Full text of Working Papers (Federal Reserve Bank of Chicago) : Gross Job Creation, Gross Job Destruction and Employment Reallocation, Working Paper 1991-5

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

GROSS JOB CREATION, GROSS JOB
DESTRUCTION AND EMPLOYMENT
REALLOCATION

Steve J. Davis and John Haltiwanger
Working Paper Series
Macro Economic Issues
Research Department
Federal Reserve Bank of Chicago
March, 1991 (WP-91-5)

G ro ss J o b C re a tio n , G ro ss J o b D e s tru c tio n
a n d E m p lo y m e n t R eallo catio n *

by

S tev e J . D avis

J o h n H a ltiw a n g e r

Graduate School of Business

Department o f Economics

University of Chicago

University of Maryland

Chicago, IL 60637

College Park, MD 20742

(312) 7027312

(301) 454-6307

February 1991
Previous Drafts: December 1988 and October 1989

* In preparing the data for this study, we have greatly benefited from the assistance
of Bob Bechtold, Tim Dunne, Cyr Linonis, Jim Monahan, A1 Nucci and other Census Bu­
reau employees at the Center for Economic Studies. We have also greatly benefited from
comments on previous drafts by an anonymous referee, Katherine Abraham, M artin Baily,
Fischer Black, T im Dunne, Larry Katz, David Lilien, Robert McGuckin, Kevin M. Mur­
phy, Ariel Pakes, Robert Topel, John Wallis, and workshop participants at the University
of Maryland, the University of Chicago, MIT, Princeton University, Stanford University,
Yale University, the Federal Reserve Bank of Chicago, the Resource M obility Session of
the Econometric Society (W inter 1988 m eetings), an N BER conference on Alternative
Explanations of Employment Fluctuations, and the N B E R ’s Economic Fluctuations Pro­
gram M eeting (Summer 1989). Scott Schuh provided excellent research assistance. Kevin
Murphy provided the gross worker flow data from the March-March matched files of the
CPS. We gratefully acknowledge the financial assistance of the National Science Founda­
tion (SES-8721031 and SES-S720931), the Hoover Institution, and the Office of Graduate
Studies and Research at the University of Maryland. Davis also thanks the National Sci­
ence Foundation for its support through a grant to the National Fellows Program at the
Hoover Institution. Much of the research for this paper was conducted while Davis was a
National Fellow at the Hoover Institution.



ABSTRACT

T his study m easures the heterogeneity o f establishm ent-level em ploym ent changes in
the U .S. m anufacturing sector over the 1972 to 1986 period. We m easure this heterogeneity
in terms o f the gross creation and destruction o f jobs and the rate at which jobs are
reallocated across plants. Our measurement efforts enable us to quantify the connection
between job reallocation and worker reallocation, to evaluate theories o f heterogeneity
in plant-level em ploym ent dynam ics, and to establish new results related to the cyclical
behavior o f the labor market.




I. I n tr o d u c tio n
This paper measures the heterogeneity of establishm ent-level employm ent changes in
the U.S. manufacturing sector over the 1972 to 1986 period. We measure this heterogeneity
in terms of the gross creation and destruction of jobs and the rate at which jobs are
reallocated across plants. Our measurement efforts enable us to quantify the connection
between job reallocation and worker reallocation, to evaluate theories o f heterogeneity
in plant-level employm ent dynamics, and to establish new results related to the cyclical
behavior of the labor market.
Our empirical work exploits a tremendously rich data set w ith approxim ately 860,000
annual observations on 160,000 manufacturing establishm ents. The data axe longitudi­
nal and include observations on all manufacturing establishm ents sampled in the Annual
Survey of Manufactures between 1972 and 1986. The com bination of establishm ent-level
longitudinal data, high frequency observations, a fifteen-year sample, and comprehensive
coverage of the manufacturing sector provides an excellent basis for developing the im pli­
cations of heterogeneity in establishm ent-level employment dynamics.
A key aspect of our study is its focus on gross job flows as opposed to gross worker
flows. Previous studies have docum ented the tremendous gross worker flows across labor
market states (i.e., employm ent, unemployment, out of the labor force) and high worker
turnover rates . 1 In the absence of evidence from longitudinal establishm ent data, it has
been difficult to determine whether large gross worker flows primarily reflect temporary
layoffs and recalls plus continual sorting and resorting of workers across a given set of jobs
or, alternatively, whether a large portion of worker turnover is driven by the destruction
and creation of employm ent opportunities.
The results that emerge from our study are striking.

Based on March-to-March

establishm ent-level employm ent changes, we calculate that manufacturing’s rates of gross
job creation and destruction averaged 9.2% and 11.3% per year, respectively. We show
that these figures reflect sim ultaneously high rates of job creation and destruction w ithin
narrowly defined sectors of the economy, e.g., four-digit industries. The impressive mag­
nitude of gross job creation and destruction has been docum ented before, perhaps most
convincingly at high frequencies by Leonard (1987) and at low frequencies by Dunne,
Roberts and Samuelson (1989b).
Summing the rates of gross job creation and destruction yields our measure of the
job reallocation rate, i.e., the rate at which employment positions are reallocated across
1See Clark and Summers (1979), Abowd and Zellner (1985), Poterba and Summers (1986),
Lilien (1980), Hall (1982), Darby, Haltiwanger and Plant (1985), Akerlof, Rose and Yellen
(1988), and Blanchard and Diamond (1990).



1

establishm ents. T he high rates of job reallocation found in this paper indicate that the
reshuffling o f em ploym ent opportunities across plants is one o f the m ost im portant reasons
that workers change employers or transit between em ploym ent and joblessness. Combining
inform ation from the LRD and the Current Population Survey, we calculate bounds on the
fraction of worker reallocation accounted for by job reallocation. Our calculations reveal
that 35-56% o f all worker reallocation between employers or betw een em ploym ent and
joblessness arises to accom m odate shifts in the distribution of em ploym ent opportunities
across work sites.
Tw o other findings docum ented below provide insight into the character of the worker
reallocation associated w ith job reallocation.

One finding is that m ost of annual job

creation and destruction reflects persistent establishm ent-level em ploym ent changes. For
exam ple, 73% o f the jobs created between March 1974 and March 1975 still existed in
March 1976, and 72% of the jobs lost in the 1974-75 interval were still lost in March 1976.
T he average one-year persistence rates for annual job creation and destruction are
and 81%, respectively.

6 8

%

This persistence indicates that the bulk of annual job creation

and destruction cannot be im plem ented by temporary layoff and recall policies. A second
finding is that job destruction is highly concentrated - only 23% is accounted for by
establishm ents that shrink by less than twenty percent over the span of a year. T his finding
indicates that the bulk of job destruction cannot be accom m odated by normal rates o f
worker attrition. Taken together, the concentration and persistence results im ply that job
reallocation is typically associated w ith long-term joblessness an d /o r worker reallocation
across employers.
T he impressive m agnitude of job reallocation and its bearing on worker reallocation
lead us to inquire into the sources o f heterogeneity in establishm ent-level em ploym ent
changes. We docum ent strong relationships between the intensity o f job reallocation and
observable plant characteristics like age, size and ownership type (single-unit versus m ulti­
unit firm). We also draw on several theories of plant-level heterogeneity and dynam ics
to identify reasons for sim ultaneous job creation and destruction w ithin narrowly defined
sectors o f the economy. Guided by these theories, we quantify the contribution o f various
sources of heterogeneity to total job reallocation and to variation in job reallocation across
groups of establishm ents defined in terms of industry, region, age, size and ownership type.
One prominent theory of heterogeneity in plant-level employm ent dynam ics stresses
the selection effects associated w ith passive learning about initial conditions . 2 We develop
a procedure for estim ating the fraction of total job reallocation accounted for by this source
2See Jovanovic (1982), Lippman and Rum elt (1982), and Pakes and Ericson’s (1990) ver­
sion of the Jovanovic m odel.



2

of heterogeneity in plant-level employment dynamics. The procedure combines information
on the distribution of employment by plant age and the rate of job reallocation by plant
age w ith sim ple and plausible identifying assumptions. D espite the attention that these
theories have received in recent empirical work , 3 we find that passive learning about initial
conditions accounts for only 11-13% of observed levels of job reallocation. In results more
favorable to this type of theory, we find that leaning about initial conditions accounts for
roughly one-third to one-half of the differences in job reallocation rates across groups of
plants defined in terms of industry, size, region and ownership type.
Long traditions in labor and industrial economics view plants w ithin industries, re­
gions or employer size classes as relatively homogeneous. Theories o f vintage effects view
plants as relatively hom ogeneous w ithin age groups. These perspectives suggest an expla­
nation for high rates of job reallocation as the natural consequence of continually occurring
sector-specific shocks, where sectors are defined in terms o f industry, region, size or age.
To evaluate this explanation, we com pute the fraction of excess job reallocation accounted
for by between-sector employm ent shifts. Excess job reallocation is defined as total job
reallocation minus the minimum amount required to accom m odate the net change in em­
ploym ent. Remarkably, we find that essentially none of the excess job reallocation in U.S.
manufacturing can be accounted for by employment shifts among two-digit industries,
Census geographic regions, eight age classes or five size classes. Even when we define sec­
tors in terms of 450 four-digit manufacturing industries, between-sector employment shifts
account for a mere 12% of excess job reallocation. Similar results hold when we define
sectors in terms of both two-digit industry and either age, size, region or ownership type.
The inability of either sectoral shock theories or theories that stress learning about
initial conditions to account for observed rates of job reallocation leads us to the following
conclusion: Any successful explanation for the m agnitude of job reallocation m ust also
explain why sim ultaneously high rates of job creation and destruction occur among mature
plants in narrowly defined sectors of the economy.
The impressive m agnitude of job reallocation and its bearing on worker reallocation
also lead us to inquire into the connection between the business cycle and the intensity
of job reallocation. In this regard, a key finding is that the job reallocation rate exhibits
significant countercyclic tim e variation. The March-to-March job reallocation rate for the
manufacturing sector ranges from a low of 17% in 1980 to a high of 23% in 1975 and 1983.
The sim ple correlation between net employment growth and the job reallocation rate is
-0.57.
3See Evans (1987ab), Hall (1987), Dunne, Roberts and Samuelson (1989a), and Pakes and
Ericson (1990).




3

We carry out several empirical exercises designed to address the question o f why the
job reallocation rate fluctuates countercyclically. T hese exercises establish two im portant
sets of results. First, the countercyclic behavior of job reallocation reflects tim e variation in
the m agnitude o f idiosyncratic plant-level employm ent m ovem ents, not sectoral differences
in the m ean em ploym ent responses to aggregate disturbances. Second, patterns of tim e
variation in job reallocation intensity differ sharply by plant age, size and ownership type.
Job reallocation rates among young (0-9 years), small (1-249 em ployees), and single-unit
plants exhibit no system atic relationship to the cycle. Job reallocation rates am ong older,
larger and m ulti-unit plants exhibit pronounced countercyclic patterns o f variation.
These results enable us to discriminate between macroeconomic theories that cannot
explain the observed cyclical behavior of job reallocation and theories that potentially can.
We conclude that standard macroeconomic theories that specify hom ogeneous firms or ho­
m ogeneous firms w ithin sectors cannot account for the tim e variation in job reallocation
intensity. Nor can cyclic m ovements in job reallocation intensity be explained by theories
that treat the idiosyncratic component of firm-level employm ent behavior as orthogonal
to the business cycle. As we discuss below, theories that stress the frictions associated
w ith the reallocation o f workers and jobs across employers im ply potentially im portant in­
teractions between aggregate employm ent growth and the pace o f reallocation. Blanchard
and Diam ond (1989, 1990), Davis and Haltiwanger (1990), and Caballero (1990) develop
theories of this sort that can explain some of the cyclical job flow findings in this paper.
We turn now to a description of the data and the gross job flow measures that we use
in this study.

I I . D a ta a n d M e a s u re m e n t
A . The L on gitu din al Research D atafile

T his study exploits annual, plant-level employm ent observations in the Longitudinal
Research D atafile (LR D ). The LRD sam pling frame encom passes all U.S. manufacturing
establishm ents w ith five or more employees. These establishm ents account for ninety-nine
percent of manufacturing employm ent, based on tabulations from either the Census of
M anufactures or County Business Patterns.
The LRD is basically a series of contiguous five-year panels w ith annual data on m any
manufacturing establishm ents, plus Census-year data on the universe of m anufacturing es­
tablishm ents. Census years in the LRD are 1967, 1972, 1977, and 1982 - annual data are
available from 1972 to 1986. From the Census-year universe, the Bureau draws a sam ple
of establishm ents that are then surveyed during five successive years. T his five-year panel,
which com m ences two years after a Census year, comprises the sam ple of establishm ents



4

that makes up the Annual Survey of Manufactures (ASM ). New establishm ents are added
to the panel as it ages to incorporate births and to preserve the representative character
of the panel. In 1977, the LRD included roughly 70,000 out of the 360,000 manufactur­
ing establishm ents. These sampled establishm ents accounted for 76% of manufacturing
employm ent.
W ith respect to the five-year ASM panels, establishm ents fall into three broad groups.
As noted, the group containing establishm ents w ith fewer than five employees is excluded
from the sam pling frame. A second group of establishm ents is included in the panel with
certainty. For the 1979-83 panel, for example, the certainty group includes all establish­
ments w ith 250 or more employees during the 1977 Census year. This certainty threshold
is lower in some industries, and many establishm ents are included with certainty based on
other criteria. Taken as a whole, the certainty cases account for about two-thirds o f man­
ufacturing employm ent during the 1979-83 period. Establishm ents that fall into neither
of the first two groups are sampled with probabilities proportional to a measure of size
determined for each establishm ent from the preceding Census. Sampling probabilities for
non-certainty establishm ents range from 1.000 to 0.005. We use sample weights, equal to
the reciprocals of the sam pling probabilities, whenever we aggregate over establishm ents.
Some, but not m ost, of the non-certainty establishm ents appear in contiguous pan­
els. Thus, our ability to link establishm ent-level observations across panels ranges from
excellent for large establishm ents to quite poor for the sm allest establishm ents. This ob­
servation implies that accurate measurement of gross employment changes is more difficult
in the first year of each panel. W hile it is possible to construct continuous series for basic
measures of job creation and destruction, and we have done so in Davis and Haltiwanger
(1990), some of the cross-tabulations presented below cannot be constructed for the first
year of a panel. Hence, we typically calculate the gross and net change measures reported
in this paper from a sam ple that excludes 1974, 1979 and 1984.
Several key features of the LRD enable us to largely overcome the selection and
measurement problems that have hampered most previous attem pts to estim ate gross
rates of job creation and destruction from plant-level or firm-level data. In this regard,
the L R D ’s key features are the comprehensive scope of its sam pling frame for a major
sector of the U.S. economy, large probability-based samples that minimize sam pling error,
the incorporation of births into ongoing panels, a careful distinction between firms and
establishm ents, and a careful distinction between ownership transfers and the birth and
death of establishm ents. Among U.S. studies on job creation and destruction, Dunne,
Roberts and Samuelson (1989b) provide the only other measurements based on a data



5

set with similar virtues. Their work exploits the Census-year observations in the LRD to
calculate five-year job creation and destruction rates.4
B. M easu rem en t o f G ross Job C reation, D estru ctio n , and R eallocation

We now introduce some notation and define measures o f establishm ent size and growth
rate. We then plot the empirical growth rate density and relate it to job creation and de­
struction m easures. We also describe the connection between these m easures and measures
of worker and job reallocation.
We m easure the size of establishm ent e at tim e t, denoted by x e*, as the sim ple average
of establishm ent em ploym ent at tim e t and t — 1 .

Sector size is defined analogously.

We define the tim e-t growth rate of establishm ent e, denoted by g et, as the change in
establishm ent em ploym ent from t — l to t, divided by x et. This growth rate measure
is sym m etric about zero, and it lies in the closed interval [—2 , 2 ] w ith deaths (births)
corresponding to the left (right) endpoint. A virtue of this measure is that it facilitates
an integrated treatm ent of births, deaths and continuing establishm ents in the empirical
analysis. T he g m easure is m onotonically related to the conventional growth rate measure,
and the two measures are approxim ately equal for sm all growth rates . 5
4
Figures l.A and l.B plot frequency distributions for the establishm ent growth rate
observations in our eleven-year sample.

Figure l.A depicts the shape of the empirical

density over the 677,000 annual observations on g€t • Figure l.B depicts the shape of
the empirical density over the size-weighted observations on ge t. B oth the weighted and
unweighted densities are slightly asymmetric w ith central peaks in the interval surrounding
zero and endpoint spikes corresponding to births and deaths.
On an unweighted basis, 25% o f all manufacturing establishm ents experienced a
growth rate in the interval (-.05,.05), and 46% experienced a growth rate in the inter­
val (-.15,-15). Births and deaths account for 14% of annual growth rate observations on
m anufacturing establishm ents. The m ass o f the size-weighted distribution is much more
concentrated about the center and much less concentrated in the tails. On a size-weighted
basis, 29% of the annual growth rate observations fall in the interval (-.05,.05), and 63%
fall in the interval (-.15,.15). Births and deaths account for only 2.4% of all size-weighted
4D avis and Haltiwanger (1989) discuss the weaknesses in other data sets that have been
used in U.S. studies o f job creation and destruction. For a full discussion o f data quality
issues pertaining to our use of the LRD, see Davis, Haltiwanger and Schuh (1990).
5Let G be the change in employm ent divided by lagged em ploym ent, i.e., the conventional
growth rate measure.
<

2 7 /(2

The two growth rate measures are linked by the identity G =

- g ).




6

growth rate observations . 6 Evidently, establishm ent turnover and employm ent volatility
Eire sharply declining functions o f establishm ent size in our sam ple, a result that is consis­
tent w ith work by Evans (1987ab), Hall (1987), Dunne, Roberts and Samuelson (1989ab),
and others.
The gross job flow measures investigated in this paper have a simple relationship to
the size-weighted frequency distribution o f establishm ent growth rates. We calculate gross
job creation by sum m ing employm ent gains at expanding and new establishm ents w ithin
a sector. Similarly, we calculate gross job destruction by sum m ing employment losses at
shrinking and dying establishm ents within a sector. To express these measures as rates,
we divide by sector size. Introducing some additional notation, we can write gross job
creation and destruction rates in sector s at tim e t as
P 0 S 3t =

5 2 ( lT ~ ) 9 e t,

and

9et>0

NEG„ =

Y . (
e€£«t

%et
x

at

)|yet|>

Set <®

where E at is the set of establishm ents in s at t . 7 As these formulas indicate, the sizeweighted frequency distribution determines the weight to attach to each growth rate value
in the calculation of job creation and destruction rates.
Two remarks are helpful in thinking about our job creation and destruction measures.
First, it seems apparent that year-to-year changes in establishm ent-level employm ent are
largely induced by changes in desired establishm ent size rather than by temporary move­
m ents in the stock of unfilled positions. For this reason, P O S at and N E G at directly reflect
the reallocation of employm ent positions or jobs, and not the reallocation of workers. Of
course, one m otivation for our research is that the reallocation of jobs partly drives the
reallocation of workers. Thus, the job reallocation concept in this paper differs from, but is
6Two caveats should be borne in mind when interpreting this aspect of the size-weighted
density. First, our size metric ( x e t) assigns only half as much weight to observations on
births and deaths as would a more conventional size metric.

For exam ple, if we were

to ask what fraction of current employment is located at establishm ents born w ithin the
past year, the birth category would appear twice as important as in Figure l.B . Second,
although births and deaths account for a small fraction of size-weighted establishm ent
growth rate observations, they account for a large fraction o f gross job reallocation. We
return to this point in section III.C.
7Sample weights are suppressed in these formulas to reduce notational clutter.




7

related to, the worker turnover concepts considered by Lilien (1980), Hall (1982), Akerlof,
Rose and Yellen (1988), and others. We spell out the contribution o f job reallocation to
worker reallocation in section III.C.
Second, since we observe only plant-level em ploym ent, we cannot determ ine whether
a given level o f em ploym ent in two different periods for the sam e plant represents the
sam e or different em ploym ent positions. T his observation and the point-in-tim e nature of
the em ploym ent data im ply that P O S at and N E G at represent lower bounds on true job
creation and destruction rates.
We use the sum of P O S at and N E G at, S U M at , to measure the gross job reallocation
rate in sector s betw een t — 1 and t. X atS U M at equals the gross change from t — 1 to
t in the number o f em ploym ent positions at establishm ents. In term s o f the frequency

distribution, the job reallocation rate S U M at can be thought o f as the size-weighted mean
of the absolute value o f establishm ent growth rates.
To relate job reallocation to worker reallocation, observe that X atS U M at represents
an upper bound on the number of workers who change jobs or switch between em ploym ent
and nonem ploym ent in response to establishm ent-level em ploym ent changes . 8 X atS U M at
represents an upper bound because som e workers move from shrinking to growing estab­
lishm ents w ithin sector s between t — 1 and t. To obtain a lower bound, we elim inate
the possibility of double counting job losers who move directly to new jobs at expanding
establishm ents in the same sector. That is, X at M A X at = X atM a x { P O S at , N E G at } rep­
resents a lower bound on the number of workers who change jobs or em ploym ent status in
direct response to job reallocation in sector s. In line w ith this discussion, we often refer to
S U M at and M A X at as upper and lower bounds on the worker reallocation rate required

to accom m odate job reallocation. W hen interpreting these upper and lower bounds, it
is im portant to recognize that the worker reallocation associated w ith job reallocation is
itself a lower bound on total worker reallocation. Worker reallocation arises in response
to life-cycle, career path, job satisfaction, and m atch quality considerations as well as in
response to job reallocation.

I I I . S o m e E le m e n ta ry F a cts a b o u t J o b C re a tio n a n d D e s tru c tio n
T his section of the paper lays out some elem entary facts about job creation and
destruction behavior in the U.S. manufacturing sector. We relate these facts to the m ag­
nitude and character of the worker reallocation associated w ith job reallocation. These
facts also set the stage for the analysis in the succeeding sections o f the paper.

8The interpretation of X atS U M at as an upper bound is subject to the qualifications about
the lower-bound nature of P O S at and N E G at discussed above.



8

A . M agn itu de and T im e V ariation

Table 1 presents annual rates of job creation and destruction, net employm ent growth,
job reallocation, and a lower bound on the worker reallocation required to accom m odate
job reallocation. The figures in Table

1

and elsewhere in this paper are based on March-

to-March changes in establishm ent-level employment.
The centred fact captured by Table

1

is the phenomenon o f simultaneous job creation

and destruction. Every year of the sample exhibits both job creation and job destruction
rates that exceed six percent of manufacturing employm ent. In 1973, when manufacturing
employm ent expanded by a robust seven percent on net, the gross job destruction rate was
six percent. In 1975, when manufacturing employm ent shrank by a dramatic ten percent,
the gross job creation rate was seven percent.
The last two columns in Table 1 point out the tremendous reallocation of jobs and
workers associated w ith sim ultaneous job creation and destruction. The job reallocation
rate ranges from 17.3% in 1980 to 23.3% in 1975.

Substantial worker reallocation is

required to accom m odate job reallocation of this magnitude. The lower bound on the
required rate of worker reallocation ranges from

. % of employm ent in 1980 to 16.6% in

1 0 2

1975. Thus, the heterogeneity of establishm ent-level employm ent movements illustrated
in Figures

1

translates into an impressive amount of worker reallocation.

One other noteworthy fact emerges from Table 1 : The pace o f job reallocation exhibits
significant countercyclic tim e variation. The range of variation in job reallocation over the
eleven years of the sample is six percentage points. The sim ple correlation between the
net job growth rate and the job reallocation rate equals -.57. Given the m agnitude of
job reallocation, its significant tim e variation, and the countercyclic pattern of the tim e
variation, one is led naturally to inquire about the connection between the pace o f job
reallocation and aggregate employment fluctuations. We take up this inquiry in section
V.
B. C ross-In du stry V ariation

Table 2 presents average annual net and gross job flow measures for the manufacturing
sector and each two-digit industry. The industry figures are X u -weighted averages of the
eleven annual industry observations, and the figures for the manufacturing sector are X ,weighted averages of the industry figures.
Employment contracted in every two-digit manufacturing industry over the sample.
Annual net contraction rates range from .2 % in Instruments to 5.4% in Primary M etals.
The manufacturing sector as a whole declined at a rate of 2 . 1 % per year. D espite perva­
sive net contractions, every two-digit industry experienced significant gross job creation.
Average March-to-March gross job creation rates range from




9

. % in Tobacco to 12.9%

5 8

in Lumber and W ood Products. March-to-March gross job destruction rates range from
7.8% in Paper to 16.0% in Lumber and W ood Products. In the m anufacturing sector as a
whole, gross job creation and destruction rates averaged

9

. 2 % and

. %, respectively.

1 1 3

T he annual average job reallocation rate shows considerable cross-industry variation,
ranging from 14.0% in Tobacco to 28.8% in Lumber and W ood Products. T he lower bound
on the rate o f worker reallocation required to accom m odate observed job reallocation
ranges from 8.9% in Chemicals and Paper to 18.8% in Lumber and W ood Products. For
the m anufacturing sector as a whole, the lower (upper) bound on the required rate of
worker reallocation equals 12.9% (20.5%) of employm ent per year.
The final colum n of Table 2 shows that sim ultaneous job creation and destruction is
an im portant phenom enon in every two-digit manufacturing industry. T his colum n reports
average industry rates of excess job reallocation, i.e., the mean difference betw een total job
reallocation and the minimum job reallocation required to accom m odate net em ploym ent
changes.

T he excess job reallocation rate varies from 9.8% to 20.6% across two-digit

industries. T he size-weighted average o f the two-digit industry excess job reallocation rates
equals 15.2% of em ploym ent. These striking facts, and their bearing on worker reallocation,
provide strong m otivation for an inquiry into the underlying sources o f the establishm entlevel heterogeneity responsible for sim ultaneous job creation and destruction. We take up
this inquiry in section IV.
C. The C on n ection to Total W orker Reallocation

The preceding results indicate that a substantial fraction of total worker reallocation
is demand driven in the sense of being induced by shifts in the distribution of em ploy­
m ent opportunities across work sites. To quantify this statem ent, we now compare the
total number o f persons who switch jobs or employm ent status to the number o f switches
required to accom m odate the reallocation of jobs.
Recall that our job reallocation figures axe based on em ploym ent changes over a
twelve-m onth interval. A meaningful comparison requires a consistent measure o f total
worker reallocation. W ith this observation in m ind, we calculate total worker reallocation
as the sum of two pieces. The first piece is the number of persons who have job tenure
of twelve m onths or less. Based on the Current Population Survey (C P S), Hall (1982, p.
317) reports that this number is 28.2% of employm ent in 1978. T he second piece is the
number o f currently jobless persons who were employed twelve m onths earlier. Sum m ing
these two pieces yields the total number o f persons who currently have a different job
an d /or em ploym ent status than they had twelve m onths earlier.
To calculate the second piece of total worker reallocation, we tabulated M arch-toMarch gross worker flows from the CPS. Gross worker flows refer to the number o f persons



10

who report a change in labor force status - employed, unem ployed, or out of the labor force
- between survey dates. Using the March-March matched files of the CPS, we obtained the
three-by-three m atrix of gross flows, F , for fifteen pairs o f years between 1968 and 1987.
Since reporting errors are known to cause a substantial upward bias in the measured flows,
we adjusted the F matrices following Poterba and Summers (1986). Letting Q denote the
three-by-three m atrix of classification error probabilities, the measured and true gross
flows satisfy the relationship F = Q 'F * Q , where F* denotes the true flows. Obtaining
Q from Table 3 in Poterba and Summers, we estim ate the true gross flow m atrix in

year t as Ft = ( Q - 1 )'F t Q ~ x. Collapsing unemployment and out of the labor force into
a single category, we then calculate the yearly number of transitions from em ploym ent
to joblessness as a percentage of employment. Averaging this transition rate over the
fifteen years, we estim ate that the number of currently jobless persons who held a job
twelve months earlier as
classification error is

. % of employment. (T he corresponding figure unadjusted for

8 6

. % of em ploym ent.)

1 1 2

Summing the two pieces, total worker reallocation equals 28.2 +

8 .6

= 36.8 percent of

employment in a typical yeax. From Table 2, the amount of worker reallocation required
to accom m odate job reallocation is bounded between 12.9% and 20.5% o f employm ent
in a typical year.

Hence, taking the ratio of the job reallocation figures to the total

worker reallocation figure, we calculate that 35-56% of total worker reallocation arises to
accom m odate shifts in the distribution of employment opportunities across work sites.
Simply put, job reallocation accounts for a major fraction o f total worker reallocation . 9
Tw o observations provide further perspective on the m agnitude of job reallocation’s
contribution to worker reallocation. First, our calculations neglect secondary waves of
worker reallocation initiated by job creation and destruction. For exam ple, a person who
quits an old job in favor of a newly-created job potentially creates a chain of further quits
9Three sources of potential bias in our calculations seem sufficiently im portant to merit
mention. First, Hall’s job tenure figure understates worker m obility (for our purposes),
because it does not include workers who, within the past twelve m onths, transferred be­
tween plants owned by the same employer. Third, the attrition rate in the March-March
matched files of the CPS may be higher for workers who change employm ent status. B oth
of these effects bias the denominator of the calculated ratio downward. Third, our job
reallocation figures are based on the manufacturing sector only. According to Leonard’s
(1987, Table

. ) tabulations for W isconsin, annual job reallocation rates are 28% higher

6 6

in nonmanufacturing than in manufacturing. Thus, Leonard’s results suggest that the
numerator of our calculated ratio significantly understates the job reallocation rate in the
economy as a whole.



11

as other workers re-shuffle across the new set of jobs. It follows that the direct plus indirect
contribution o f job reallocation to total worker reallocation exceeds the figure we derived
above.
Second, a certain amount of worker reallocation inevitably arises from life-cycle con­
siderations as old workers retire and young workers enter the workforce. If the typical
person works forty-five years, then retirement and initial labor force entry directly cause
transitions betw een employm ent and nonemployment equal to roughly 4.4% o f the work­
force in a typical year. It follows from our figure for total worker reallocation that sim ple
life-cycle effects account for roughly 12% of total worker reallocation. After accounting
for job reallocation and life-cycle effects, the residual amount o f worker reallocation equals
11.9-19.5% o f em ploym ent, or 33-53% o f total worker reallocation. T his com ponent of
worker reallocation reflects temporary exits from the workforce and the sorting and re­
sorting o f workers across existing jobs for a variety of reasons.
We conclude this discussion w ith a caveat. Recall that our job and worker reallocation
figures are based on changes between two points in tim e twelve m onths apart. Carrying out
similar calculations for data based on, say, m onthly sam pling would place greater em phasis
on seasonal disturbances and other factors that lead to transitory flows o f workers and
jobs. To the extent that these factors disproportionately affect worker or job flow rates,
a different calculation of job reallocation’s contribution to total worker reallocation would
emerge.
D . C on cen tration and P ersisten ce

T he high rates of job reallocation reported in Tables

1

and 2 prompt two further

factual questions. First, what role do plant births and deaths play in the creation and
destruction of jobs? Or, to restate the question in a more general way, how are job creation
and destruction distributed by establishm ent growth rate? Second, do the high rates of
job creation and destruction reported in Tables

1

and 2 reflect primarily transitory or

persistent establishm ent-level employm ent changes? We address these questions in turn.
Gross job creation and destruction axe distributed over establishm ents experiencing
the full range o f expansion and contraction rates. Figure 2 displays the distributions of
job creation and destruction over this range. The right half o f Figure 2 plots the fraction
of job creation accounted for by establishm ents experiencing growth rates in the intervals
[0, .1), [.1, . 2 ) , . . . [1.9,2.0). A final category shows the fraction of job creation accounted
for by establishm ent births. The left half of Figure 2 provides a sym m etric partition of
gross job destruction.



12

Figure 2 highlights two noteworthy aspects of job creation and destruction. First,
both large discrete changes and smaller incremental changes account for significant frac­
tions of job creation and destruction. Establishm ents experiencing m odest growth rates
(| < 1 < .20) account for 29% of job creation and 23% of job destruction. Establishm ents ex­
7
periencing dram atic growth rates (|g| > 1.0) account for 28% of job creation and 34% o f job
destruction. Births (deaths) alone account for 20% (25%) o f job creation (destruction ) . 1 0
Second, Figure 2 reveals a clear asymmetry between the distributions o f job creation
and destruction by establishm ent growth rate. Relative to job creation, job destruction
exhibits greater concentration at establishm ents that experience dramatic growth rates.
This aspect of job creation and destruction behavior provides support for theories o f plantlevel employm ent dynamics that generate greater lum piness in employm ent contraction
than employm ent expansion.
We now turn to the persistence of the March-to-March establishm ent-level employ­
ment changes that underlie our annual job creation and destruction measures. The per­
sistence question is especially pertinent to an assessment o f the character o f worker reallocation associated w ith job reallocation. To the extent that job creation and destruction
represent short-lived establishm ent-level employment changes, these changes can be im ple­
mented largely through temporary layoffs and recalls. To the extent that establishm entlevel employm ent changes are persistent, they must be associated w ith long-term jobless­
ness an d /or worker reallocation across plants.
In thinking about how to measure persistence, we stress that our focus is on the
persistence of the typical newly-created or newly-destroyed job.

This focus is distinct

from a focus on the persistence of the typical existing job (e.g., Dunne and Roberts, 1989)
or the persistence of establishm ent size (e.g., Leonard, 1987). In line with our focus, we
measure persistence as follows. Let F P O S ti denote the fraction of newly-created jobs in
March of year t that continue to be present in March of year t + l . 1 1 Also, let F P O S t 2
10An earlier version of this paper reports partitions o f job creation and destruction by
year and partitions by two-digit industry.

These more detailed results show that the

im portant role of dramatic establishm ent-level employment changes illustrated in Figure
2

is pervasive across industries and years. For example, the fraction of job destruction

accounted for by establishm ent deaths ranges from 14-36% across years and 15-35%, on
average, across two-digit industries.
1 1 Let E M P et denote tim e-t employment at establishm ent e. Newly-created jobs at e in
t equal E M Pet — E M P e, t - i , assuming positive growth. If E M P e,t+ i > E M P et, then all

of these newly-created jobs are present in t +
of the newly-created jobs are present in t +



1

13

1

. If E M P e,t+ i < E M P e, t - i , then none

. If E M P e,t+ i G [ E M P e^ - i , E M P et), then
t

denote the fraction o f newly-created jobs in March o f year t that are present in March of
year t +

1

and March of year t + 2. Define F N E G tn analogously.

Table 3 reports the persistence measures for a set o f base years determ ined by the
lifecycle o f the ASM panels. T he key fact captured by the table is the highly persistent
nature o f the establishm ent-level employm ent m ovements underlying annual job creation
and destruction. To take the m ost pronounced exam ple, the one-year persistence rate for
jobs destroyed betw een March 1980 and March 1981 is

8 8

%, and the two-year persistence

rate for these lost jobs is 82%. T he average one-year persistence rates for newly-created
and newly-destroyed jobs are

6 8

% and 81%, respectively.

T hese facts on concentration and persistence shed further light on the connection
between job reallocation and worker reallocation. Since only 23% of job destruction is
accounted for by establishm ents that shrink by less than twenty percent over the span of a
year, the bulk of job destruction cannot be accom m odated by normal rates o f worker attri­
tion. Since annual job creation and destruction primarily reflect persistent establishm entlevel em ploym ent changes, the bulk of annual job creation and destruction cannot be
im plem ented by tem porary layoff and recall policies.

IV . E x p la n a tio n s fo r S im u lta n e o u s J o b C re a tio n a n d D e s tru c tio n
T he preceding section established that job reallocation is large in m agnitude and that
it accounts for a large fraction of total worker reallocation. T his section investigates the
sources of establishm ent-level heterogeneity that lead to sim ultaneous job creation and
destruction w ithin industries. We draw on several theories o f plant-level heterogeneity
and dynam ics to identify potential driving forces behind sim ultaneous job creation and
destruction. We then quantify the contribution of various sources of heterogeneity to total
job reallocation and to variation in job reallocation across groups of establishm ents defined
in terms of industry and other observable characteristics.
A.

Theories o f H eterogen eity th a t E xplain Sim ultaneous Job C reation and D e stru c tio n

One prominent theory of heterogeneity in firm- and plant-level em ploym ent dynam ics
stresses the selection effects associated with passive learning about initial conditions. In
this type of theory, plants face ex ante uncertainty about certain cost parameters or their
level of efficiency. A plan t’s underlying efficiency level cannot be directly observed but is
learned over tim e through the process of production. A plant that accum ulates favorable
E M P e<t + 1 — E M P ett - 1 of the newly-created jobs are present in t +

1

. Carrying out this

calculation for all growing establishm ents in t and dividing the result by P O S t yields
FPOSn.



14

inform ation about its efficiency expands and survives, whereas a plant that accumulates
sufficiently unfavorable information chooses to exit. W ell-articulated theories of this sort
include Jovanovic (1982), Lippman and Rumelt (1982), and Pakes and Ericson’s (1990)
version o f the Jovanovic m odel. Much of the empirical analysis in recent studies of firmlevel and plant-level employm ent dynamics is explicitly couched in terms of this type of
theory - see Evans (1987ab), Hall (1987), Dunne, Roberts and Samuelson (1989a), and
Pakes and Ericson (1990).
As a stand-alone theory, passive learning and selection cannot explain perpetual plant
turnover w ithin an industry. Eventually, plants learn their underlying efficiency level and
decide whether to exit or remain indefinitely. The transitory, idiosyncratic cost distur­
bances present in the Jovanovic model generate transitory, plant-level em ploym ent fluctu­
ations that continue indefinitely, but the existence of sunk costs associated w ith entry and
exit insures that the set of surviving plants eventually becomes fixed in the absence of some
other type of disturbance. Hence, we view passive learning and selection as a mechanism
that magnifies the job reallocation and plant turnover response to other disturbances. For
example, passive learning might explain why an industry that experiences steady growth
also exhibits gross job destruction associated with plant deaths.
Another reason for the co-existence of job creation and destruction is technical in­
novation that leads to the replacement of old, outm oded plants by new, technologically
superior plants. Sim ultaneous job creation and destruction accompanies the technological
upgrading and plant turnover process. Bresnahan and Raff (1990) pursue this them e in
their analysis of technological heterogeneity in the American auto industry during the
1920’s and 1930’s. However, Dunne, Roberts and Samuelson (1989b, Table 5) present ev­
idence that the rate of job destruction associated w ith plant death declines as plants age.
This fact is difficult to reconcile with a simple monocausal theory o f plant turnover and
job reallocation driven by exogenous technical change, but it does not preclude a major
role for technical change in a full explanation for job reallocation and plant turnover.
Ericson and Pakes (1989) and Pakes and Ericson (1990) develop a theory of firm and
industry dynamics in which investment outcom es involve idiosyncratic uncertainty. The
stochastic outcom es of an individual plant’s investm ents, coupled w ith com petitors’ invest­
ment outcom es, determine the probability distribution over future profitability streams.
A plant’s investm ent outcom e may improve its position relative to its com petitors, thus
leading to expansion, or it may cause a relative deterioration, thus leading to contraction
and, possibly, exit. Investm ent in the Ericson-Pakes m odel thus involves elem ents of ac­
tive learning and selection. Unlike the passive learning and selection m odel of Jovanovic
(1982), the Ericson-Pakes m odel builds in an explanation for perpetual entry and exit 


15

the outside industry or com petitors stochastically, but exogenously, advance along an effi­
ciency path. Hence, the active learning theory em beds technical change into a rich m odel
of plant-level heterogeneity and selection.
Another class o f theories stresses differences in initial conditions, or uncertainties
about future conditions, that lead firms to commit to different factor intensities and pro­
duction techniques. T hese differences in turn lead to heterogeneity in plant-level responses
to common cost and dem and shocks. These sources o f heterogeneity, and their connection
to the sim ultaneous entry and exit of plants, are nicely analyzed in Lambson (1990).
Finally, even plants that produce identical products w ith identical technologies can
face idiosyncratic cost disturbances. As exam ples, energy costs and tax burdens are often
heavily influenced by local conditions. Exogenous, idiosyncratic cost disturbances lead to
contraction at som e plants and, simultaneously, expansion at other plants. In related work,
Davis and Haltiwanger (1990), we develop a general equilibrium m odel o f em ploym ent
reallocation and job turnover driven by exogenous, idiosyncratic cost disturbances.
The preceding remarks identify several theories or factors that plausibly account for
sim ultaneously large job creation and destruction rates w ithin narrowly-defined sectors
o f the economy. W hile a full assessment of each theory is beyond the scope o f a single
paper, we exploit several observable plant characteristics to quantify the contribution
of som e potentially im portant factors to job reallocation. In addition to industry, the
observable plant characteristics we consider are plant age, size, geographic region, and
ownership type. (Ownership type refers to whether a plant is owned by a single-plant firm
or a m ulti-plant firm.) We interpret these plant characteristics as observable correlates of
technical change, choice of production technique, differences in initial conditions, locationspecific disturbances, organizational scale, and the progressive resolution of uncertainty
about initial conditions.
B. V ariation by Region, Size, A ge and O w nership Type

Table 4 displays net and gross job flow rates cross-tabulated by plant size, age, own­
ership type, and geographic region.

The rightm ost column reports the distribution of

m anufacturing em ploym ent by plant characteristic. Except for plant age, the figures in
Table

4

represent average annual rates over the eleven years in our sam ple. Since our

ability to construct detailed age categories is greatest in the last year o f com plete panels,
we report averages of 1978 and 1983 values for the age figures.
According to Table 4, every region except the M ountain region experienced net job
loss over the sam ple period. The variation in net job loss rates is quite sm all across plants
o f different average sizes and ownership types. In contrast, net job loss rates vary greatly
by plant age. Young plants grow rapidly on average, while older plants shrink on average.



16

The gross job flow measures exhibit strong patterns of variation w ithin each grouping
of plants in Table 4. T he western geographic regions exhibit noticeably higher job reallocation rates than the rest of the country. Job reallocation rates for single-unit plants
are half again as large as reallocation rates for plants operated by m ulti-unit firms. Job
reallocation rates decline sharply with average establishm ent size, ranging from 14% at
plants w ith 1000+ employees to 30% at plants with 1-99 employees.
The m ost dramatic variation in gross job flow rates occurs w ith respect to plant age.
For plants that are one-year old in the base year, the annual job reallocation rate averages a
remarkable 48%.12 The job reallocation rate drops off rapidly to roughly 26% by age three,
and it declines further to 16% for plants that axe at least fifteen years old. Unreported
results reveal that this sharp relationship between plant age and the job reallocation rate
is pervasive across two-digit industries, geographic regions, plant size classes, and plant
ownership types.
T hese facts about variation in job reallocation rates by plant characteristic are con­
sistent w ith the existing literature on heterogeneity in firm dynamics. Our measure of
dispersion and our scheme for weighting establishm ent-level observations differ from pre­
vious studies, but sharp declines in employment volatility w ith plant size and age are
robust findings in the literature. In seeking explanations for sim ultaneous job creation
and destruction, the especially sharp and pervasive relationship between job reallocation
and plant age im pels one toward theories that can also explain this fact. Theories based on
passive learning and selection suggest an interpretation of this fact as the natural outcom e
of a progressive resolution of uncertainty about initial conditions. In the next section, we
quantify the extent to which this type o f theory can explain the m agnitude o f job reallocation and the variation in job reallocation rates across industries, regions, plant size
classes, and plant ownership types.
C. Q uantifying the Role o f P a ssive Learning about In itia l C on dition s

Consider the following counterfactual question: How much would gross job realloca­
tion be dim inished if selection effects associated with passive learning about initial condi­
tions were absent from the economic environment determining firm dynamics? We provide
an answer to this question by bringing some simple identifying assum ptions to bear on the
age-related information in Table 4.

12This figure is not inflated by including re-openings of previously idled plants. As the
LRD enables us to track re-openings o f older plants, their contribution to job creation and
reallocation is allocated to the appropriate plant age category.



17

Iden tifyin g A ssu m p tio n s and M ethodology

If plants accum ulate information over tim e about an unknown, but tim e-invariant,
cost parameter, then the posterior distribution eventually converges in probability to the
true value. A ssum e that this convergence process is largely com plete w ithin n years o f
plant birth. This is our central identifying assumption. It follows from this assum ption
that none of the job reallocation among m ature plants (age > n) reflects selection effects
associated w ith passive learning about initial conditions.
Now, consider how we m ight exploit this identifying assum ption to answer the counterfactual question. Besides the passive learning mechanism , m any factors contribute to
sim ultaneous job creation and destruction w ithin industries or sectors o f the economy.
Except as described below, we assum e that these other factors have age-neutral effects on
job reallocation rates. This assum ption means that these other factors generate the sam e
base job reallocation rate for younger and older plants. Thus, as our second identifying
assum ption, we take the “base” reallocation rate to be age invariant.
Combining the two identifying assum ptions, the fraction of job reallocation caused by
passive learning and selection is

Eq<n g (q )[K q) - r(aSe > ” )]

^2

rX

m
a<n

H a) ~ r(ase - n)l /r>

()
1

where x ( a ) / X is the ath age group’s share of sectoral em ploym ent, r(a) is the measured job
reallocation rate of age group a, and r denotes the sectoral job reallocation rate. T he term
r(age > n ), equal to the measured job reallocation rate among m ature plants, represents
the base rate o f job reallocation that is assumed to be age invariant. Thus, the formula
counts all job reallocation in excess of the base amount as arising from passive learning
effects.
Besides passive learning, other factors may lead to age-nonneutral effects on the job
reallocation rate.

For example, consider the following characterization o f an industry

equilibrium like the one articulated by Jovanovic. Suppose that industry dem and and
em ploym ent grow at a constant rate through tim e. Job reallocation am ong m ature plants
arises because of the transitory, idiosyncratic cost (or demand) disturbances present in
the Jovanovic m odel. Among younger plants, however, job reallocation also arises because
o f selection effects associated with passive learning about initial plant conditions. Thus,
along the stationary growth path for the industry, new plants continuously enter to ac­
com m odate net industry expansion and to replace dying plants. Because o f the ongoing
selection process associated w ith passive learning, gross job destruction exceeds zero, and



18

gross job creation exceeds net job creation by an equal am ount. Since job reallocation
among m ature plants reflects transitory and idiosyncratic disturbances, there is no rea­
son to expect sharply different job reallocation rates among younger plants in response
to these disturbances. However, given diminishing returns at the plant level, net long-run
industry growth will occur entirely through the entry o f new plants rather than through
higher job creation rates among existing plants. Thus, long-run net growth generates agenonneutral effects on job reallocation rates. For this particular case, the adjustm ent of ( 1 )
is straightforward and given by

P =

^2 ~
x

m
a<n

"

~ r (age - n )l ~

s

/r,

()
2

where g denotes the net industry employment growth rate. This alternative formula counts
all job reallocation in excess of the base amount and the amount required to accom m odate
net industry expansion as arising from passive learning effects.
The appropriate adjustment is less clear for a contracting industry, because industry
contraction is likely to occur through shrinkage (and death) among plants of various ages.
Taking these considerations into account, we modify our second identifying assum ption by
assum ing that (a) net industry contraction has age-neutral effects on the reallocation rates,
and (b) net industry expansion has the age-nonneutral effects described above. In line with
this modification, the empirical results below estim ate the fraction of job reallocation due
to passive learning and selection as

[r(a) - r(age > n)] - max{$r, 0 } /r .

P =

(3)

.a<n
Tw o additional remarks regarding the use of (3) to estim ate passive learning’s contri­
bution to job reallocation are in order. First, it is conceivable that factors other than net
industry growth lead to age-nonneutral effects on the job reallocation rate. These factors
potentially bias our estim ate of passive learning’s contribution to total job reallocation . 1 3
To the extent that these age-nonneutral factors reflect transitory or industry-specific dis­
turbances, their im pact on the calculation of P for the entire manufacturing sector will
13

Age-nonneutral disturbances to employment growth rates, as opposed to reallocation

rates, do not bias the estim ate of passive learning’s contribution. For exam ple, a techno­
logical innovation that causes an equal rise in gross job destruction among m ature plants
and in gross job creation among young plants has offsetting effects on the calculation of
P in (l)-(3 ).



19

be negligible. To the extent that some unspecified factor system atically causes higher job
reallocation rates am ong young plants, our estim ate of passive learning’s contribution will
be upwardly biased.
Second, stink costs associated w ith plant entry im ply that transitory fluctuations
in industry dem and will be largely accom m odated by the expansion and contraction of
existing firms. T he industry response to these disturbances is unlikely to involve a sharply
age-nonneutral response in job reallocation rates. Hence, we interpret g in equations (2)
and (3) as the long-run net growth rate. Empirically, we estim ate g as the average annual
employm ent growth rate in our sam ple for the industry or sector.
E m pirical R esu lts

We im plem ent equation (3) using pooled sam ple data from 1978 and 1983, the only
years for which we can tabulate the r ( a ) function by the detailed age categories in Table
4. Carrying out the calculations for n = 4 years, we find that selection effects associated
w ith learning about initial conditions account for

1 1

% of total job reallocation in the U.S.

m anufacturing sector. R epeating the calculations under the assum ption that plants com ­
pletely learn their underlying efficiency level by age six, learning about initial conditions
explains 13% of job reallocation in the manufacturing sector. T he key finding contained
in these results is that learning about initial conditions explains only a sm all fraction of
total job reallocation.
This finding is unlikely to be overturned by reasonable m odifications of our procedure
or identifying assum ptions. Table 4 indicates why: nearly nine-tenths of manufacturing
employm ent is located at plants more than six years old, yet these plants exhibit substantial
job reallocation rates. Learning about initial conditions is not a plausible explanation for
high job reallocation rates among these plants.
T he sm all contribution of learning to total job reallocation does not preclude a large
role for learning in the cross-sectoral variation in job reallocation rates (Tables 2 and
4). For exam ple, selection effects associated w ith learning about initial conditions m ight
play a more im portant role among small establishm ents than am ong large establishm ents.
Selection effects associated w ith uncertain im itability a la Lippman and R um elt (1982)
are likely to be more im portant for single-plant than m ulti-plant firms, because technology
transference between plants w ithin a firm is relatively easy. Other things equal, selection
effects associated w ith learning about initial conditions will be more im portant in rapidly
growing sectors than in m ature or contracting sectors.
To investigate cross-sectoral differences in the im portance o f learning about initial
conditions, we im plem ented equation (3) for each two-digit industry, geographic region,
size class, and plant ownership type. Table 5 reports selected results. T he results show



20

considerable cross-sectoral variation in the fraction of job reallocation explained by learning
about initial conditions, although learning never explains more than one-fifth of sectoral
job reallocation. Learning about initial conditions is relatively im portant in the western
states, am ong sm all plants, and among plants owned by a single-unit firm. As indicated
in Tables 2 and 4, these sectors also display high total rates o f job reallocation. Thus, the
detailed results in Table 5 suggest that cross-sectoral differences in learning about initial
conditions account for much of the observed cross-sectoral differences in job reallocation
rates.
Table

6

quantifies the ability o f the passive learning story to explain cross-sectoral

differences in job reallocation rates. The first row of the table reports the cross-sectoral
standard deviation of job reallocation rates for alternative sectoral classification schemes.
The next two rows present estim ates of the fraction of the cross-sectoral variance in job
reallocation rates explained by the passive learning story. In com puting these estim ates,
we rely on equation (3) to com pute sectoral job reallocation rates net of the estim ated
contribution of learning about initial conditions.
The Table

6

results indicate that differences in the im portance o f learning about initial

conditions explain a major portion of observed cross-sectoral differences in job reallocation
rates. Learning about initial conditions explains one-third or more of the variation in job
reallocation rates among two-digit industries and census geographic reasons. Learning
explains over half of the variation in job reallocation rates among plants o f differing sizes
and between single-unit and m ulti-unit plants.
In terms of explaining job reallocation behavior, we can summarize the empirical
performance of the passive learning story as follows. Learning about initial conditions
provides a plausible explanation for the sharp and pervasive relationship between job
reallocation rates and plant age. This aspect o f our results confirms closely-related findings
by Evans (1987ab), Hall (1987) and Dunne, Roberts and Samuelson (1989a). In addition,
Table

6

indicates that the passive learning story also explains much o f the cross-sectoral

variation in job reallocation intensity. These results lead us to conclude that the passive
learning story is quite useful for interpreting variation in job reallocation intensity across
different types of plants.
On the more fundam ental matter of explaining the overall m agnitude o f job reallo­
cation, the passive learning story is far less successful. Learning about initial conditions
accounts for a sm all portion, 11-13%, of total job reallocation and only a slightly larger
fraction of excess job reallocation. This result prompts us to investigate another potential
explanation for high rates of excess job reallocation.




21

D . Q uan tifyin g the R ole o f B etw een -S ecto r E m p lo ym en t Shifts

D isturbances that cause a reshuffling of employm ent between different sectors or
groups of plants generate sim ultaneous job creation and destruction. T his sim ple point im ­
m ediately raises two questions: W hat fraction of excess job reallocation can be explained
by the reshuffling o f em ploym ent between groups o f plants defined in terms o f interesting
observable characteristics? And, which observable plant characteristics axe m ost useful
in accounting for excess job reallocation? As before, we define excess job reallocation as
total job reallocation minus the minimum level required to accom m odate net em ploym ent
expansion or contraction.
We address these questions by decomposing excess job reallocation for the m anufac­
turing sector, and for each two-digit industry, into two com ponents . 1 4 One com ponent
represents the contribution of reshuffling employm ent between sectors, and the other com­
ponent represents the contribution of excess job reallocation w ithin sectors.

Sum m ing

the two com ponents yields overall excess job reallocation. The com ponent o f excess job
reallocation due to between-sector employm ent shifts is given by

s
y ; |Net Employment Change in s\ — |Overall Net Em ploym ent Change|,
5=1

where s indexes sectors. The component due to excess job reallocation w ithin sectors is
given by

s
^ ( J o b R eallocation in s — |Net Employment Change in s |) .
5=1

Table 7 reports the results of decom posing excess job reallocation for sectoral classi­
fication schemes defined in terms of plant age, size, region, ownership type, and industry.
Our earlier discussion identified these variables as observable correlates of the factors that
underlie heterogeneity in plant-level employm ent dynamics. Each entry in Table 7 reports
the fraction of excess job reallocation explained by between-sector em ploym ent shifts for
the indicated sectoral classification.
The m ost remarkable aspect o f Table 7 is the inability of betw een-sector em ploym ent
shifts to account for excess job reallocation. According to the top panel, em ploym ent shifts
among plants of different ages, sizes, regions, ownership types, and tw o-digit industries
account for virtually none of the excess job reallocation in the m anufacturing sector as a
14

D unne, Roberts and Samuelson (1989b) exploit an equivalent decom position in their

analysis of gross job flows over five-year intervals.



22

whole. C u ttin g sectors m uch m ore finely by defining th em in te rm s of age, size, region,
a n d ow nership sim ultaneously, betw een-sector em ploym ent shifts account for only 15%
o f excess jo b reallocation.

E m ploym ent shifts am ong th e 450 four-digit m a n u factu rin g

in d u strie s account for a m ere 12% of excess jo b reallocation. E ven w hen we define sectors
in term s of all five p la n t ch aracteristics sim ultaneously, betw een-sector em ploym ent shifts
account for only 39% of excess jo b reallocation. To a p p reciate th e level of d etail c a p tu re d
by th is classification schem e, we re m a rk th a t th e average n o n em p ty “secto r” co n tain s only
a b o u t five sam pled p la n ts.
T h e industry-level decom positions in T able 7 caxry th e sam e basic m essage, a lth o u g h
som e ad d itio n a l p a tte rn s em erge. M ost notably, for every tw o-digit in d u stry th e age-based
sectoral classification schem e yields a m ore successful acco u n tin g o f excess jo b reallo catio n
th a n any o th e r classification schem e based on a single p la n t ch aracteristic. E m ploym ent
shifts betw een age groups account for o n e-ten th o r m ore of excess jo b reallo catio n in m ost
in d u strie s an d roughly one-fifth of excess jo b reallo catio n in a h an d fu l of in d u stries. T h e
age resu lts suggest th a t theories stressing em bodied technical change o r o th e r sources
of v intage effects axe likely to provide a p a rtia l ex p lan atio n for hig h ra te s of excess jo b
reallocation.
T h e re su lts in T able 7 argue stro n g ly ag ain st th e view th a t hig h ra te s o f excess jo b
reallocation arise p rim arily because of sectoral d istu rb an ces o r econom yw ide d istu rb a n ce s
w ith differential sectoral effects. In stead , T able 7 argues th a t excess jo b reallo catio n is
fu n d am en tally a phenom enon re la ted to plant-level heterogeneity in la b o r d e m an d b eh av ­
ior. L earning a b o u t in itia l conditions is one reason for plant-level hetero g en eity in la b o r
d em an d , b u t we found th a t th is sto ry has lim ited ability to explain th e m a g n itu d e of
jo b reallocation. T heories th a t stress active learning an d selection am ong young a n d old
p la n ts (E ricson an d Pakes, 1989), theories th a t stress endogenous p reco m m itm en t to h e t­
erogenous p ro d u c tio n technologies (L am bson, 1990), an d theories th a t stress exogenous
plant-specific cost or d em an d d istu rb an ces (D avis an d H altiw anger, 1990) all seem consis­
te n t w ith th e resu lts in T able 7. F u rth e r investigation in to th e ab ility of these theories to
explain high ra te s of excess jo b reallo catio n m u st aw ait fu tu re research.

V . A c c o u n tin g for T im e V a ria tio n in J o b R e a llo c a tio n In te n s ity
T able 1 show ed th a t th e pace o f jo b reallo catio n ex h ib its significant countercyclic
v ariatio n in o u r sam ple. For exam ple, betw een th e business cycle tro u g h in 1975 a n d
th e p eak in 1980 th e jo b reallocation ra te fell by six p ercen tag e p o in ts.

T h is cyclical

p a tte rn is confirm ed in subsequent research th a t relies on d a ta for o th e r tim e p eriods,
sectors, a n d countries. B lan ch ard an d D iam ond (1990) d e m o n stra te a close relatio n sh ip



23

betw een o u r jo b cre atio n a n d d e stru c tio n figures a n d a p p ro p ria te ly a d ju ste d m easu res o f
jo b tu rn o v e r in th e BLS m an u fa ctu rin g tu rn o v e r series. T h ey find th a t th e ir re la te d jo b
reallo catio n m e asu re flu c tu a te s c o u n tercy c lic a l^ over th e 1958 to 1981 p erio d . B ased o n
BLS estab lish m en t-lev el d a ta , B ronaxs (1990) finds significant countercylical v a ria tio n in
th e jo b reallo ca tio n ra te for every one-digit in d u stry g ro u p in th e U .S. over th e 19721989 p erio d . T a b u la tio n s in B aldw in an d G orecki (1990, T ab le 3.5) reveal countercyclic
jo b re a llo catio n b eh av io r in th e C an a d ia n m an u factu rin g secto r d u rin g th e 1970 to 1981
p erio d . R egev (1990) re p o rts countercyclical v a riatio n in jo b re allo catio n ra te s for Israel
d u rin g th e 1980’s.
T hese em pirical resu lts p o in t to a close relatio n sh ip b etw een th e b u sin ess cycle a n d
th e in te n sity of jo b reallocation , b u t th ey do n o t ad d ress th e q u estio n of w hy th e jo b
reallo catio n ra te flu c tu a te s countercyclically. In view o f th e links b etw een jo b re a llo ca tio n
a n d w orker reallo catio n , a n answ er to th is q u estio n will pro v id e in sig h t in to th e source a n d
n a tu re of ag g reg ate la b o r m a rk e t flu ctu atio n s. To ad d ress th e q u estio n of w hy jo b reallo ca­
tio n m oves countercyclically, we first ad d ress tw o sim pler questions: How m u ch o f th e tim e
v ariatio n in jo b reallo catio n is acco u n ted for by m ean tra n sla tio n s o f th e estab lish m en t-lev el
grow th ra te d en sity a n d differential m ean secto ral responses to ag g reg ate d istu rb a n ce s?
A nd, how does th e cyclical b eh av io r of jo b reallo catio n differ by in d u s try ty p e , p la n t size,
age a n d ow nership ty p e? D raw ing on o u r answ ers to th ese q uestions, we th e n d iscrim in a te
betw een m acroeconom ic theories th a t can n o t explain th e observed cyclical b eh av io r o f jo b
reallo catio n a n d theories th a t p o te n tia lly can.
A. A n A ccounting Fram ew ork
C onsider th e lin ear m odel for establishm ent-level em ploym ent g ro w th ra te s
9et = get + 9 s t + 9 t ,

(4)

w here gt is th e m an u fa ctu rin g g ro w th ra te , g at is th e sector g ro w th ra te (d e v ia ted a b o u t
th e m a n u fa c tu rin g grow th ra te ), a n d g f f js th e resid u al id io sy n cratic co m p o n en t o f th e
e stab lish m en t g ro w th ra te . A ccording to e q u atio n (4), each e sta b lish m e n t’s g ro w th ra te
a t t is th e su m of a n ag gregate-tim e effect, a secto r-tim e effect a n d a tim e-v ary in g idiosyn­
cra tic effect. T im e v ariatio n in th e realized aggregate a n d secto ral g ro w th ra te s in d u ce
tim e v a ria tio n in th e location a n d sh ap e of th e d ensity over th e (size-w eighted) get, th e re b y
g en eratin g tim e v ariatio n in gross jo b creatio n , d e stru ctio n a n d reallo catio n . T h e crosssectional variance a n d higher m o m en ts of th e id io sy n cratic co m p o n en t, g f f , also influence
th e sh ap e of th e g ro w th ra te density, th ereb y g en eratin g fu rth e r tim e v a ria tio n in th e jo b
flow m easures.



24

Several a lte rn a tiv e views a b o u t th e n a tu re of aggregate flu c tu a tio n s can b e couched
in term s of eq u atio n s like (4). Prevailing views o f th e business cycle stress th e role o f
aggregate d istu rb an ces as driving forces. T h e sim plest version of th is view im plies th a t all
tim e v ariatio n in gross jo b creation, d e stru ctio n a n d reallo catio n reflects by tim e v ariatio n
in th e ag gregate-tim e effects. T h is view encom passes a tim e-in v arian t, b u t possibly large,
cross-sectional variance of th e idiosyncratic com ponent o f th e gtf W e rep resen t th is p u re
aggregate shifts sto ry by th e hypothesis th a t th e d istrib u tio n over th e gjt = get — gt is
tim e in v arian t.
A less sim plistic c h aracterizatio n of prevailing views a b o u t th e business cycle w ould
in c o rp o ra te differences in th e tim in g a n d m ag n itu d e o f secto ral responses to ag g reg ate dis­
tu rb a n c es. S ystem atic cross-sectoral differences in th e responses to ag g reg ate d istu rb an ces
are a n im p o rta n t elem ent of tra d itio n a l views a b o u t th e business cycle. See A b ra h a m an d
K atz (1986) on th is p oint.
To c a p tu re th is aspect of tra d itio n a l views, we allow for com pletely u n re stric te d sec­
to ra l responses to aggregate d istu rb an ces.
tim e-invariant d istrib u tio n over th e

In p a rtic u la r, consider th e h y p o th esis of a

In view o f (4), th e secto r-tim e effects gat c a p tu re

an y sy stem atic o r no n -sy stem atic cross-sectoral differences in th e m ean resp o n se to ag­

g regate d istu rb an ces. N either linearity, m ag n itu d e, n o r tim in g re stric tio n s a re placed on
th e m ean sectoral responses to aggregate d istu rb an ces u n d e r th is in te rp re ta tio n o f th e g9t
.
T h e only re strictio n s placed on m ean sectoral responses are th o se in h eren t in th e secto ral
classification schem e itself.
B ased on th e decom position in (4), we m easure th e relativ e im p o rtan c e of aggre­
g ate, sectoral a n d idiosyncratic com ponents for tim e v ariatio n in jo b creatio n , d e stru c tio n
a n d reallocation. We also m easure th e covariation betw een th e com ponents. To see o u r
p ro ced u re, consider th e d istrib u tio n over th e g f f , from w hich we co m p u te jo b creatio n ,
d e stru ctio n a n d reallocation ra te s a d ju sted for th e ag g reg ate-tim e a n d th e secto r-tim e
effects:

fosf=
N E G f=

£

£

!% ? ■ ),

^ (Is ifl),

e S t<0
,«

S U Mf

(5)

and,

(6)

*

= £

^ i | j 5 r |.
« x ‘

(7)

T im e v ariatio n in these a d ju sted m easures reflects only th e co n trib u tio n s o f th e id- •ST
-

iosyncratic effects. T h u s,



S U M t m easures
25

th e gross ra te o f change in th e n u m b e r of

establishm ent-level em ploym ent positions as a result of idiosyncratic establishm ent-level
------ S

T

em ploym ent m ovem ents. From a statistical perspective, S U M t

equals the size-w eighted

average absolute deviation of establishm ent growth rates around the overall and sectoral
means.
Now consider the identity

S U M t = S U M * 7 + ( S U M t - S U M * 7 ),

(8 )

which im plies the variance decom position for gross job reallocation,
—

—_

S T

V a x (S U M t) = V a i( S U M t

S T

)+ V a x (S U M t - S U M t

— _

—_ -

S T

)+ 2 C o v ( S U M t

S T

,S U M t - S U M t

).

(9)
_ . 5j»
_

If the distribution over the g

is time-invariant, then the ratio of Vai ( S U M t

) to

Var( S U M t) equals zero. Conversely, a large value for this ratio indicates that tim e varia­
tion in the cross-sectional variance (and higher m om ents) of g f f accounts for m uch o f the
tim e variation in gross job reallocation. We interpret the covariance term as reflecting the
part of tim e variation in gross job reallocation that cannot be unam biguously assigned to
either the aggregate and sectoral effects or to the idiosyncratic effects.
We also decom pose the variance of job creation and destruction rates along the lines
of ( 8 ) and (9). Variance ratios provide information on the relative contribution of aggregate/sectoral versus idiosyncratic effects to tim e variation in job creation and destruction.
T he covariance terms indicate whether the idiosyncratic effects reinforce or counteract the
im pact of aggregate and sectoral effects on job creation and destruction rates.
B. R esults
Table

8

decom poses the tim e-series variance of annual job reallocation, creation and

destruction rates using several sectoral classification schemes. According to the first row
of the first panel, aggregate and sectoral effects account for 4.2-10.5% of the tim e variation
in job reallocation, depending on the classification scheme. A ssigning all of the covariance
term to the aggregate and sectoral effects, they still account for no more than

2 0

% percent

o f tim e variation in annual job reallocation rates. These results show that tim e variation in
the structure o f mean employm ent growth rates across regions, detailed industries, plant
size classes, age groups, and ownership types account for remarkably little o f the tim e
variation in job reallocation. The flip side of the sam e coin is that idiosyncratic effects
account for 80% or more of the variability in annual job reallocation rates.



26

Thus, Table 8 finds that only 4-20% of the time variation in job reallocation rates
can be accounted for by mean translations of the growth rate density and differential
mean sectoral responses to aggregate disturbances. T his finding refutes the hypothesis
that some system atic pattern of sectoral responses to aggregate disturbances can account
for the significant tim e variation in gross job reallocation displayed in Table 1. Instead,
the tim e variation in gross job reallocation results overwhelmingly from tim e variation
in the m agnitude of idiosyncratic effects. This result is especially striking in that our
narrow definition o f idiosyncratic effects imposes neither linearity, m agnitude nor tim ing
restrictions on the mean sectoral responses to aggregate disturbances.
The second and third panels of Table 8 shed further light on the time-series behavior
of gross job reallocation. These panels indicate that aggregate-year effects play a dom inant
role in accounting for tim e variation in job creation and destruction rates. The variance of
the idiosyncratic component of job creation amounts to only 12-16% of the overall variance
of job creation, and the variance of the idiosyncratic component of job destruction amounts
to only 6-8% o f the overall variance of job destruction. The covariance results for job
creation and destruction link their behavior to the behavior o f job reallocation. For job
destruction, the positive sign and large magnitude of the covariance terms indicate that
idiosyncratic effects strongly reinforce the countercyclic m ovements in gross job destruction
associated w ith aggregate mean effects. For job creation, in contrast, the negative sign
and large m agnitude of the covariance terms indicate that idiosyncratic effects strongly
counteract the procyclic fluctuations in job creation associated w ith aggregate mean effects.
Taken together, the covariance terms from the P O S and N E G decom positions explain
how the idiosyncratic component dom inates fluctuations in job reallocation. W hile P O S
falls and N E G rises during economic contractions, idiosyncratic effects counteract the fall
in gross job creation while reinforcing the rise in gross job destruction.
We turn now to a more detailed accounting for tim e variation in job reallocation in­
tensity. Table 9 provides information on the cyclical behavior of sectoral job reallocation
rates. The top panel of the table captures two points. First, whether we define sectors
in terms of industry, region, size, age or ownership type, m ovem ents in both raw and
adjusted sectoral job reallocation rates are predominantly countercyclical. For exam ple,
all twenty of the two-digit manufacturing industries show countercyclic m ovem ents in raw
and adjusted job reallocation rates. Second, under each sectoral classification schem e the
adjusted job reallocation rate shows a stronger, and typically more pervasive, pattern of
countercyclical m ovements than the raw rate. Focusing on the two-digit industry break­
down again, the m ean correlation between the net industry job growth rate and the raw
own-industry job reallocation rate equals -.51. Adjusting the empirical growth rate density
for aggregate and sectoral effects, and computing the adjusted job reallocation rates, yields



27

a mean correlation of -.55. Thus, rather than providing an explanation for countercyclical
fluctuations in job reallocation, sectoral differences in mean growth rates actually m itigate
the countercyclicality of job reallocation.
The bottom panel of Table 9 shows how the cyclical behavior of job reallocation varies
across sectors. Countercyclic movements in job reallocation rates axe more pronounced for
larger plants, older plants, m ulti-unit plants, and plants th at produce durable goods.
The results by plant age and size axe especially striking. Segregating plants into
groups of young (0-9 years) and old (10+ yeaxs), and then interacting with two-digit
industry, yields forty industry-by-age sectors. For the twenty sectors representing older
plants, the size-weighted average correlation between rates of net sectoral growth and
adjusted gross job reallocation equals -.71. In shaxp contrast, the younger plant sectors
show no systematic relation between net job growth and gross job reallocation. These
results reveal th a t the countercyclicality of job reallocation rates entirely reflects greater
heterogeneity in the establishment-level employment movements of m ature plants during
contractions. A similar characterization of cyclical movements in job reallocation rates
holds in term s of small versus large plants. Cross-classifying on two-digit industry and our
five size classes yields 100 industry-by-size sectors. The average correlation between net
sectoral growth and adjusted job reallocation for the forty large plant sectors is -.63. In
contrast, the average correlation for the forty small plant sectors is only -.20.
It is helpful to place the results in the bottom panel of Table 9 alongside the vari­
ance decomposition results in Table 8. The variance decomposition results show th a t the
great bulk of time variation in gross job reallocation cannot be accounted for by sectoral
differences in mean responses to cyclical impulses. The bottom panel of Table 9 indicates
th at the bulk of tim e variation in job reallocation can be accounted for by especially sharp
countercyclical job reallocation movements among sectors made up of older, larger and
m ulti-unit plants.
While the results in Table 9 provide insight into the basic p attern of time variation in
sectoral job reallocation rates, they provide little information about the m agnitude of the
covariances between net overall and sectoral growth rates, on the one hand, and sectoral job
reallocation rates, on the other hand. To investigate the covariance structure, we regress
the adjusted sectoral reallocation rates defined by (7) on net sectoral and m anufacturing
growth rates plus interactions of these net rates with age, size and ownership dummies. The
regressions also contain sectoral fixed effects to control for perm anent sectoral differences
in the intensity of job reallocation.
Table 10 summarizes the regressions and reports key results. Column (1) of the top
panel, for example, regresses adjusted industry-level job reallocation rates on industry
fixed effects and two time-varying covariates: gt and gst. These covariates are highly



28

significant (t-statistics greater than five in absolute value), and they account for 27%
of the time variation in industry job reallocation rates. The bottom panel summarizes
the implications for the covariance structure. Here, we use the regression to estimate
the response of adjusted job reallocation rates to one standard deviation increases in gt
and gat. Based on regression (1), for example, a one standard deviation decline in the
manufacturing (own-industry) net growth rate is associated with an increase in sectoral
job reallocation rates of 1.15 (.24) percentage points. Relative to regression (1), regressions
(2)-(4) add the age, size, and ownership interaction terms, respectively.
Two main results stand out in Table 10.15 First, large movements in sectoral job reallocation rates axe associated with movements in total m anufacturing employment growth
rather than movements in own-sector employment growth. This result occurs primarily
because the average time-series standard deviation of gst is small relative to the standard
deviation of gt. The regression coefficients on gt and gat differ significantly only for old
plants in regression (2).
Second, the covariation between the manufacturing employment growth rate and sec­
toral job reallocation rates is much larger among old plants than among young plants,
among medium-sized and big plants than among small plants, and among m ulti-unit plants
than among single-unit plants. Indeed, there is no evidence of statistically significant co­
variation between manufacturing or own-sector net employment growth and rates of job
reallocation among younger, smaller, and single-unit plants. There is clear evidence of
large and highly significant covariation between manufacturing employment growth and
rates of job reallocation among older, larger and multi-unit plants.
A similar, but less pronounced, pattern emerges with respect to the covariation be­
tween own-sector employment growth and sectoral job reallocation rates. Point estimates
indicate greater negative covariation between own-sector employment growth and job re­
allocation rates among older, larger and multi-unit plants. These differences are statisti­
cally significant at the five percent level except for the comparison between m ulti-unit and
single-unit plants. The negative covariation between own-sector employment growth and
job reallocation rates is highly statistically significant for old and large plants.
C. Interpretation of Cyclical Findings
We have established the following cyclical facts: (1) Job reallocation rates fluctuate
countercyclically; this pattern is pervasive across industries and regions. (2) The countercyclic behavior of job reallocation reflects time variation in the magnitude of idiosyncratic
15The main results in Table 10 are unaffected if we use the raw job reallocation rates as
dependent variables in the regressions.



29

plant-level employment movements, not sectoral differences in the mean employment re­
sponses to aggregate disturbances. (3) Job reallocation rates among young (0-9 years),
small (1-249 employees), and single-unit plants exhibit little or no systematic relationship
to the cycle. (4) Job reallocation rates among older, larger and m ulti-unit plants exhibit
pronounced countercyclic patterns of variation.
W hat classes of macroeconomic models can explain these facts? It is useful, and
perhaps easier, to first identify im portant classes of models th at cannot explain these facts:
(i) Models th at specify or treat all firms as homogenous, (ii) Sectoral models of the business
cycle th at specify homogenous firms within sectors. Examples include simple versions
of the model described by Lilien (1982), in which sectoral disturbances drive aggregate
fluctuations, and the model described by Abraham and Katz (1986), in which aggregate
disturbances drive differential sectoral responses, (iii) Sectoral or aggregate models th at
treat the idiosyncratic component of firm-level employment behavior as orthogonal to the
business cycle. This class includes models that specify a cyclically invariant natural rate
of unemployment as in Phelps et al (1970), Hall (1979), and Johnson and Layard (1986).
We stress th at appending idiosyncratic establishment-level shocks to simple sec­
toral or aggregate models is not sufficient to explain our cyclical findings. Idiosyncratic
establishment-level shocks clearly generate an underlying rate of gross job reallocation
within sectors, but they do not necessarily generate a relationship between aggregate fluc­
tuations and the pace of job reallocation. This point is nicely made by Caballero (1990).
He posits an asymmetry in firm-level hiring and firing costs in a model th at accommodates
aggregate and idiosyncratic labor demand disturbances. His adjustm ent cost specification
implies a higher time-series variance in job destruction rates than in job creation rates
at the firm level. This feature of the microeconomic structure in Caballero’s model is
consistent with the pattern displayed in our Figure 2. If this firm-level result carried
over to the aggregate level, it would provide an explanation for countercyclic variation in
job reallocation rates.16 However, Caballero shows th at the asymmetry in firm-level job
creation and destruction behavior is smoothed away by aggregation when firms exhibit id­
iosyncratic components to their employment movements. Empirically, we have seen th at
the idiosyncratic components are large and pervasive.
To explain our findings requires a macroeconomic model th at generates simultaneous
job creation and destruction within narrowly defined sectors and countercyclical rates of
job reallocation within sectors. Progress along these lines is made in recent work by Blan­
chard and Diamond (1989, 1990), Davis and Haltiwanger (1990), and Caballero (1990).
16The raw job reallocation rate, S U M t, is negatively correlated with N E T t if, and only
if, V ax(N E G t) exceeds Var(P O S t ).



30

These authors specify alternative models that allow both common aggregate and idiosyn­
cratic allocative shocks to influence establishment-level employment dynamics. The mod­
els differ in the frictions th at they ascribe to the process of reallocating workers and jobs
across establishments, but in each model labor market frictions imply potentially impor­
tant interactions between aggregate employment growth and the pace of reallocation.
These models identify four types of potentially im portant interactions between the
pace of job reallocation and the stage of the business cycle. First, time-series fluctuations
in the intensity of allocative shocks can cause aggregate employment fluctuations, as well
as countercyclic movements in the job reallocation rate. Second, aggregate shocks can
influence the timing of the job reallocation that ultimately arises from allocative shocks,
and thereby lead to a bunching of job reallocation activity during downturns.17 Third,
aggregate downturns may induce a shake-out of less efficient firms and establishments,
leading to both aggregate contraction and increased heterogeneity in plant-level employ­
ment movements. Fourth, if negative aggregate shocks are more severe (and less frequent)
than positive aggregate shocks, then the endogenous evolution of the cross section dis­
tribution over plant-level employment growth can generate countercyclic variation in job
reallocation intensity.
In light of the findings reported in this paper, disentangling these and other connec­
tions between aggregate activity and the pace of job reallocation is an im portant area
for future research. None of the interpretations of countercyclic job reallocation intensity
offered by Blanchard-Diamond, Davis-Haltiwanger, and Caballero incorporate an expla­
nation for the findings in this paper related to pronounced differences in the magnitude
and cyclicality of job reallocation intensity by plant age, size and ownership type.
V . C o n clu sio n
This study paints a sharp picture of gross job flow behavior in U.S. m anufacturing
industries. Gross rates of job creation and destruction are remarkably large - they amount
to roughly ten percent of manufacturing employment in a typical year. The phenomenon
of simultaneously high rates of job creation and destruction is pervasive across industries
and across groups of plants defined in terms of plant age, size, region and ownership type.
In large part, the gross job flows that we measure reflect establishment-level employment
changes th at are highly persistent and concentrated at plants experiencing sharp expansion
or contraction.
17Darby, Haltiwanger, and Plant (1985) and Davis (1987) also discuss this reallocation
timing effect.



31

The m agnitude and character of gross job flows bear directly on the reasons for gross
worker flows in the labor market. Combining longitudinal information from household
and establishment surveys, we calculate th at the reallocation of employment opportunities
across establishm ents accounts for 35-56% of all worker reallocation between employers or
between employment and joblessness.
The m agnitude and cyclical variability of gross job flows differs systematically across
plants w ith different observable characteristics. On average, job reallocation rates are
substantially higher among younger, smaller and single-unit plants. At the same time, job
reallocation rates among these plants show no systematic cyclical variation, whereas job
reallocation rates among older, larger and m ulti-unit plants show pronounced countercyclic
variation.
This paper provides partial explanations for several aspects of gross job flow behavior.
Further research designed to explain gross job flow behavior and to develop its implications
for labor m arket dynamics, for the evolution of firms and industries, and for the nature of
business cycles m erits a high priority.




32

References
Abowd, John and Arnold Zellner (1985) “Estim ating Gross Labor Force Flows,” Journal
of Economic and Business Statistics, 3, 254-283.
Abraham, K. and L. Katz (1986) “Cyclical Unemployment: Sectoral Shifts or Aggregate
Disturbances?” Journal of Political Economy, 94, 507-522.
Akerlof, George, Andrew Rose, and Janet Yellen (1988) “Job Switching and Job Satis­
faction in the U.S. Labor M arket,” Brookings Payers on Economic Activity, 1988:2,
495-582.
Baldwin, John R. and Paul K. Gorecki (1990) Structural Change and the Adjustment
Process: Perspectives on Firm Growth and Worker Turnover, Canadian Government
Publishing Centre, Ottawa.
Blanchard, Olivier and Peter Diamond (1989) “The Beveridge Curve” Brookings Papers
on Economic Activity 1989:1, 1-60.
Blanchard, Olivier and Peter Diamond (1990) “The Cyclical Behavior of Gross Flows of
Workers in the U.S.,” Brookings Papers on Economic Activity 1990: 2, 85-155.
Bresnahan, T im othy'and Daniel Raff (1990) “Technological Heterogeneity, Adjustment
Costs, and the Dynamics of Plant Shut-Down Behavior: The American Motor Vehicle
Industry in the Time of the Great Depression,” working paper, Stanford University.
Bronars, Stephen G. (1990) “Employment and Hours Variation Over the Business Cycle,”
paper in progress, Bureau of Labor Statistics.
Caballero, Ricardo (1990) “A Fallacy of Composition” Discussion Paper Series No. 479,
Columbia University.
Clark, Kim B. and Lawrence H. Summers (1979) “Labor Market Dynamics and Unemploy­
ment: A Reconsideration,” Brookings Papers on Economic Activity, 1979:2, 13-60.
Darby, Michael, John Haltiwanger, and Mark Plant (1985) “Unemployment Rate Dynam­
ics and Persistent Unemployment under Rational Expectations,” American Economic
Review, 75, 614-637.
Davis, Steve J. (1987) “Fluctuations in the Pace of Labor Reallocation,” CamegieRochester Conference Series on Public Policy, 27, 335-402.
Davis, Steve J. and John Haltiwanger (1989) “Gross Job Creation, Gross Job Destruction,
and Employment Reallocation,” Hoover Institution Working Papers in Economics,
E-89-33.
Davis, Steve J. and John Haltiwanger (1990) “Gross Job Creation and Destruction: Mi­
croeconomic Evidence and Macroeconomic Implications,” NBER Macroeconomics
Annual, 5, 123-168.



33

Davis, Steve J., John Haltiwanger, and Scott Schuh (1990) “Published Versus Sample
Statistics from the ASM: Implications for the LRD” forthcoming in Proceedings of
the American Statistical Association, Business and Economics Statistics Section.
Dunne, Tim othy and Mark J. Roberts (1989) “The Duration of Employment O pportunities
in U.S. M anufacturing,” working paper, Pennsylvania St. University.
Dunne, Timothy, Mark J. Roberts, and Larry Samuelson (1989a) “The Growth and Failure
of U.S. M anufacturing Plants,” Quarterly Journal of Economics, 104, 671-698.
Dunne, Timothy, Mark J. Roberts, and Larry Samuelson (1989b) “Plant Turnover and
Gross Employment Flows in the U.S. Manufacturing Sector,” Journal of Labor Eco­
nomics, 7, no. 1, 48-71.
Ericson, Richard and Ariel Pakes (1989) “An Alternative Theory of Firm and Industry
Dynamics,” working paper, Yale University.
Evans, David S. (1987a) “Tests of Alternative Theories of Firm Growth,” Journal of
Political Economy, 95, no. 4, 657-674.
Evans, David S. (1987b) “The Relationship between Firm Growth, Size and Age: E sti­
mates for 100 M anufacturing Industries,” Journal of Industrial Economics, 15, 567581.
Hall, Bronwyn H. (1987) “The Relationship between Firm Size and Firm Growth in the
U.S. M anufacturing Sector,” Journal of Industrial Economics, 15, 583-606.
Hall, Robert E. (1979) “A Theory of the Natural Unemployment Rate and the D uration
of Employment,” Journal of Monetary Economics, 5, 153-169.
Hall, Robert E. (1982) “The Im portance of Lifetime Jobs in the U.S. Economy,” American
Economic Review, 72, 716-724.
Johnson, G.E. and P.R.G. Layard (1986) “The N atural Rate of Unemployment: Explana­
tion and Policy,” in O. Ashenfelter and Layard, eds., Handbook of Labor Economics,
Volume 2, Amsterdam: North-Holland.
Jovanovic, Boyan (1982) “Selection and the Evolution of Industry,” Econometrica, 50, 3,
649-670.
Lambson, Val E. (1989) “Industry Evolution with Sunk Costs and Uncertain Market Con­
ditions,” forthcoming in the International Journal of Industrial Organization.
Leonard, Jonathan (1987) “In the Wrong Place at the Wrong Time: The Extent of Fric­
tional and Structural Unemployment,” in Unemployment & the Structure of Labor
Markets, edited by Kevin Lang and J. Leonard, Basil Blackwell: New York.
Lilien, David (1980) “The Cyclical P attern of Temporary Layoffs in United States Manu­
facturing,” Review of Economics and Statistics, 112, no. 1, 24-31.



34

Lilien, David (1982) “Sectoral Shifts and Cyclical Unemployment,” Journal of Political
Economy, 90, 777-793.
Lippman, S.A. and R.P. Rumelt (1982) “Uncertain Imitability: An Analysis of Interfirm
Differences Under Competition,” Bell Journal of Economics, IS, no. 2, 418-438.
Pakes, Ariel and Richard Ericson (1990) “Empirical Implication of Alternative Models of
Firm Dynamics,” working paper, Yale University.
Phelps, Edmund S., et al. (1970) Microeconomic Foundations of Employment and Inflation
Theory, New York: Norton.
Poterba, James and Lawrence Summers (1989) “Reporting Errors and Labor Market Dy­
namics,” Econometrica, 54, 1319-1338.
Regev, Haim (1990), unpublished tabulations from Central Statistical Bureau, Israel.




35

Figure IA
Unwaithtod Growth Rato Distribution
Percent

0

0

0

2

0

2

Growth Rate Intervals (Width

Figure tB
Slze~Weightsd Growth Rato Distribution
Percent




zt
xr

e
a
t
h

•1
)

^<xxxxxxxxxxxxxxxxxxxxg

OQ -H M JJ J3
E

K
t

Figure 2.
Job Creation and Destruction Partitioned by Establishment Growth Rate

g

in in

K

E

o in

E

EE

in

I o

cn

I o

OOOOOOOOOOOCKX
<x<>o<^oo<x

i o

o in
o in

in in

o

SEE
BE
E

in

o in

E

IE

E
E

in in

g
Q

in

o

CN

CN




in
a. a) u

in
o

a a
>

jj

a) <d

A

I n t e r v a ls

cn

R ate

7
>oooo<^ooo<
oo<x>oooooooo^oo<
>^>0<>000<>^<X><x>0<X><X
XXx*XxX>00000<X
<xxxxxxxxxx

in in

G row th

KXEXXX

(W id th

e s s

Table 1. Net and Gross Rates by Year, M anufacturing Sector
Year
1973
1975
1976
1977
1978
1980
1981
1982
1983
1985
1986

PO St
0.132
0.067
0.113
0.112
0.116
0.080
0.070
0.064
0.086
0.084
0.088

NETt 1
0.071
-0.100
0.017
0.018
0.041
-0.012
-0.049
-0.087
-0.056
-0.033
-0.044

NEG
0.061
0.166
0.096
0.096
0.075
0.093
0.118
0.152
0.142
0.117
0.132

Pearson Correlations:2
p (P O S t ,N E G t ) = -0.864

(.001)

SU M t
0.194
0.233
0.209
0.206
0.191
0.173
0.188
0.216
0.227
0.201
0.220

M AX,
0.133
0.166
0.122
0.117
0.117
0.102
0.119
0.152
0.143
0.121
0.133

p (N E T t , S U M t ) = -.5 6 5

(.07)

Notes:
1 N E T t = P O S t — N E G t is the net employment growth rate.
2 Marginal significance level in parentheses.

Table 3. Persistence Rates for Job Creation and Destruction1
Year2 (t)
FPO Sn
F N E G ti
FPO Sn
.72
.54
.73
1975
.79
.75
.58
1976
.79
.76
1977
.82
.43
.63
1980
.88
.44
.60
1981
.86
.60
1982
.84
.63
1985
.81
Simple Mean
.67
.50
Notes:
1 F P O Stn (F N E G tn ) is the fraction of jobs created (destroyed)
year t — 1 and March of year t that persists through March of year t +
-

FNEGn
.62
.69
—

.77
.82

-

-

—

—

.73
between March of
n.

2
Given the ASM panel structure, the persistence measures can be calculated for all
plants only in the indicated years.




TABLE 2
Net and Gross Rates by Industry
Size-Weighted Averages:
^

P0S

NEC

NET

SIM

MAX

slm -| et |
n

0.089
0.058
0.074
0.116
0.129

0.104
0.082

0.193
0.140
0.185
0.272
0.288

0 . 1 2 1

0.063
0.091
0.068
0.066
0.107
0.091

0.078
0.087
0.080
0.091
0.118
0.144

0.108
0.090
0.124
0.168
0.188
0.143
0.089
0.099
0.089

0.169
0.098

0 . 1 0 1

-0.015
-0.024
-0.036
-0.040
-0.031
-0.019
-0.015
-0.004
-0.013
-0.025
-0.011
-0.053

0.093
0.059

0.123
0.114

-0.031
-0.054

0.095

0 .1 2 0

0.096

Industry:
Food
(2 0 )
Tobacco
(2 1 )
Textile
(2)
2
Apparel
(23)
linker
(24)
Furniture
(25)
Paper
(26)
Printing
(27)
Chemicals
(28)
Petroleum
(29)
Rubber
(30)
Leather
(31)
Stone, Clay
and Glass
(32)
Primary Metals (33)
Fabricated
Metals
(34)
Nonelectric
Machinery
(35)
Electric
Machinery
(36)
Transportation (37)
Instruments
(38)
Miscellaneous (39)
Total
Manufacturing
Size-weighted
Cross-Industry
Standard
Deviation

Cross-Industry: 2

0 .1 1 0

0.156
0.160

0 . 1 2 1

0.207
0 .2 0 2

0.143
0.152

0.157
0.105
0.158
0.118
0.114
0.163
0.166

0.216
0.173

0.136
0.126

0.160
0.094

-0.025

0.215

0.137

0.156

0 . 1 2 1

-0.025

0.217

0.141

0.152

0.097
0.094
0.093
0.108

0.109
0.099
0.093
0.145

-0.011
-0.006
- .0 0 2
0
-0.037

0.206
0.193
0.186
0.253

0.130
0.123
0.156

0.152
0.140
0.149
0.193

0.092

0.113

0

.0 2 1

0.205

0.129

0.152

0.016

0 .2 1

0.015

0.034

0.023

0.028

p(FOS.NBG) - 0.764
(0.0001)

0 .2 2 2

0.141
0.178
0.148
0.157
0.225
0.235

0 .1 0 0

0 . 1 1 2

p(NET,SIM) - -0.347
(0.135)

^Size-wei^ited average based on annual values with t - 1973-1986 (excluding 1974, 1979,
1984).
^Marginal significance levels in parentheses.




Table 4. Net and Gross Rates by Type of P lan t1
No. of Employees
1-99
100-249
250-499
500-999
1000-1Age in Years
Births
1
2
3
4-5
6-10
11-14
15+

NEG
.164
.120
.105
.093
.078

Size2
NET
-.023
-.021
-.019
-.023
-.019

SU M
.304
.219
.191
.163
.138

M AX
.180
.133
.120
.106
.090

Share3
.246
.185
.162
.134
.273

POS

NEG

A ge
NET

SU M

M AX

.270
.169
.139
.133
.120
.102
.065

.206
.167
.117
.134
.121
.111
.097

.476
.336
.257
.267
.240
.213
.162

.299
.200
.148
.154
.135
.123
.103

Share
.008
.018
.015
.015
.045
.143
.110
.645

POS
.140
.099
.086
.070
.060

.064
.003
.022
-.001
-.001
-.010
-.033

O w n e rsh ip T y p e
SU M
NEG
NET
.184
.103
-.023
-.016
.277
.146

M AX
.115
.170

Share
.768
.232

Census Region
New England
Middle Atlantic
South Atlantic
E. South Central
W. South Central
E. North Central
W. North Central
Mountain
Pacific

G e o g ra p h ic R e g io n
NEG
NET
SU M
-.018
.198
.108
.121
-.036
.205
.111
-.032
.190
.107
-.015
.198
.101
-.009
.193
.107
.198
-.016
.115
-.010
.220
.114
.232
.005
.128
.246

M AX
.122
.127
.126
.122
.117
.124
.134
.138
.149

Share
.073
.175
.238
.068
.154
.070
.077
.026
.120

POS
.090
.085
.079
.092
.092
.091
.105
.118
.118

i
O
oO
C

Firm Operates:
POS
Multiple Mfg. Plants .080
A Single Mfg. Plant .131

Notes:
1 Figures are size-weighted averages of eleven annual values, except for age group figures.
Age group figures are size-weighted averages of 1978 and 1983 values.
2 A plant’s size is measured as its mean number of employees over all sample observations
with positive employment.
2 Group share of total employment, using the size metric described in the text.




Table 5. Estim ated Fraction of Job Reallocation Due to
Learning about Initial Conditions
T o ta l M a n u fa c tu rin g
S elected I n d u s tr ie s
Food
Tobacco
Printing
Prim ary Metals
Fabricated Metals
Transportation
S e lec te d R e g io n s
Middle Atlantic
South Atlantic
Pacific
Size C lasses
0-99 employees
100-249
250-499
500-999
1000+
O w n e rsh ip T y p e
Single Plant
Multiple Plants

n = 4
.11

n = 6
.13

.10
.02
.17
.05
.11
.08

.11
.02
.19
.05
.13
.10

.10
.08
.14

.11
.10
.16

.18
.08
.09
.06
.03

.21
.09
.09
.07
.04

.17
.08

.20
.10

Note:
(1)
All table entries are based on equation (2) in the text using pooled sample data for
1978 and 1983. (2) The table shows entries for selected industries and regions, including
the extremes.




Table 6. Estim ated Fraction of Cross-Sectoral Variation in Job Reallocation Rates
Due to Learning about Initial Conditions
Sectoral Classification by:
Industry Region
Size
Cross-Sectoral Standard
Deviation of Job
Reallocation Rates
Fraction of Cross-Sectoral
Variance Explained by
Learning About Initial
Conditions, Assuming n = 4
Fraction of Cross-Sectoral
Variance Explained by
Learning About Initial
Conditions, Assuming n = 6

Ownership

.042

.026

.056

.050

.32

.39

.51

.56

.36

.48

.57

.62

Notes:
(1) All table entries are based on the pooled sample data for 1978 and 1983.
(2) Rows two and three report the quantitiy 1 — {V / V ). V is defined as the crosssectoral variance of job reallocation rates. V is defined as the cross-sectoral variance of
adjusted job reallocation rates. The adjusted sectoral reallocation rate equals the observed
rate minus the contribution of learning as estim ated from equation (2).




Table 7. Fraction of Excess Job Reallocation Due to Between-Group Employment Shifts,
Means of 1978 and 1983 Values by Two-Digit Industry
Age
8

Size
5

Region
9

Ownership
2

A ll1
720

.13
.12
.18
.20
.01
.08
.12
.08
.06
.26
.16
.12
.04
.07
.05
.12
.10
.09
.11
.20

Group Type
No. of Groups
Industry
Food
Tobacco
Textiles
Apparel
Lumber
Furniture
Paper
Printing
Chemicals
Petroleum
Rubber
Leather
Stone, Clay, Glass
Primary M etals
Fabricated M etals
Nonelectric Mach.
Electric Machinery
Transportation
Instruments
M iscellaneous

.01
.03
.01
.11
.00
.03
.08
.06
.00
.07
.12
.03
.00
.01
.00
.00
.00
.00
.00
.05

.01
.05
.01
.10
.02
.04
.08
.05
.00
.21
.01
.06
.02
.05
.00
.00
.02
.01
.02
.05

.00
.01
.03
.02
.00
.00
.00
.00
.00
.00
.00
.01
.00
.00
.00
.00
.00
.00
.00
.06

.36
.62
.39
.46
.36
.45
.43
.39
.39
.65
.48
.52
.39
.31
.23
.31
.33
.35
.50
.56

Means of 1978 and 1983 Values for Total M anufacturing
Group

Age

Region

Type
Number 8
.06

5
.00

9
.00

Owner­

2-Digit

4-D igit

All,

ship

Size

Ind.

Ind

ex. Ind.1

2
.00

20
.01

450
.12

720
.15

AH2

14400
.39

Notes:
1 Based on a grouping o f plants by age, size, region, and ownership type simultaneously.
2 Based on a grouping of plants by age, size, region, ownership type, and two-digit industry
simultaneously.3
3 Approxim ately 11,000 group cells are nonempty.




Table 8. Decomposition of Time-Series Variance of Job Reallocation, Creation and Destruction
S ecto ral C lassification Schem e

# of Sectors

T otal.
Mfg.
1

4-digit

2-digit

450

20

2-digit,
Size
100

2-digit,
Age
40

2 -d ig it,
O w n er
40

2-digit,
R egion
180

Fraction of Job Reallocation Variance (S U M t) Accounted for by:
(a) sec to ra l/a g g .
m ean effects
(b ) id iosyncratic
effects
2C ov(a,b)

0.03

0.105

0.044

0.044

0.042

0.051

0.053

1.026

0.797

0.876

0.816

0.879

0.838

0.917

-0.056

0.098

0.079

0.140

0.078

0.111

0.030

Fraction of Job Creation Variance (P O St) Accounted for by:
(a) sec to ra l/a g g .
m ean effects
(b ) idiosyncratic
effects
2C ov(a,b)

1.44

1.318

1.395

1.431

1.388

1.459

1.385

0.16

0.124

0.136

0.142

0.138

0.149

0.142

-0.60

-0.442

-0.531

-0.573

-0.526

-0.609

-0.526

Fraction of Job Destruction Variance (N E G t) Accounted for by:
(a) sec to ra l/a g g .
m ean effects
(b ) idiosyncratic
effects
2C ov(a,b)

0.63

0.705

0.658

0.726

0.664

0.680

0.665

0.079

0.062

0.068

0.063

0.067

0.066

0.071

0.287

0.233

0.274

0.211

0.288

0.254

0.264

N otes:
(1) E n trie s in th e to p p anel a re b a se d on th e variance d eco m p o sitio n in eq u atio n (9).
E ach e n try re p o rts th e ra tio o f th e in d icated te rm on th e rig h t side of (9) to th e
te rm on th e left side. E n trie s in th e second a n d th ird p an els axe b ased on analogous
variance decom positions for jo b creatio n a n d d e stru ctio n .
(2) Size, region a n d ow nership secto rs axe defined as in T ables 4-6.
(3) T h e re axe tw o age groups: young p la n ts (0-9 years) a n d old p la n ts (1 0 + years).




T a b le 9. C y clical B e h a v io r o f S e c to ra l J o b R e a llo c a tio n R a te s:
T im e-S erie s C o rre la tio n s B etw een N E T st a n d J o b R e a llo c a tio n M easu res

S e c to ra l C lassific atio n S ch em e
T o ta l.

4 -d ig it

2-d ig it

M fg.

2-d ig it,
Size

2 -d ig it,
A ge

2-d ig it,
O w ner

2-d ig it,
R eg io n

Summary Statistics on Correlations of N E T st with SUM st •
S ize-w eighted
A vg. C o rre la tio n

-0.57

-0.35

-0.51

-0.37

-0.49

-0.39

-0.39

( # < 0 ) /T o ta l

1/1

3 6 3 /4 5 0

20/20

8 5 /1 0 0

2 7 /4 0

2 9 /4 0

1 4 6 /1 7 7

Summary Statistics on Correlations of N E T st with S U M S :
J
S ize-W eighted
A vg. C o rre la tio n
( # < 0 ) /T o ta l

-0.64

-0.36

-0.55

-0.41

-0.50

-0.45

-0.42

1/1

3 6 0 /4 5 0

20/20

8 7 /1 0 0

2 9 /4 0

3 1 /4 0

1 5 2 /1 7 7

S e c to r T y p e
D u ra b le

N o n d u r.
S m all

T w o -D ig it I n d u s tr y by:
L arg e
Y oung
O ld

S in gle

M u lti

Summary Statistics on Correlations of N E T st with S U M st:
S ize-w eighted
A vg. C o rre la tio n

-0.61

-0.35

-0.13

-0.61

-0.08

-0.70

-0.09

-0.48

( # < 0 ) /T o ta l

1 0 /1 0

10/10

3 1 /4 0

3 7 /4 0

7 /2 0

20/20

10/20

1 9 /2 0

„-----

S T

Summary Statistics on Correlations of N E T st with S U M st :
S ize-w eighted
A vg. C o rre la tio n

-0.65

-0.40

-0.20

-0.63

0.06

-0.71

-0.19

-0.53

( # < 0 ) /T o ta l

1 0 /1 0

10/10

3 3 /4 0

3 8 /4 0

9 /2 0

20/20

11/20

20/20

N otes:
(1) S ecto rs a re d efin ed as in T a b le 8.
(2) “S m all” refers to th e fo rty se c to rs w ith p la n ts in th e 0-99 a n d 100-249 size classes.
“L a rg e ” refers to th e fo rty se c to rs w ith p la n ts in 500-999 a n d 1 0 0 0 + size classes.




Table 10. Regressions of Adjusted Sectoral Job Reallocation Rates on
Own-Sector and Manufacturing Net Growth Rates
Dependent Variable in Regressions: Adjusted Sectoral Job Reallocation Rates
S u m m a ry o f R e g re ssio n s a n d G o o d n e ss-o f-F it M e a s u re s
Sectoral Classification Scheme
2-digit
industry
industry
industry
by age
by size
Regression Number
# of Observations
# of Fixed Effects
(one for each sector)
# of O ther Regressors
Regression R 2
Fraction of Time-Series Variation
Explained by O ther Regressors

industry
by ownership

(1)
220
20

(2)
440
40

(3)
1100
100

(4)
440
40

2
.78
.27

4
.78
.08

6
.53
.05

4
.75
.11

Im p lie d R e s p o n s e o f J o b R e a llo c a tio n R a te to N e t J o b G ro w th
Sector Type:

Ind.

Young Old

Small

Med.

Large

Single

Estimated response of sectoral job reallocation rate to a one st. dev. increase in:
-1.82
-1.15
-.13
-.18
-1.51
Mfg. Net Growth R ate
-1.38
.20
.14
.60
.21
.17
Standard E rror
.12
.39
.25
-.24
.11
-.26
.55
Own-Sector Net Gr. Rate
-.13
-.38
-.16
.06
.79
.10
.07
Standard E rror
.05
.19
.10
Based on regression # :
(2)
(2)
(3)
(3)
(3)
(4)
(1)
(1)

(2)

(3)
(4)

(5)

Multi
-1.46
.13
-.23
.13
(4)

Notes:
In regression (1), “other regressors” refers to the m anufacturing net growth rate (gt)
and the own-sector net growth rate deviated about the m anufacturing growth rate
(g9t). Relative to regression (1): regression (2) adds interactions of these variables
with one age-group dummy; regression (3) adds interactions with two size-class dum ­
mies; and regression (4) adds interactions with one ownership-class dummy.
Time-series variation equals the residual variation after eliminating sectoral fixed ef­
fects. Regressing this residual variation on the “other regressors” yields the entries
titled “Fraction of Time-Series Variation Explained by O ther Regressors.”
E stim ated responses in the bottom panel are multiplied by 100.
In computing the estim ated responses in the bottom panel, a one standard deviation
increase in the own-sector net growth rate is measured as the size-weighted average
of the time-series standard deviations of sectoral growth rates. This m easure isolates
the m agnitude of time-series variation in the gst.
The standard errors in the bottom panel are heteroscedasticity consistent.