Labor market flows, business dynamics, and unemployment.
Davis, Steven J.
Measured from establishment-level data on employment gains and
losses, job creation and destruction average nearly 8 percent of
employment per quarter in the U.S. private sector. Worker flows in the
form of establishment-level hires and separations are more than twice as
large. (1) These facts summarize the remarkable extent of job and worker
flows in U.S. labor markets. They provide powerful motivation for
theories of frictional unemployment.
In recent research with several coauthors, I explore the
relationship of job flows to worker flows, develop methods to improve
the measurement of worker flows, investigate job loss and business
volatility trends, and provide new evidence on the determinants of
long-term movements in the unemployment rate.
Job Flows and Worker Flows in the Cross Section
Data from the Job Openings and Labor Turnover Survey (JOLTS)
display a very tight link between job flows and worker flows in the
cross section of employers. In Figure 1 we see that hires rise a bit
more than one-for-one with establishment-level job creation. Separations
rise a bit more than one-for-one with job destruction. (2) Further
investigation reveals that layoffs are the main margin of employment
adjustment for establishments with high job destruction rates, while
both quits and layoffs are important margins at moderate destruction
rates. Many studies find, not surprisingly, that layoffs are much more
likely than quits to result in unemployment spells. (3) Thus, higher
rates of job destruction bring higher layoff rates and greater worker
flows into unemployment.
Pitfalls in Measuring Worker Flows from Employer Survey Data
A striking feature of Figure 1 is the highly nonlinear relationship
of hires and separations to employer growth rates. These relations
exhibit pronounced kinks at zero, steep slopes moving away from zero in
one direction, and mild slopes with an opposite sign in the other
direction. Similar patterns hold for quits and layoffs.
These highly nonlinear relations create potential pitfalls in the
measurement of worker flows from survey data. To see the issue, observe
that aggregate hires, for example, are the weighted sum of hires at
establishments with different growth rates, with weights given by the
amount of employment at each growth rate. In order to accurately measure
aggregate worker flows, it is necessary to combine good estimates for
the type of cross-sectional relations in Figure 1 with an accurate
measure of the (weighted) cross-sectional distribution of employer
growth rates.
[FIGURE 1 OMITTED]
Using survey data to construct an accurate measure of the growth
rate distribution is challenging for two reasons. First, employer
surveys typically capture new establishments with a considerable lag.
Entrants account for a disproportionate share of hires and, more
generally, newer establishments exhibit a much higher incidence of
extreme growth rates. (4) Second, survey response rates are correlated
with employer growth rates in the cross section. More to the point, and
borrowing a line from Robert Hall: the first employee let go from a
declining establishment is the person who fills out government surveys.
For both reasons, employer surveys tend to produce growth rate
distributions with too little mass in the tails. Inspecting Figure 1, it
is easy to see why missing tail mass generates a downward bias in worker
flow estimates.
My coauthors and I study this issue in the JOLTS program, a leading
source of information about worker flows and job openings for the U.S.
economy. (5) We verify that the growth rate distribution generated by
the JOLTS sample has much less tail mass than that implied by the
comprehensive Business Employment Dynamics (BED) database. We also
develop a method to correct the problem. The key idea is to reweight the
cross-sectional distributions of employment growth rates in JOLTS to
match the corresponding distributions in the comprehensive BED.
Our adjusted statistics for hires and separations exceed the
published statistics by about one third. The adjusted layoff rate is
more than 60 percent greater than the published layoff rate. Our
adjustments significantly alter time-series properties as well.
Aggregate hires are 50 percent more variable than separations in
published JOLTS statistics, as measured by the variance of quarterly
rates, but 20 percent less variable according to our adjusted
statistics. Quarterly quit rates are more than twice as variable as
layoffs in published statistics but equally variable according to our
adjusted statistics.
Secular Declines in Job-Loss Rates before the Great Recession
American workers faced lower risks of job loss in the years leading
up to the Great Recession of 2007-9 than ten, twenty, or thirty years
earlier. I review some of the evidence for this claim in "The
Decline of Job Loss and Why It Matters." (6) There, my attention
centers on "unwelcome" job loss: employer-initiated
separations that lead to unemployment, temporary or persistent drops in
earnings, and other significant costs for job losers. Since there is no
fully satisfactory statistic for the incidence of job loss, I consider
several measures and data sources.
New claims for unemployment benefits as well as
employment-to-unemployment flows in the Current Population Survey show
dramatic declines in the risk of job loss since the 1970s and early
1980s. Job destruction measures from various sources also point to large
declines in the risk of job loss, with a generally downward drift since
the 1970s. (7) The much-studied Displaced Worker Survey is an outlier in
suggesting that essentially the entire long-term decline in the risk of
job loss reflects a recovery from the deep recession of the early 1980s.
Other measures point to continuing declines in the risk of unwanted job
loss long after the early 1980s. All of this evidence pertains to the
period before the Great Recession. Whether job loss rates will return to
relatively quiescent levels in the near future remains to be seen.
Business Volatility Trends: Privately Held Versus Publicly Traded
Firms
Declining rates of job destruction in the decades leading up to the
Great Recession appear puzzling when set against evidence that publicly
traded firms became more volatile over the same period. (8) My coauthors
and I tackle this puzzle using the Longitudinal Business Database (LBD).
(9) This comprehensive database contains annual employment observations
for all nonfarm establishments and firms in the U.S. private sector. The
LBD enables us to extend the study of business volatility to privately
held firms and, together with COMPUSTAT data, to distinguish publicly
traded from privately held firms.
We first use LBD employment data to confirm that business-level
volatility trended upward for publicly traded firms, rising more than 50
percent from 1978 to 2001. Our central finding, however, is a large
secular decline in the cross-sectional dispersion of business growth
rates and in the average magnitude of business volatility. This result
holds whether we define "businesses" in terms of firms or
establishments. Using the same measure as in previous research, the
employment-weighted mean volatility of firm growth rates fell by more
than 40 percent from 1982 to 2001.
Resolution of the puzzle turns on a remarkable finding: the large
upward trend in volatility among publicly traded firms is overwhelmed by
a large downward trend in volatility among privately held firms. It
turns out that widespread perceptions of deteriorations in employment
stability placed too much weight on developments at publicly traded
firms. Privately held firms, hitherto little studied in this context,
account for more than two-thirds of U.S. private-sector employment, and
they dominate the overall volatility trends.
Digging deeper, we find that two basic patterns hold across major
industry groups. First, the volatility and dispersion of business growth
rates are much greater among privately held firms. As of 1978, the
average standard deviation of firm-level employment growth rates is 3.7
times larger for privately held than for publicly traded firms. This
volatility ratio ranges from 2.3 in Services to 6.3 in Transportation
and Public Utilities. Second, volatility and dispersion decline sharply
among privately held businesses in the period covered by the LBD, and
they rise sharply among publicly traded firms. The overall
private-public volatility ratio falls to 1.6 by 2001, and it drops
sharply from 1978 to 2001 in every major industry group. In other words,
there was a pronounced "volatility convergence" between
privately held and publicly traded firms.
Employment shifts toward older businesses account for more than a
quarter of the volatility decline among privately held firms. The story
for publicly traded firms is very different. There was a large influx of
newly listed firms after 1979, with about 10 percent of listed firms new
each year from 1980 to 2001. Newly listed firms are much more volatile
than seasoned listings. Moreover, firms newly listed in the 1980s and
1990s exhibit greater volatility on an age-adjusted basis than earlier
cohorts.
These observations point to a major evolution in the economic
selection process governing entry into the set of publicly traded firms.
Indeed, we find that simple cohort dummies for the year of first listing
account for 67 percent of the volatility rise among publicly traded
firms from 1978 to 2001. Other researchers find that later cohorts of
publicly traded firms are riskier in terms of equity return variability,
profit variability, time from IPO to profitability, and business age at
time of first listing. (10)
Implications for Unemployment
In the canonical equilibrium model of search and matching in the
labor market, less job destruction means fewer job-losing workers,
smaller unemployment inflows, and lower unemployment rates. (11) It is
natural to ask--motivated by the trend declines in business volatility
and job destruction--whether this simple mechanism played a significant
role in the downward drift of U.S. unemployment rates after the early
1980s.
To address this question, my coauthors and I investigate the
low-frequency relationship of unemployment inflows to job destruction
and business variability measures. (12) At the aggregate level, the
secular decline in these measures roughly coincides with a marked
decline in the magnitude of unemployment flows. Inflows, for example,
fell from 4 percent of employment per month in the early 1980s to about
2 percent per month by the mid-1990s, and they remained low until the
Great Recession.
While suggestive, this aggregate relationship is confounded by
other factors that affect the evolution of unemployment flows, including
the aging of the workforce. (13) Thus, we turn to industry-specific
movements in unemployment flows and their relationship to
industry-specific movements in business variability and job destruction.
Unlike previous research on unemployment flows, ours focuses on
low-frequency relationships and interprets the evidence in light of
steady-state properties of a frictional unemployment model.
The industry-level data provide strong evidence that job
destruction and business variability measures can explain large changes
in the incidence of unemployment. For example, we estimate that a
decline of 100 basis points in an industry's quarterly job
destruction rate lowers its monthly unemployment inflow rate by 28 basis
points with a standard error of 4 basis points. This estimate reflects a
specification that controls for industry and time fixed effects.
Ignoring time aggregation, the estimate indicates that the response of
unemployment inflows over one quarter is 84 percent (three months times
28 basis points per month), as large as the movement in the number of
jobs destroyed.
To put this result in perspective, the quarterly job destruction
rate for the private sector fell 174 basis points from 1990 to 2005.
Multiplying this fall by its estimated effect in the industry-level
analysis yields a decline of 48 basis points in the unemployment inflow
rate. This response amounts to 55 percent of the drop in the
unemployment inflow rate from 1990 to 2005 and 22 percent of its average
value. Analogous estimates and calculations based on a different data
source imply that falling job destruction rates account for 28 percent
of the larger drop in unemployment inflow rates from 1982 to 2005. In
short, secular declines in job destruction rates were a major factor
behind the long-term drop in unemployment inflows.
What do these results say about the determinants of long-term
movements in the rate of unemployment? The average unemployment rate
fell by 43 log points from the period 1976-1985 to 19962005. Simple
accounting shows that this decline is almost entirely attributable to a
drop in the inflow rate. This accounting result, when combined with our
estimates, implies that the secular fall in job destruction explains
about a quarter to one half of the long-term decline in the aggregate
unemployment rate. In terms of the canonical equilibrium model of search
and matching, this result is consistent with a significant downward
trend in the intensity of idiosyncratic labor demand shocks in the
quarter century before the Great Recession.
(1) See S.J. Davis, R.J. Faberman, and J. Haltiwanger, "The
Flow Approach to Labor Markets: New Evidence and Micro-Macro
Links," NBER Working Paper No. 12167, April 2006, and Journal of
Economic Perspectives, 20(3) (Summer 2006) pp. 3-24; and S.J. Davis and
J. Haltiwanger, "Measuring Gross Worker and Job Flows," NBER
Working Paper No. 5133, May 1995, and Labor Statistics Measurement
Issues, J. Haltiwanger, M. Manser, and R. Topel, eds., University of
Chicago Press, 1999.
(2) Figure 1 is reproduced from S.J. Davis, R.J. Faberman, J.
Haltiwanger, R. Jarmin, and J. Miranda, "Business Volatility, Job
Destruction, and Unemployment," NBER Working Paper No. 14300,
September 2008, and American Economic Journal: Macroeconomics, 2(2)
(April 2010), pp. 259-87.
(3) See the discussion on pages 7-8 of Davis, Faberman, and
Haltiwanger (2006).
(4) See, for example, S.J. Davis and J. Haltiwanger, "Gross
Job Creation, Gross Job Destruction and Employment Reallocation,"
NBER Working Paper No. 3728, June 1991, and Quarterly Journal of
Economics, 107(3) (August, 1992), pp. 819-63; and S.J. Davis et al.,
"Measuring the Dynamics of Young and Small Businesses," NBER
Working Paper No. 13226, July 2007, published in Producer Dynamics: New
Evidence from Micro Data, T. Dunne, J.B. Jensen, and M.J. Roberts, eds.,
University of Chicago Press, 2009.
(5) S.J. Davis, R.J. Faberman, J. Haltiwanger, and L Rucker,
"Adjusted Estimates of Worker Flows and Job Openings in
JOLTS," NBER Working Paper No. 14137, June 2008, forthcoming in
Labor in the New Economy, K. Abraham, M. Harper, and J. Spletzer, eds.
(6) American Economic Review: Papers and Proceedings, 98(2) (May
2008), pp. 263-67. See also Figures 2 and 3 of Davis, Faberman, and
Haltiwanger, 2006; Figure 4 of R. Shimer, "Reassessing the Ins and
Outs of Unemployment," NBER Working Paper No. 13421, September
2007; Figure 2 of M. Elsby, R. Michaels, and G. Solon, "The Ins and
Outs of Cyclical Unemployment," NBER Working Paper No. 12853,
January 2007, and American Economic Journal: Macroeconomics 1 (1)
(January 2009), pp. 84-110.
(7) See Davis, Faberman, and Haltiwanger (2006); Davis et al.,
"Business Volatility, Job Destruction, and Unemployment" and
R.J. Faberman, "Job Flows, Jobless Recoveries, and the Great
Moderation," Federal Reserve Bank of Philadelphia Working Paper No.
08-11 (June 2008).
(8) See, for example, J.Y. Campbell et al., "Have Individual
Stocks Become More Volatile?" NBER Working Paper No. 7590, March
2000, and Journal of Finance 56(1) (February 2001), pp. 1-43; and D.
Comin and S. Mulani, "A Theory of Growth and Volatility at the
Aggregate and Firm Level," NBER Working Paper No. 11503, August
2005, and Review of Economics and Statistics 88(2) (May 2006), pp.
374-83.
(9) S.J. Davis, J. Haltiwanger, R. Jarmin, and J. Miranda,
"Volatility and Dispersion in Business Growth Rates: Publicly
Traded versus Privately Held Firms," NBER Working Paper No. 12354,
July 2006, and NBER Macroeconomics Annual 2006, Volume 21 2007.
(10) See, for example, E. Fama and K. French, "New Lists:
Fundamentals and Survival Rates," Journal of Financial Economics,
73(2) (August 2004), pp. 229-69; and G. Brown and N. Kapadia,
"Firm-Specific Risk and Equity Market Development," Journal of
Financial Economics, 84(2) (May 2007), pp. 358-88.
(11) The literature is vast. A seminal contribution is D.T.
Mortensen and C.A. Pissarides, "Job Creation and Job Destruction in
the Theory of Unemployment," Review of Economic Studies 61(3) (July
1994), pp. 397-415.
(12) S. J. Davis et al., "Business Volatility, Job
Destruction, and Unemployment," op. cit.
(13) See R. Shimer, "Why Is the U.S. Unemployment Rate So Much
Lower?" NBER Macroeconomics Annual 1998, Volume 13, 1999.
Steven J. Davis *
* Davis is a Research Associate in the NBER Programs on Economic
Fluctuations and Growth, Labor Economics, and Environmental and Energy
Economics, and a professor of economics at the University of
Chicago's Booth School of Business. His profile appears later in
this issue.