term higher level of unemployment as the “new normal.” This inference is taken to imply that we should not be so concerned with stimulating aggregate demand through monetary and fiscal policies. For example, here is an August 17, 2010 statement by Narayana Kocherlakota, President of the Minneapolis Federal Reserve Bank.
What does this change in the relationship between job openings and unemployment connote? In a word, mismatch. Firms have jobs, but can’t find appropriate workers. The workers want to work, but can’t find appropriate jobs. There are many possible sources of mismatch—geography, skills, demography— and they are probably all at work. Whatever the source, though, it is hard to see how the Fed can do much to cure this problem. Monetary stimulus has provided conditions so that manufacturing plants want to hire new workers. But the Fed does not have a means to transform construction workers into manufacturing workers.
This statement has set off a flurry of reactions, to which I will add.
There is no surprise that we are hearing claims of higher structural unemployment – such statements appear when unemployment is high. A similar debate unfolded as I was a new student of economics. And in 1964, Bob Solow devoted his Wicksell Lectures to rebutting claims that the high unemployment in the late ’50s and early ’60s was structural rather than a result of inadequate aggregate demand. Indeed there is a long history of claims that the latest technological or structural developments make for a new long-term high level of unemployment, but these have repeatedly been proven wrong (Woirol, 1996).
It is likely to be more informative to think about the state of the labor market by focusing on the matching function, relating hires to unemployment and vacancies, rather than the Beveridge curve, which only considers the latter two. The natural interpretation is that the Beveridge curve movements would appear as a decrease in the efficiency of matching workers and jobs. Figure 5 shows the ratio of hires to vacancies over the last decade. Consistent with the picture from the Beveridge curve, over the last year we have had a drop in the rate of hiring relative to vacancies even though unemployment has stayed steady – a drop in the level of the matching function.
Note Kocherlakota’s statement that “Firms have jobs, but can’t find appropriate workers. The workers want to work, but can’t find appropriate jobs.” This is a static view of the labor market that does not make sense when thinking of the millions of hires that happen each month. While many workers, far too many workers, remain unemployed for a long time, many workers are finding jobs and many vacancies are being filled. Figure 1 shows that over the last two decades, on average, 37 percent of unemployed found employment each month. That percentage has dropped, but is still roughly 20 percent. Moreover, the large increase in the number of unemployed roughly offsets the fall in the exit rate, leaving monthly hires at a similar level to before - for the 12 months from November 2009 to October 2010, 5.7 million workers found a job per month, not hugely different from the 6 million average over the last 20 years. So we still have to think about large flows into and out of employment.
The matching function is not a technologically given structural relationship. Rather it is a reflection at the aggregate level of a complex and varied pattern of hiring at the level of individual employers and workers. Thus it is useful to examine some of the details at a less aggregative level to see how the current slump might be affecting the aggregate relationship - empirically we are outside the range of values of the ratio of vacancies to unemployment that were used in most estimates of the matching function. That is, a key question for interpreting the data in this recession and recovery compared to earlier ones is how the pattern of hires, unemployment and vacancies is different in recessions of different sizes and also different because of specific events, such as the large and continuing issues in both banking and housing markets.
The severity of the current recession in both depth and length has resulted in a great deal of long-term unemployment. Figure 6 shows the distributions of unemployment durations as of October 2010 and a year earlier, October 2009, before the Beveridge curve started moving vertically. The low vacancies we experienced raised long-term unemployment. In addition, we have had extended unemployment benefits. Such benefits somewhat reduce job search efforts and also discourage movement out of the labor force. Any lowering of the job search effort of the long-term unemployed is not likely to have much effect on aggregate unemployment, as there are many other workers who are seeking jobs and relatively few vacancies. And a reduction in the flow of unemployed out of the labor force increases measured unemployment, while having little effect on hiring, both because of the large numbers of remaining unemployed and because those outside the labor force take jobs as well. Reducing the flow of unemployed out of the labor force shifts the Beveridge curve up and the measured matching function down as hires divided by the number of unemployed is lower because the denominator is higher.
Long-term unemployment is very hard on the workers experiencing it and on their families. Moreover, over time, extended durations of unemployment affect behavior – the long-term unemployed are less good at maintaining their connection to employment and so we may have a slower-responding labor force after the economy grows significantly, which may be relevant for inflation concerns once we are nearing full employment, but not now. The deleterious effects of long-term unemployment are a reason to be particularly concerned about how long the economy does badly. Historically, recovery is slow after financial crises. The impact of a slow recovery on the long-term unemployed emphasizes the importance of stimulating aggregate demand enough to speed up recovery. And it emphasizes the importance of experimenting with programs to help the long-term unemployed find and hold jobs.
Just as measured unemployment does not fully reflect the availability of workers to be hired, so too the measured level of vacancies does not fully reflect the availability of jobs. Some hiring is done by firms that do not have measured vacancies, with some of these happening at firms that hire without posted vacancies, and some at firms that fill posted vacancies too quickly to be picked up in the data. John Haltiwanger provided me an estimate that about 40 percent of hires in the raw data are associated with establishments that begin the month with zero vacancies; with an estimated two-thirds due to the timing issue and the rest due to hiring without posting. Thus, measurement of the aggregate matching function may well vary with shifts in the makeup of hiring.
On a cross-section basis, the speed of hiring varies widely in systematic ways (Davis, Faberman and Haltiwanger, 2010). There are large differences across industries, with construction having a very high ratio of hires to vacancies, compared to industries like education and health. While generally cyclically sensitive, construction has been particularly hard hit this recession, which would lower the measured efficiency of the matching function compared with a time with a smaller relative impact on construction. Establishments that are growing fast fill vacancies much more quickly than those growing more slowly.
I do not know of data, but there may be a larger change in the mix of vacancies at fast and slow growing firms in this slump compared with smaller and less prolonged periods of high unemployment. Small firms are much more likely to hire without measured vacancies than large firms. Giuseppe Moscarini and Postel-Vinay (2010) report on the relative roles of large and small firms over the business cycle. Moscarini reports that gross job creation by large firms minus that by small firms has been unusually high, more so than in the other recoveries since 1980. This is consistent with a differential impact of credit market changes on firms of different size that seems to be happening.
The drop in house values has also impacted the ability of small firms to finance hiring by borrowing against the houses of the firm owners. The resulting smaller share of hiring by small firms lowers the measured matching function. Some types of positions are filled much more readily and rapidly than others. Hall (2010) has suggested that positions that a firm wants to fill after a quit are filled more quickly than newly created positions because quits are most likely to occur in high-turnover jobs with low and generic skills, such as fast-food restaurants.
And of course quits are way down, possibly reducing the average speed of filling jobs.
A key question for interpreting the pattern of aggregate unemployment and vacancy rates in this recession and recovery compared to earlier ones is whether the prime difference is in a changed difficulty of hiring at the disaggregated level or from a changed mix of diverse, but basically unchanged, hiring patterns across different firms and sectors, given that this is such a large and prolonged slump and with large and continuing issues in both the capital and housing markets. Complementing this analysis of hiring on a disaggregated basis is consideration of what Kocherlakota’s assertion would suggest might be found. Is there really a widespread difficulty in hiring in some industries or locations? I have not seen such reports.40 Thus we may be having shifts in the Beveridge Curve and the matching function that do not signal change for the underlying functioning of the economy once a recovery is well-established. That is, the pattern would return to normal after a sufficient rise in aggregate demand, apart from the lingering effects of long-term unemployment.
Having looked at the data, let me now look at possible policy inferences from whatever shifts may still be there. First, whatever one’s view on the magnitude of recent slippage in matching efficiency, more education, better education, good retraining all make for a more productive labor force and, done well at a reasonable cost, are policies to pursue. And carefully evaluated experiments in helping the long-term unemployed get and hold jobs seem likely to be worthwhile. Indeed a time of high unemployment is likely to be a time when further education is less socially costly by using time that would otherwise not be so well spent. The policy debate is not about whether to do more on the structural side; it is about what to do on the aggregate demand side, which is particularly an issue now with concern about projected long-run debt levels. Second, for the current moment, the argument about the aggregate demand side is academic, in the negative sense of the word. Current estimates I have seen of how much of the increase in unemployment from a few years ago is “structural,” rather than due to inadequate aggregate demand, still leaves enough need for aggregate demand stimulation that it is clear what direction is needed for further policies.
Third, I am skeptical of the value of attempting to separate cyclical from structural unemployment over a business cycle. When firms evaluate candidates for positions, they consider the quality of the match of available candidates, projections of the availability of new candidates, and the value to the firm of filling the slot. That is, the willingness to hire for a given quality of match depends on expectations about the profitability of investing in a new worker and about the likely pool of future applicants.
The tighter the labor market and the more valuable the filling of a vacancy, the more a firm is willing to hire a worker who is a less good match, who may need more training. In other words, a worker who might be viewed as structurally unemployed, as facing serious mismatch in the current state of the economy, may be readily employable in a tight labor market. The common practice of thinking about the extent of unemployment as a sum of frictional, structural and cyclical parts misses the point that the tightness of the labor market affects worker quitting decisions and affects employers’ willingnesses to hire an applicant who needs more training. Since direct measures of frictional or structural unemployment are dependent on the tightness of the labor market, they have limited relevance for the role of demand stimulation policies. The idea that the US economy is not adaptable and capable of dealing with the need for skills and jobs to adapt to each other is peculiar, given the long history of unemployment going up and down.
When the labor market is tight and firms have trouble finding workers, they reach out to places they have not looked before and extend training in order to find workers who can fill their needs. Supporting current stimulus policies as very good for the economy is entirely compatible with taking care to avoid future inflation...
