Recessions, Beveridge Cycles and Labor Market Adjustment, Part 2
More to matching efficiency than meets the eye...
An interesting, but deeply flawed analysis from the Peterson Institute for Economics has been brought to my attention. I will be addressing why the natural rate of unemployment has likely fallen, rather than increased, in a note later today. This note, was distributed to my institutional clients on 3 October 2020.
Email info@acrosstime.net or reply to this email to become an institutional client.
· Demand for labor depends, in part, on the cost of finding workers and that measure is where the Beveridge curve provides the most value, not as a prima facie leading indicator of economic activity.
· In 2019 employers reacted by to rising recruiting costs by reducing their spending on posting job openings. However, that does not mean their demand for labor was any lower than it had been before or that sufficient labor resources could not be found.
· The Beveridge curve is best used as an indicator of when the labor market has stopped deteriorating and has begun the process of rebuilding. A Beveridge curve indicates the U.S. labor market reacted to the COVID shock in a normal manner and is following past precedent in returning to a steady state.
New Regime, Same Beveridge Curve
Part 1 of this note examined “Beveridge cycles” prior to 1982. A “Beveridge cycle” analysis of the labor market views demand for labor as driven by expectations for revenue per match (“RPM”) relative to the cost of recruiting. As discussed in that note, the natural rate of unemployment rose after each of the recessions from 1950 to 1982 (Chart 1). The culprit was a decline in matching efficiency, which manifested as a rightward shift of the Beveridge Curve. After 1982, Beveridge cycles began tracing out a counterclockwise loop with structural unemployment falling each time (Chart 2). This note will examine the workings of the post-1982 Beveridge curve and implications for the U.S. in the near term.
Of course, the lines on the charts above are not showing us “the” Beveridge curve, because no single static curve exists. Rather, the scatterplot shows us where the shifting and twisting Beveridge curve and Job Creation Conditions curve intersect as they independently adjust to changing conditions. The factors affecting the shape and position of the two curves are shown in diagrams A, B and C below.
Expectations of higher revenue per match are self-fulfilling because firms are stimulated to open vacancies, which results in higher employment and higher RPM in the future. It would seem, the marginal cost of labor is the only impediment to further hiring as aggregate employment rises. However, the efficiency and elasticity of the labor market’s matching capabilities shift over the course of a business cycle. Congestion externalities at the peaks and troughs of business cycles drive up the cost of recruiting new employees, resulting in the cyclicality of matching efficiency and elasticity. Demand for labor depends, in part, on the cost of finding workers and that measure is where the Beveridge curve provides the most value, not as a prima facie leading indicator of economic activity.
Variable Search Effort
A major driver of recruiting costs is the varying effort put into searching by firms and workers across the business cycle. Changing recruiting and searching intensity drive changes in the elasticity of the Beveridge curve, represented by its slope in the scatterplot. Search intensity determines the number of hires that are likely to be made for each job vacancy posted.
Theoretically, worker job search effort should be pro-cyclical because wages are higher during an expansion, so the opportunity cost of not working is higher. However, two factors drive a counter-cyclical search effort (Charts 3 & 4). The first being that a more intense search effort is needed to generate the same return in a difficult job market. Average weekly job applications per worker are significantly higher in cities with more slack in their labor markets.
Second, during recessions, a compositional change skews the pool of job searchers towards people who search harder and try for longer. The composition shift allows aggregate search effort to rise without a cyclical change in behavior of any individual. The “high search types” find jobs easily during a boom so they disappear from the labor market. However, when unemployment is high, they are searching at full tilt. Employers are flooded with resumes by “high search types” during recessions and then struggle to find “low search types” in a tight labor market.
For firms, the cost of finding labor is determined by labor market tightness and the share of unemployed among jobseekers. In a tight labor market congestion is found among businesses competing fiercely for scarce labor resources. Competition for labor can make hiring so costly that firms reduce vacancies before reaching a steady state. If job destruction then overcomes job finding, the process is sent into reverse. Expectations for total employment will be lower – as will expected revenue per match - so firms reduce hiring. The process then becomes self-reinforcing.
In contrast, at the trough of a recession the market congestion is on the worker side of the labor market. The unemployed are eager to find a job so they submit eleven times as many applications as their employed peers. The unemployed spend an average of thirty-four minutes per day looking for work versus less than thirty seconds by the employed. Employed workers perform “on-the-job” sorting and will only switch from their current job if they believe another job is a better match. Self-sorting among workers reduces the cost of hiring for employers and increases demand for labor.
Conventional models assume the arrival rate of potential matches can rise without a deterioration of the quality of candidates. However unemployed workers apply for positions they might not be a fit for. Screening is roughly half the cost of recruiting, so the cost screening many unqualified applicants is significant. The higher share of mismatches during and soon after a recession causes job creation to be conditional on unemployment, reducing matching efficiency and pushing the Beveridge curve out. The cost of screening the flood of additional candidates outweighs the benefits of receiving them.
Beveridge Cycles Before 2001
As mentioned above, while the labor is tightening matching efficiency rises because the employed are self-selecting to high-quality matches. At the same time, the elasticity of recruiting is falling because a falling supply of available labor drives a rising marginal cost to find more labor. The result is that the Beveridge curve is moving to the left as efficiency increases and steepening as elasticity falls. This path is traced out by the Job Creation Conditions curve as it rotates clockwise due to rising demand for labor.
However, at some point, structural unemployment is reached, and no amount of recruiting will attract idle labor – the Beveridge curve is vertical. In periods prior to 1982 the result was a reversal of the cycle and a plunge into recession. After 1982 the situation changed, and employers were able to continue creating jobs – thus reducing unemployment – without spending more money on recruiting. The process is laid out in three steps in Diagram D below. The 1982-1992 and 1992-2001 Beveridge cycles trace out remarkably similar paths of the Beveridge curve, implying the repetition of a stable process (Charts 5 & 6).
In the JOLTS report, the “Openings” category includes jobs in newspapers, television, radio, internet, “help wanted” signs, “word of mouth” and contacting employment agencies. However, forty two percent of hiring takes place at establishments with no reported vacancies. Employers rely heavily on other instruments, in addition to vacancies, as they vary their hiring rate. Other instruments can include advertising, screening methods, hiring standards and compensation packages. These “recruiting intensity” factors affect applications per vacancy, applicant screening times and acceptance rate of offers. Investigating the cause of a declining natural rate of unemployment is beyond the scope of this note, but alternative recruiting channels could play a role.
The Post-2001 Super-Cycle
The 2001 to 2009 Beveridge cycle appears not to follow the counterclockwise loop observed in the two prior cycles (Chart 7). However, looking across the period from 2001 to 2019 it becomes clear that the mid-2000s labor market “cycle” was really just one portion of a larger super-cycle (Chart 8). That larger cycle does in-fact display the counterclockwise loop seen before, but with two sub-cycles encompassed by the same loop. In this case, the “transit periods” where the Beveridge curve suddenly flattens and shifts outward are 2001, 2008 and 2009 (Chart 9). Between those transit periods are “consolidation periods” where the labor market slowly returns to a steady state (Chart 10). Firms increase recruiting efforts as unemployment falls in a self-reinforcing cycle.
Matchmaking
Two factors can push the Beveridge curve out to the right: a rise in separations or a decline in matching efficiency. One might expect that separations would rise during a recession as layoffs create unemployment, but the opposite takes place as workers stop quitting their jobs (Chart 11). The implication is that the rightward shift of the Beveridge curve during recessions is driven by a decline in match efficiency rather than an increase in separations.
Chart 12 below uses the job finding rate for the very-short-term unemployed as a proxy for matching efficiency. Matching efficiency is clearly procyclical, but also displays structural changes as the labor market’s efficiency varies over long periods of time. Note that in 2019 matching efficiency still had a way to go before reaching levels seen prior to the Great Recession. There may have been further room for employment growth via improved matching efficiency. Some of the additional demand for labor would be satisfied by reducing the unemployment rate. However, the very low level of unemployment in 2019 meant that an increase in matching efficiency would need to be fed by pulling non-participants into the labor force. Whether that would happen or not remains an unknown.
By 2019, matching efficiency was moving upwards, and aggregate employment was still growing – demand for labor was strong. Employers did pull back on their job postings and hiring plans as the economy slowed coming out of 2018 (Chart 13). However, this was done because the cost of finding new workers was rising rapidly. Producer price inflation for employment services was at boom levels (Chart 14) and employers were reporting serious difficulties filling open positions (Charts 15 & 16). Matching elasticity had fallen to zero and the Beveridge curve became vertical by 2018. In 2019 employers reacted by reducing their spending on posting job openings. However, that does not mean their demand for labor was any lower than it had been before or that sufficient labor resources could not be found.
Conclusion
The implications of the discussion above, and my other analysis on the state of the U.S. economy in 2019[1], is that the decline in job postings seen in 2019 was a function of rising search costs rather than falling demand for labor (Chart 17). The result could have been a grinding down of structural unemployment or an increase in participation. Alternatively, the Beveridge curve could have stopped shifting and additional demand for labor could only elicit higher wages. In either case, the Beveridge curve was not a warning sign of oncoming recession.
Looking across the available data and literature it appears that the Beveridge curve is not a useful tool for forecasting changes in business activity. The number of job openings does not drive the unemployment rate because it is not solely determined by the demand for labor. The fact that a plurality of hiring is done by firms with no jobs posted is evidence enough of that. The cost of search plays a major role in determining the resources firms are willing to pour into recruiting new employees. When finding the right employees is difficult because there are too many or too few job seekers, the relationship between unemployment and job openings is different than it is mid-expansion. T
The Beveridge curve is best used as an indicator of when the labor market has stopped deteriorating and has begun the process of rebuilding. Indeed, that very indication was provided in May when the curve began the long climb back to where it had been pre-COVID (Chart 18). Note that job posting have reached their 2018 peaks despite unemployment being very high.
The labor market is healing but has not returned to steady state yet. Watch for signs of continued migration of the Beveridge curve to the left as unemployment falls. If unemployment stops falling while job openings remain stable the implication is that the labor market remains unsettled.
References
Ahn, Hie & Crane, Leland. (2020). Dynamic Beveridge Curve Accounting.
Diamond, Peter A. and Sahin, Aysegul, Shifts in the Beveridge Curve (August 1, 2014). FRB of New York Staff Report No. 687.
Engbom, Niklas. 2019."Application Cycles,"2019 Meeting Papers 1170, Society for Economic Dynamics.
Ghayad, Rand & Dickens, William. (2012). What Can We Learn by Disaggregating the Unemployment-Vacancy Relationship?
Lubik, Thomas, The Shifting and Twisting Beveridge Curve: An Aggregate Perspective (October 3, 2013). FRB Richmond Working Paper No. 13-16.
Raines, Richard & Jungho Baek, 2016. "The Recent Evolution of the U.S. Beveridge Curve: Evidence from the ARDL Approach," Review of Economics & Finance, Better Advances Press, Canada, vol. 6, pages 14-24, August.
Sniekers, Florian. (2018). Persistence and volatility of Beveridge cycles. International Economic Review.
[1] See my note “Was the USA Headed for Recession?” of 27 July 2020.