No.

24, 1980
THE DURATION OF UNEMPLOYMENT AND
UNEXPECTED INFLATION - AN EMPIRICAL
ANALYSIS
by
Anders Bjrklund and Bertil Holmlund
June 1978
Revised August 1979
This paper has been accepted for
publication in the American Economic
Review.
l. lntroduction*
A distinguishing message of the theory sea unamploy-
ment is that short-run unemployment fluctuations are explain-
able by surprises. Unemployment is basically
viewed as productive investment in job search, chosen by employ-
ees in order to enhance their lifetime earnings. An increase 1n
aggregate demand will imp1y a temporary fall lin unemployment
due to short-run deviations between actual and expected wages;
workers are faoled into accepting more employment.
Th; s information-l ag interpr'etat ion of changes in unemp 1 oyment
might be compared to an alternative view, where the quantity-
rationing rules of the labor market are emphasized. A rising
aw of abor ft'om unemp l oyment to emp l oyment i s, accordi ng to
this theory, caused by the relaxation of job-rationing con-
straints rather than unanticipated inflation.
In this paper we address ourselves to the question of the
empirical importance of the two competitive explanations. The
two stories are, of course, not mutually exclusive; we try,
via a fairly simple specification, to capture both views in
one equation. The principal contribution of our study lies
in its ability to provide information about the relative
importance of unexpected inflation and job opportunities as
explanations of the duration of unemployment. l Another in-
teresting feature of our paper is its comparative per-
spective; we app same model to both Swedish and U.S.
data, thereby being able to reveal certain important dif-
ferences between the labor markets in the tilJa count es. ~ e
find e.g., perhaps somewhat surprisingly, that the U.S. un-
employment duration is more or less unaffected by unexpected
inflation, whereas the results for Sweden, on the other hand,
give some support for the information- is.
2
novelty of our study ;s the disaggregated data used (for Sweden
only). By focussing the analysis on transition probabilities
for workers with different lengths of (incompleted) spells,
same interesting behavioral differences are observed; one
fi is s e information-l story is more valid
for the short-term unemployed.
The paper is organized as fol1ows: Section II below introduces
the basic theoretical framework that guides Gur empiricai
estimation procedures; the latter are described in section III.
Section IV presents the data employed and section V the empiri-
cal results. Some interpretations of Dur findings are discussed
the final ian.
II. Optimal Search Policies and the Duration of Unemployment
Microeconomic explanations of unemployment have been focussing on
the behaviour of the household, whereas the demand side generai-
ly has been eonsi as exogenous. We will follow that rtial
equilibrium approach, using a simple job search model as our
theoretical framework.
Consider the behaviour of an unempioyed worker according to
$ e an
3
which assures him an income greater than what he might have
received by continued search. The decision is affected by
the perceived location of the wage offer distribution. If
a monetary contraction produces a left-ward shift of the
wage offer distribution - or a lower rate of wage inflation -
this change in general market conditions is assumed to be
imperfectly detected by job seekers, who mistakenly blame
local circumstances rather than changes in aggregate demand.
Unemp 1 oyed ltIorkers wi 11 search a longer time causing
l ength of spe 11 s unemployment to rise.
A common assumption in standard search models is that the
number of job offers received per period equals one. The pro-
bability of leaving unemployment - the transition probability -
is then sol ely determined by the job seeker's offer-acceptance
probability. The simplifying job offer assumption is, however,
tion the case with random number of job offers is easily in-
corporated into the b a s i ~ search theoretic framework. Consider
the job-seeker's transition probability, which - in the absence
of labor force exits - equals the hiring probability. Decompos-
ing the transition 1
.
l ~ ) into two components the
job offer probability (8) and the acceptance probability (P)
we have
(1 ) ]J = ep = 8[1-F(a)J
8 S.
- where a is the reservation wage and F(o) the distribution
4
function of wage offers. If the transition probability is
eons tant during search, the expected duration of unemploy-
ment (D)i s
( 1
Which are then the characteristics of an optimal search
policy? In the simple case of infinite time horizon and no
diseounting) the n t
variant reservation wage obtained as the solution to
GO
(3)
c = 8P[E(wiw>a)-aJ = ej(w-a)
a
where C is the (constant) marginal search cost and f()
known density function of wage offers. L. Eq. (3) imp!ies
that the reservation wage declines as the job offer probabi-
lit Y e decreases. Likewise, a known leftward shift of the
wage offer di bution 11 a l so reduce res on
vJe have so far br; efly outl ined the bas ie search story, str; ct-
ly valid only in a stationary world. Now consider the possibili-
ty of fluctuations in aggregate demand, influencing the job-
seekerls transition probability via the job offer probability
(more vacancies) and/or via imperfect reservation wage adjust-
ments. Three different effects may be identified:
Li' i numoer of vacancies
means a higher job offer probability, thereby reducing the dura-
5
tion of unemployment.
2. A permanent increase of the job offer pro-
bability will increase the expected returns from search, thus in-
creasing the workerIs reservation wage. It follows that the un-
employment effect of a rising number of vacancies is ambigous
a p n (1977) , demonst the
availability effect will outweigh the supply effect under cer-
tain reasonable assumptions.
affect location of the wage offer distribution. Assuming
a lag in the d,iscernment of a rising rate of inflation, reserva-
tion wages will be unaffected in the short run, implying a ris-
ing flow of new hires from the pool of unemployed.
Summarizing these three effects we have:
(4 ) II = 8(V)P(V ,w/w*) == g(V
+ +
where V is the number of vacancies, w the actual average wage
and w* the expected average wage.
We would argue that Eq. (4) represents the kernel of the search
theory of cyc1ica1 unemployment. The standard search model out-
lined does rely on some very restrictive assumptions, e.g. a
stationary wage offer distribution, fixed leisure time and a con-
stant job offer probability. More complex search models, e.g. those
Siven (1979) and Seater (1977, 1978
10
1...1 ) are, however, fairly
6
r is on
3
unexpected inflation and vacancy contacts. We are suppressing
other plausible determinants of unemp var-
htions in unemployment compensation and the discount rate.
These simplificat should not be too severe, since the cy-
clical uctuati na:t i We have also
excluded changes in the pr;ce level from consideration, perhaps
a more questionable simplification. Unexpected price inflation
does affect unemployment in same models within the microfounda-
tions literature, although it is absent in the standard search
v:ition regressor is, however,
quite different in e.g. the Lucas-Rapping model compared to
the Siven model (rnisperception of future prices versus mis-
perception of current prices) and the theoretical predictions
are completely opposite; a higher rate of unexpected price in-
flation will increase unemployment in Siven's model and decrease
1
4 .
unemployment in the Lucas-Rapping mode. It 1S interesting to
uni G rescnts 0.
tions in unemployment are totally unaffected by how workers
5
\.if: i ..... c
to ude
"""' ....-
the price inflation variable from the regressions, thereby
avoiding troublesome problems of interpretation.
III. Empirical analysis
A straightforward method of investigating the validity of the
detection-l hypothesis is to specify explicit transition pro-
bability equations with vacancies and unexpected wage increases
as explanatory variables, i.e. to represent Eq. (4) above by a
suitable functional form. The basic specification used will be:
(5) xn]l
v t
The obtained a
2
-estimate reflects the net result of the
positive availability effect and the negative supply effect;
7
intuition and some theoretical predictions suggest that a
2
(the
net availability effect) will have a positive sign.
6
The main problem with the approach chosen is, of that it
an ana is as \;Je 11 as nce
no direct data about expected wages or wage-changes are available
some model of the formation of expectations must be used. The ex-
panding literature about the formation of expectations give sever-
al alternatives which all are quite plausible. However, no model
v/hich can made operati can be considered ;jcorrect
H
'in a l
respects. Our approach has been to try three different models in
order to investigate how robust the information-lag-hypothesis
is with respect to the different specifications. Two of the ap-
plied forecasting functions are.consistent with the idea that
workers 1 earn past reestimating the parameters of
their forecasti equations when more information is obtained.
The first model used is a type of adaptive expectations. These
expectations are formed according to a finite distributed lag
of past wage changes, i.e., wi
for ):
(6a)
4
= l:
i
quarterly data (which is used
w .
9,. ( _)
l W 4'
t- -l
where
4
4
( Z
(5-i )
:::;
i == 1
i == 1
and viith monthly data (which is used for the U. S. )
w* 12 ItJ ,
(7a)
(t )
r_1
= L: Q,.
(. ~ ~ )
W l
t ~ 2
i = 1
v". ') .
~ l __ -l
where
12
l
12
(7b) E Q,. =
E (13-i) ==
1.
Models like these - where the sum of the weights has been
constrained to,vne - are of ten used in empirical work even
though ;t has been pointed out that the theoretical basis is
quite weak. ( e.g. Persson (1979), where it is shown that
the sum should equal one on1y in very special cases if the
forecast is to optimal. )
Even though the simplici of the simple adaptive model is
-. .
! ; ~ since it mi r-s
expectations in a simple and cheap way - 1t could also be argued
that individuals have some knowledge about historica1 regulari-
ties of wage changes, and that they use this information when
forming their expectations. One possible way to represent
these regularities is to apply a time-series approach. The as-
sumption is that people have in their mind an auto-regressive-
moving average-process (ARMA) which is generating forecasts from
9
ad to 80th s ficaticn e pe,r;;,mGters of
this process are, however, likely to be revised when peop1e re-
ceive more information about wage-changes. Therefore we have
proceeded as 1 Oi'/S:
The process has been reestimated each period and reidentified
each fourth period (with quarterly data) and each twelfth period
(with monthly data). 7 For Sweden the character of the process
changed over time; when observations from 1960 onwards were used
the appropriate process changed from an AR(l) to an AR(1)MA(2),
back again to an AR(l) and final1y - during the past two years
(1976-1977) - an MA(10) on the first differences of the variable
(i.e. the process was non-stationary). All the time autoregressive
seasonal terms had to be used.
For the U. S. the process \tIas stat; onary when data from 1960 un-
til 1969 were used.-AR(l) with first a seasonal autoregressive
term and then a seasonal moving average term. From then on the
process became non-stationary with an MA(l} term and a seasonal
moving average term on the first differences.
It could, finally, be argued that workers are still more rationa1
than using only from an ARMA-process of wage-changes.
They might even have in an empirical model incorporating
different economic variables. An unemployed worker forming his
10
expectations may e.g. use a wage-equation of the illips
curve ions
(quarterly data) like:
where
data from the last five years. On the whole the estimated equa-
tians performed reasonably wel1 for Sweden according to standard
This approach was less successful for U.S.; the available vacancy
indicators turned out to be bad predictors of wage inflation. We
to ude this expectations-formation scheme for the
U.S. regressions.
IV. The data
Swedish transition probabilities have been estimated as fol1ows:
The rotating system of the Swedish Labor Force Surveys is con-
structed so that almost 90 % of those who are interviewed in ane
survery are interviewed again three months later, whereas dif-
ferent individuals are interviewed in two subsequent manths.
In order to improve the estimates we decided to campute quarter-
ly transition p lities.
11
Oenoting the number of unemployed for at least a
less than b weeks at time t by and the weekly inflow
inta unemployment by f we can describe the estimates as follows:
1,14
13
(9) = f Z; ( l
)1
i =0
1 ')
]13
( 10) ==
[ 1
(11 )
G
27
,39
::::
r.:: 14,26 [ l _1 ] 13
t+26
\.:lt+13 11
3
Three transition probabilities are obtai
can be regarded as conditional upon the length of the spel1 of
unemployment. By using available data on f, etc.
from
( 12)
whereas
11
2
and 11
3
are calculated as
1/13
rG
14
,
27
1
(13 ) 1
I t I
Jl
2
::::
-
I
G
l,l3 ,
t t J
1/13
fG27,391
(14 )
Jl
3
::::
l
I t+26 i
-
I I
I G
14
, 26 I
l t+13 J
The Swedish vacancy statistics are from labor market statistics,
published by the National Labor Market Board. Quarterly wage data
are obtained from the labor market issues of Statistical Reports,
published by National Bureau of Statistics. All data used
to manufacturing industry.
The U.S. transition probabilities refer to the labor market
as a whole. They were computed by uSing the method proposed
Barror. (1 ). The essential idea is to campare the number
of people in one week who have been unemployed less than ve
weeks with the number of people four weeks later who have been
unemployed five to eight weeks. The difference consists of
people who have left the pool of unemployed. The duration data
reported in Ernployment and Earnings are grouped in the classes
12
1-4 weeks, 5-14 weeks etc., which requires a slight modification
out1i deta 15, see rran.
The U.S. wage data are average hourly earnings in manufacturing
'd t td' r l dE' SA d
111 repor e ln cmp oyrnent an arnlngs. s vacancy ata
for the period 1965-1 we used the Help-wanted advertising in-
dex (HWA) published in Main Economic Indicators (OECO). For the
period 1969.4-1973.10 manufacturing vacancies (Vm) according to
establishment data were also tried (Employment and Earnings);
the latter series are available only for (approximately) this
period.
V. rical results
Tables 1 and 2 below. The estimation method is weighted-least-
squares and ate weights are derived in an appendix.
Let us first have a look at the results obtained for Sweden.
We observe, in the first place, that the detection-lag variable
is significant both for the short-term (1-13 weeks)
13
Table l. Transition probability equati9ns for
-Quarterly data 1968.1-1977.3
Adaptive expectations V w/w*
-2
R DW

( 1 ) 0.81 10:'30 0.60 1. 75
(4.29) Ut. 1O}
(2) 1.11 0.42 1. 57
(5.36)
l'":!. \
\v) 14. 0.41 1. 05
(5.18)

(4) 0.34 1. 97 0.33 2.27
(3.24) (1.69)
(5) 0.41 0.30 2.29
(4. 11 )
(6) 3.42 O. 16 1.84
(2.83)
(7) 0.39 -3.35 0.09 2.16
(2. 19) (-1. 57)
(8) 0.31 0.05 2.28
(1. 78)
(9) -L 0.004 2.03
(-0.93)
ARMA expectations

(10) 0.98 7.47 0.50 1.43
(4.94) (2.59)
(11 ) 11.00 O. 18 0.79
(3.08)
conto
ARMA expectations
v w/w*
-2
R DW
(12 )
(13 )
unemploved (11-)

(14 )
Expectations from wage-
equations
)
(15 )
(16 )

( 17)
(18 )

(19)
Nate:
R
2
.
1.8 the fraction of
0.36
(3.56)
0.36
)
1. 13
(5.93)
0.40
(4.23)
0.30
(1.72)
2.21
(L 64 )
3.56
(2.41)
-2.16
(-0.81)
8.43
(2.79)
7.76
(1. 85)
3. 10
(2.38)
3.37
(2.14)
-3.20
(-1.31)
0.33
O. 11
0.05
0.51
0.06
O.
0.09
0.07
the weighted variance of the
dependent variable explained by the weighted independent
2. 19
1.72
2.27
1.66
0.65
"
L.LI
1.63
2.20
-2 .
variables, adjusted for degrees of freedom. The R obta:i.ned
A
when regre8sing on 11
1
from Eq. (1) was 0.62.
14
15
Table 2. Transition probability equation for the U.S.
Monthly data 1969.4-1973.10 and 1965.2-1975.12,
respectively
jl,daDt i ve
expectations HWA
Vm
w/w* TIME Fi
2
DW
P
1969.4-1973.10
--------------
0.23 1. 62 -0.0008 0.73 1. 19
(11.27) (1. 59) (-1.61)
2 0.21 0.91 -0.0002 n.a. 2.02 0.29
(8.57) (1.11) (-0.41)
3 0.24 1. 42 0.72 1.13
(11.81) (1. 38)
4 0.21 0.87 noa.
')
<-.
(8.71) (1. 08)
5 O. 14 -0.0021 0.07 0.37
(0.08)
! ';f"'" \
\
.:Jo;
6 0.50 1. 36 -0.0031 0.72 1. 18
(1. 33) (-6.39)
7 0.44 0.71 -0.0022 n.a. 1. 97 0.31
(8.21) (0.86) (-3.84)
8 0.45 0.21 0.51 0.67
(7.64) (0.15)
1965.2-1975.12
--------------
9 0.52 0.70 -0.0025 0.83 1. 34
(16.81) (1. 28)
10 O. 0.47 -0.0025 n.a. 2.03
0.34
(11. 45) (0.81) (-13.26)
11 0.49 1. 55 0.33 0.34
(7.93) (1. 43)
12 0.71 -0.0024 0.47 0.44
(0.73) (-10.62 )
conto
16
Adaptive
expectations HWA Vm w/w* TIME
-2
D ~ J R p
1969.4-1973.10
--------------
13 0.23 2.57 -0.0007 0.74 1. 15
(11.27) (2.07)
. , ~ 4)
(- I j .
14 0.20 1. 96 -0.0002 n.a. 2.03 0.30
(8.63) (1. 98) (-0040)
15 O. 2.48 0.73 1. 10
(11.92) (1. 98)
16 0.20 1. 93 n.a. 2.04 0.31
(
fl
';. ~ "-
17 0.24 0.71 1. 12
(11. 64)
18 0.20 n.a. 2.04 0.30
(8.68)
19 3.03 -0.0021 0.10 0.36
l 31) (-2.49)
20 0.49 ~ 4 8 -0.0030 0.73 1. 17
(11.23) (2.00) (-6.44)
21 0.44 1. 81 -0.0022 n.a. l. 99 0.30
(8.36) (1. 80) (-3.96)
22 0.44 2.23 0.53 0.65
(7.71) (1. 34)
1965.2-1975.12
--------------
23 0.52 -0.37 -0.0026 0.83 1. 33
(16.7) (-0.50) (-19.19)
24 0.53 -0.07 -0.0025 n.a. 2.04 0.35
(1 L 37) (-0.10) (-l3.01)
25 0.49 3.45 0.35 0.41
(8.03) (2.48)
26 -0.09 -0.0025 0.46 0.44
(-0.07) (-10.31)
Note: p is
the first-order autocorrelation coefficient obtained
by using the Cochrane-Orcutt approach.
17
and for the medium-term unemployed (14-26 weeks). These results
Q
hold for all models of expectations.
J
For the long-term unemployed,
on the other hand. no significant detection-lag effect is revealed;
the coefficient has even a wrong sign.
all regressions, even for the long-term unemployed. Dropping this
variable produces in most cases a marked decrease in the DW-value,
indicating the presenee of specification errors.
Which are then the economic interpretations of the different re-
sults for the three groups of unemployed? No straightforward
answer is available, partly because the "hypothesis-testing in-
cludes a joint test of the underlying model and the expectations-
la
generating mechanism
ll
The absence of any significant detection-
lag effect for the long-term unemployed may have at least two
and/or the variable reservation wage hypothesis could be errone-
DUS. There are arguments in favourof both these interpretations.
In the first p l c e ~ it makes sense to hypothesize that the 10ng-
term unemployed (more than six months in our data) are better
informed about the actual wage offer distribution, simply because
they have experienced a longer period of "learningll through full
tlme job search. This es
the forecasting function might differ across workers with dif-
ferent unemployment histories.
18
The second interpretation stated above (the possible unrealism
of the variable reservation wage hypothesls) may be elucidated
by recal1ing same familiar results from search theory: The re-
servation. wage of a job-seeker with finite search horizon will,
under some stationary conditions, fall with the duration of un-
emp1oyment, a theoretical prediction which has been given empiri-
cal support. llEventually the reservation wage will coincide with
the minimum value of the wage offer distribution, implying an
acceptance probabi1ity equal to one. In that extreme case a11
job offers are accepted and there is no detection-lag effect.
Both of the hypotheses outlined are consistent with the results
obtained. Intuition would suggest that both of the mechanisms are
in operation to same extent, reinforcing each other and thereby
producing the observed results.
Since both the (net) availability effect and the detection-lag
effect are significant, it is important to find out the relative
i ca 1
variations of the duration of unemployment. To find out this
of the independent variables inta account. The question might be
illuminated by comparing the predicted transition probabil ities
using estimates from regressions in the table
'"
( 15)
]Jt
=
al
.
V
t
.
(w*)
t
j
l ~
{, ~ '
r;ea eion is t- e:; .
,
perfectly foreseen (w
t
= w*)
t
(16)
a
2
]l = a . V
t t l
Using the results from the adaptive mode l Figure 2 below de-
monstrates the relative unimportance of the detection-lag
effect for the medium-term unemployed. Inflation surprises
produce, on the other hand, quite important unemployment effects
for the short-term unemployed during the peak years 1969-70 and
1974-75. (Figure 1.) The main part of the variation is, however,
attributable to the vacancy-variable.
Turning now to the U.S. regressions, the dominant avail lit Y
effect is even more pronounced than in the Swedish case. The
vacancy variables used are highly significant in all regressions
whereas the detection-lag coefficient is fairly sensitive with
respect to the choice of expectations mode l and pe-
riod. A significant detection-lag effect is obtained only by
applying an ARMA-expectations-generating mechanism for the pe-
riod 1969;4-1973;10. These results are independent of the choice
of vacancy variable. Exclusion of the latter alsogives rise to
a strong decline in the DW-statistic, indicating specification
errors. When the estimation period is extended (1965;2-1975;12),
the significance of unexpected inflation 12 It should
a1so be noted that a negative and significant trend-coefficient
is obtained when HWA is used as vacancy variable.
The main conelusian from these excercises on U.S. data is that
the job-availability variables are the dominant determinants of
19
20
Figure I. The effects of unexpected inflation - short-term unemployed
in Sweden
0.25 ~ .
0.20
I
r
i
\
i.-
I
L
\
I
0.15
!
I
t\
\
l-
I
L
l\
i
i
\
r
\
\
i I
O. 10
I
t
~
I
I
r
ti
0.05
'I
i
'" '
~ \
I
\=J
t-
o
I
,
. I I ! . I ; ,
. ! ! .
,
! . ! I ! i .
,
I
68 69 70 71 72 73 74 75 76 77
predicted transition probability
predicted transition probability when inflation is perfectly foreseen
21
Figure 2. The effects of
flation medium-term un
in Sweden
0.14
0.
06
1
----- ---------------:
T, i I t i t f , t r f i , t ; t f t , , I f I i I
68 69 70
]l
72 73 74 75 76 77
predicted transition probability
____ predicted transition probabi1ity when inflation is perfectly foreseen
however, rule out the possibility of same deteetion-lag ef-
feets in operation, at least during certain time-periods and
- espeeial1y - if the expectations are farmed according to an
ARMA-process rather than adaptively.
VI. Concluding remarks
job search literature there a tendency tu ove ook
22
the impartance of vacancy contacts as determinants of the dura-
tion of unemployment; the emphasis instead being placed on in-
flationary surprises. This (mis)use of the seareh stOt'y does not
necessarily follm'i from the 10gic of the theory; most search models
do recognize the significance of the stream of jab offers. The pop-
ularity of the detection-lag view is, probably, its ability to
provide a reasonable interpretation or the short-ruQ Phillips
curve. The transmission mechanism of aggregate demand policies is
explicated in a fairly simple ""ay: an increase in the money growth
rate \'1111 increase inflation thereby faoling the acceptance deci-
sions of job seekers.
In this paper we have that this ew has some empi cal
validity, at least for the short-term unemployed and forc alabor
market lika But we have also shown that unexpected in-
flation can explain only a small part of the actual fluctuations
in unemployment duration. Since the flovl inta unemployment is
fairly stable over the eyele. our results imply, moreover. that
cyclical changes in the unemployment rate are only sliqhtly affect-
ed by inflationary surprises.
23
The elementary search model - where variations in the job offer
probability are disregarded - is then clearly inadequate as an
curv so
rule out one of the mechanisms which imply a vertical long-run
Phillips curve; the natural rate theory must of course be valid
cyclical changes in unemployment. The results are thus more in
accordance with the II mainline
ll
view of inflation and unemployment
stressing that aggregate demand influences employment and unemploy-
ment via the relaxation of job rationing constraints rather than
via misperceptions of relative wages. It is possible that unanti-
cipated price inflation may be of some importance even with;n
the latter framework - as a determinant of the flow of vacancies
inta the labor market. We are, however, unaware of solid theoret-
ical work on that issue.
Let us, finally, offer some comments to the observed differences
ly unionized labor market and wage bargaining at the national
level gives rise to relatively uniform and long-term wage con-
tracts. One would be inclined to expect that this institutional
setting would produce fast dissemination of information about
the wages in general. thus reducing the importance of information-
lag effects. The less unionized U.S. labor market is probably
more r ~ l i o.r
13
than the Swedish is and the scope for temporary wage-mispercep-
tians would therefore be greater. In fact, we find the opposite.
ore itional significant di
24
labor market functioning in Sweden u.s. - the importance
of temporary layoffs. Temporary layoffs constitute - as Martin
14
Feldstein has out - an source of U.S. un-
employment. The U.S. manufacturing layoff rate has varied be-
tween 10 and 20 percent (of the number of employed workers)
per year whereas the corresponding Swedish figures are 2 - 4
percent. The major part (60 - 70 percent) of the U.S. layoffs
are temporary, implying that most workers are ultimately rehired
by the same employer. Temporary layoffs in Sweden are, on the
other hand, very unusual. Unemployed workers on temporary lay-
accounted r 2 - 3 percent of ish du ng
the period 1975-1978. The corresponding U.S. figures seem to have
fluctuated between la and 20 percent.
15
Feldsteinls view of those
laid off as liwaiting" rather than IIsearching" has been questioned
on empirical grounds.
16
The Feldstein-hypothesis mi.ght, however,
be considered as modest ly corroborated by our results; one in-
teresting interpretation of our revealed U.S.-Sweden differences
would be that the extent and intensity of job-search among the
unemployed is lower in the U.S. If unemployed workers on layoff
act as if they will be recalled - and therefore abstain from
search - there is little scope for detection-lag effects of
the traditional type.
A laid off worker IIhas a job" in some sense; he is attached to
a particular firm and expects to be recalled by his employer.
He is probably also we1l informed about wage changes in his firm.
How would then a non-seeking lmemployed worker on layoff respond
to unexpected general wage inflation? He would, most likely, be
25
inclined to search, s
fel1ows; a familiar implication of search theory is that quits will
decrease to sea as a response
to unexpected wage increases. Clearly, tempora layoffs re-
present a middle state between loyment and unemployment.
Economic theories designed to explain individual behavior
in the polar cases would obviously be less suitable when
applied to the middle state.
REFERENCES
R. Axelsson and K. G. Lfgren, IIThe Demand for Labor and Search
Activity in the Swedish Labor Market", Europ. Econ. Rev., 9,
19
J.M. Barron, "Search in the Labor Market and the Duration of Un-
employment: Some Empirical Evidence
il
, Amer. Econ. Rev" Dec. 1975,
65, 934-42.
T. F. Bradshaw and J. L. Scho 11, IIThe Extent of Job Search during
Layoff
ll
, Brookings Papers, Washington 1976, 2, 515-26.
Darby, "Three-and-a-Half Million U.S. Employees Have Been
fvtislaid: Or, and Explanation of Unemployment, 1934-1941
11
,
J. Polit. Econ., Febr. 1976, 84, 1-16.
14. Fel dstein, "The Importance of Temporary Layoffs: An Empi ri ca l
AnalysisIi, Brookings Papers, Washington 1975, 3, 725-44.
R. Feinburg, "Search in the Labor Market and the Duration of Un-
employment: Note
ll
9
Amer. Econ. Rev., Dec 1977, 67,1011-13.
R. Gronau, IIInformation and Frictional Unemployment"., Amer. Econ.
l 1. 1. 290-301.
H. Kasper, "The Asking Price of Labor and the Duration of Unemploy-
ment!!, Rev. Econ. and Statis. ) May 1967, 49, 165-72.
J.R. Kesselman and N.E. Savin, "Three-and-a-Half Million Workers
Never Were Lost", Econ. Inquiry, April 1978, 16, 205-225.
N.M. and G.R. Neuman, liAn Empirical Job-Search Model with/
a Test of the Constant Reservation-Wage Hypothesis
ll
, J. Po1it.
Econ., Febr. 1979,87,89-107.
S.A. Lippman and J.J. fvlcCall, II The Economics of Job Search:
A Survey: Part I!!!j Econ. Inquiry, June 1976, 14, 155-89.
R.E. Lucas, JR.and L.A. Rapping, "Real Wages, Employment and
Inflation", in S. Phelps et al. (1971), pp. 257-305.
C.R. Nelson, Time Series Anal Fore- ___ _________ =_ ___ __ __
casting
ll
, Holden-Day, San Francisco 1973.
M. Persson, Inflat ions and the Natural Rate

Hypothesis, Stockholm School of Economics (Diss.), 1979.
E.S. Phelps, "Introduction: The New Microeconomics in Employment
and Inflation Theory"., in E.S. Phelps et al. (1971), pp. 1-23.
et al, ilMicroeconomic Foundations of Employment
and Inflation Theory", !vlacmillan, London 1971.
J
Il
Off: A Critique of the Literature
ll
, J. Econ Lit., June 1978,
499-544.
J.J. Seater, ilA Unified Model of Consumptian, Labor Supply, and
27
"Util ity fv1aximizati on, Aggregate Labor Force Behavi or,
and the Phillips Curve
ll
, J. Econ., Nov. 1978, 4, 687-713.
, IIJob Search and Vacancy Contacts
ll
, Amer. Econ. Rev.,
June 1979, 69, 411-19.
North Holland, Amsterdam 1979.
H. Theil, "Economics of Information Theoryii, North Holland,
Amsterdam, 1967.
Arbetsmarknadsstyre 1 sen (AMS), II Arbetsmarknadsstati st; k ", (The
National Labor Market Board, lILabor Market Statistics
ll
), various
issues, Stockholm.
5,
Statistiska Centralbyrn (Se8), "Arbetskraftsunderskni
("Swedish Labor Force Surveys", AKU), yearly averages 1975-78.
U.S. Bureau of Labor Statistics (BLS), IIEmployment and Earnings",
various issues, Washington.
28
29
FOOTNOTES
* We are indebted to George Borts, Ned Gramlich, Mats Persson
and an anonymous referee for helpful comments on an earlier
vers; on.
1
The question has earlier beenadd ( 19
Axelsson and Lfgren (1977). Their methods differ from ours.
~ for a praof of (3), see e.g. ppman and Iv1cc,aii (19
\
) .
3
The worker in Sivenls
Seater's models is maximizing his
lifetime utility by using search in the labor market as one
important cholce variable. Siven a1so considers search in the
goods market but assumes leisure to be fixed; maximization of
the uti1ity functional is therefore equivalent to maximization
i earn; , on d, t ~ e s account
of variable leisure but ignores search on the goods market.
4
priceinflation implies in the Siven-model a re-
allocatian of time from search in the labar market to search in
the goods market thereby causing a decline of the job offer
probabiiity. The reservation wage will also increase} reinforc-
ing the effect on unemployrnent duration. The Lucas-Rapping mode 1
is hardly suitable for analyzing the length of spells of unemploy-
ment s it sregards j search and considers unemployment
as pure leisure, resulting as a difference between actual and
normal emp l 9 ) ana Kesselman-Savin (1978) have
run unemployment regressions for the U.S. including un-
anti ce i as an exp l anator'y va e.
results turn out to be unsatisfactory; the coefficients are
as a rule insignificantly different from zero and the signs
are unstable across di ressions.
5 Seater (1978).
6 The crucial trick in Barranis approach - fol1owed by Axe1sson
and Lfgren - is to construct a model which gives an explicit
s fication rel ip the il r es
(V) and offer probability (8). Given such a relationship,
8 = f(V), the acceptance probability is obtained as P =
re is interesti since i can validate a pro-cyclical
reservation wage pattern (i.e. P and Vare inversely correlated) .
.
The approach requires, however, some fairly restrictive assump-
tians regarding the relationship between e and V; Barron assumes
that 8 = k . V, implying that the elasticity equals one,
on
vacancy in each occupation. It can be shown that less restrictive
assumptions produce an elasticity lower than one. Barron's proce-
dure is, moreover, unable to separate the supply effect from the
detection-lag effect. Our approach, on the other hand, can quanti-
fy the detection-lag effect but captures only the net availability
effect.
7
A Box-Jenkins-program cal1ed T-series available at the Stockhoim
30
School of Economics has been used. For identification criteria, see
son 19 ),
31
8
In some regressions we a1so tried average hourly earnings for
l sector. The results were basical-
ly the same.
9
have also tried logit-specifications in same cases, as weIl
as adaptive expectations with shorter lags. The results turned out
to be fairly robust with respect to these changes.
10 Santomero & Seater (1978) p. 525.
11 See articles by Gronau 1 9 7 1 ) ~ Kasper (1967) and Kiefer &
( l
\
),
12 The coef+, ,'c,'ent f / *. " 'f" t" .... (25\ b t th
_ o w W l S S l gn, l can 1 n t.q., ) u e
value indicates that the t-ratio should not be taken s ously.
13 Phelps (1971) pp. 6-7.
14 See Feldstein (1975).
15
1\
(1975) figures imply that 18 percent of those unemployed in
!v1arch 1974 were on temporary 1 ayoff. The correspondi ng fi gure
for March 1978 is 11 percent (Employment and Earnings).
6
See the paper by & Scholl (1975) l10wing
discussion in the Brookings Paper.
APPENDIX
An estimation problem arises because the dependent variable
is an estimate of the Iitrueli transition probability. This
of i ~ ~ i ~ t to samo1ing
variation and this variation obviously enters in the regres-
sion equation as stochastic disturbances. Since this variation
is not constant the assumptions of ordinary least squares are
violated.
Thell (1 ) has rived the following variance of the dis-
turbances for the logit model:
(A. 1 )
where ~ l is the estimate of the transition probability and N
t
the number of observations. By using the same procedure as The;l
(see belo\'J)
(A.2)
The appropriate weights are given by l/Var(E? ).
_t
The derivation of (A.2) proceeds as follows: Consider the basic
relation bebJeen the IItrue
H
transition probability for individual
i at time t and the explanatory variable X
t
A: 1
A:2
(A.3)
Since the explanatory variables have the same values for all in-
dividuals in these applications index i has been omitted.
When the estimate V
t
of the death-risk is inserted inta the equa-
tion (A.3) instead of ~ t the sampling variation necessitates the
inclusion of a distrubance St
(A.4) ln" = ln
3 1 n
Now, the problem is to express the variance of St in terms of
The average of (A.3) is
(A.5)
By subtracting (A.5) from (A.4) we obtain:
(A.6)
- _1_ ~ (ln ]l - ln - I . ~
N
t
1. t l.L
The expression in parentheses can be simplified to:
(A.7)
The last expression can be simplified to
jl -]1.
8)
-"

if and other.
If (A.8) is inserted inta 6) we nave
(A.9)
The variance of Et now becomes:
(A. 10)
If as Theil, disregard var(]Jit) and approximate ']J by
we obtain:
(A.11 )

't
A:3