MIT LIBRARIES
DUPL
3 9080 02618 0353
Digitized by the Internet Archive
in
2011 with funding from
Boston Library Consortium IVIember Libraries
http://www.archive.org/details/effectivelaborreOOcaba
DBAIEY
HB31
.M415
30
Massachusetts Institute of Technology
Department of Economics
Working Paper Series
EFFECTIVE LABOR REGULATION AND
MICROECONOMICS FLEXIBILITY
Ricardo
J.
Caballero
Kevin N. Cowan
Eduardo M.R.A. Engel
Alejandro Micco
Working Paper 04-30
August
Room
1,
2004
E52-251
50 Memorial Drive
Cambridge, MA 02142
This
paper can be downloaded without charge from the
Social Science Research Networl< Paper Collection at
http://ssrn.com/abstract=581021
Effective Labor Regulation
AND MICROECONOMIC FLEXIBILITY
Ricardo
J.
Caballero
Kevin N. Cowan
Eduardo M.R.A. Engel
Alejandro Micco*
August, 2004
Abstract
Microeconomic
nomic growth
is
modem
by
facilitating the process
marlcet economies.
the presence of adjustment costs,
latter is
is
in
flexibility,
rather weak. In this paper
some of them
for
why
is at
we
revisit this
the core of eco-
this process is not infinitely fast,
technological, others institutional. Chief
While few economists would object to such a view,
labor market regulation.
sectoral panel for
of creative-destruction,
The main reason
its
hypothesis and find strong evidence for
60 countries and a methodology suitable for such a panel.
We
among
the
empirical support
it.
We
new
use a
find that job security
regulation clearly hampers the creative-destruction process, especially in countries where regulations are
likely to
be enforced. Moving from the 20th to the 80th percentile in job
strong rule of law, cuts the annual speed of adjustment to shocks
percent from annual productivity growth. The same
by a
movement has
security, in countries
with
third while shaving off about
one
negligible effects in countries with
weak rule of law.
JEL
Codes:
Keywords:
E24, J23, J63, J64, KOO.
Microeconomic
rigidities, creative-destruction,
job security regulation, adjustment costs,
rule of law, productivity growth.
'Respectively:
Bank.
MIT
and
NBER;
Inter-Anierican Development Bank; Yale University and
NBER;
Inter-American Development
We thank Joseph Altonji, John Haltiwanger, Michael Keane, Norman Loayza and participants at the LACEA 2003 conference
for useful
comments. Caballero thanks the
NSF
for financial support.
Introduction
1
Microeconomic
nomic growth
by
flexibility,
modem
in
facilitating the
US
is
core of eco-
at the
market economies. This basic idea has been with economists for centuries, was
brought to the fore by Schumpeter
texts.' In
ongoing process of creative-destruction,
fifty
years ago, and has recently been quantified in a wide variety of con-
Manufacturing, for example, more than half of aggregate productivity growth can be directly
linked to this process.-^
The main obstacle faced by microeconomic
flexibility is
purely technological, others are institutional. Chief
The
ular job security provisions.
literature
among
adjustment costs.
the latter
is
Some of
these costs are
labor market regulation, in partic-
on the impact of labor market regulation on the many diiTerent
economic, political and sociological variables associated
and contentious. However, the proposition
to labor
markets and their participants
is
extensive
that job security provisions reduce restructuring is a point
of
agreement.
Despite
consensus, the empirical evidence supporting the negative impact of labor market regulation
this
on microeconomic
flexibility
has been scant at best. This
cal success are legions, including
variables, both within a coimtry
is
not too surprising, as the obstacles to empiri-
poor measurement of restructuring activity and labor market institutions
and more so across countries. In
this
develop a methodology that allows us to bring together the extensive
by Djankov
tion constructed
and output from the
ofiicial labor
new
we make
a
new
attempt.
We
data set on labor market regula-
(2003) with comparable cross-country cross-sectoral data on employment
(2002) data-set.
We
also
emphasize the key distinction between effective and
market regulation.
The methodology builds on the simple partial-adjustment idea that
larger adjustment costs are reflected
The accumulation of
limited adjustment to these shocks
in slower
employment adjustment
wedge between
builds a
this
UNIDO
et al.
paper
approach.
to shocks.^
frictionless
and actual employment, which
is
the
main
right
hand side variable
in
We propose a new way of estimating this wedge, which allows us to pool data on labor market
legislation with
comparable employment and output data for a broad range of countries. As a
able to enlarge the effective sample to 60 economies,
studies in this literature.''
Our attempt
to
measure
more than double
result,
we
are
the coimtry coverage of previous
effective labor regulation interacts existing
measures of
job security provision with measures of rule of law and government efiiciency.
Our results
to
are clear
and robust: countries with
less effective job security legislation adjust
more quickly
imbalances between frictionless and actual employment. In countries with strong rule of law, moving
fi^om the
and
'
20m to
the 80th percentile of job security lowers the speed of adjustment to shocks
cuts annual productivity grov/th
by 0.86 percent. The same
by 35 percent
movement for countries with low rule of law
See, e.g., the review in Caballero and Haniraour (2000).
^See,
e.g., Foster,
Haltiwanger and Krizan (1998).
^For surveys of the empirical
''To our
countries.
OECD
literature
on partial-adjustment see Nickell (1986) and
Other recent studies, such as
economies.
Hammermesh
(1993).
- Nickell and Nuziata (2000) - included 20 high income OECD
Burgess and Knetter (1998) and Burgess et al. (2000), pool industry-level data from 7
knowledge, the broadest cross-coimtry study
to date
only reduces the speed of adjustment by approximately
The paper proceeds
Section 3 discusses the
results
percent and productivity growth by 0.02 percent.
Section 2 presents the methodology and describes the
as follows.
main
1
and explores
new
data
set.
4 gauges the impact of effective
their robustness. Section
labor protection on productivity growth. Section 5 concludes.
Methodology and Data
2
Methodology
2.1
Overview
2.1.1
The
starting point for
number of
(filled)
our methodology
is
where the change
a simple adjustment hazard model,
jobs in sector j in country c between time
?
—
1
and
t
in the
a probabilistic (at least to the
is
econometrician) function of the gap between desired and actual employment:
Gapjc,=e*^-ejc,,-i,
^ejct^^jctGapjct
where
ejct
variable
and
\|fycr,
e*^,
denote the logarithm of employment and desired employment, respectively. The random
which
has country-specific
is
assumed
mean
i.i.d.
both across sectors and over time, takes values in the interval
Xc and variance C0cXc(l
to the standard quadratic adjustment
Tie
maximum. As
Equation (1) hints
at
— Xc),
model, the case
captures microeconomic flexibility.
flexibility is
(1)
As
cOc
with
=
Xc goes to one,
1
all
<
cOc
<
The case
1.
cOc
=
to the Calvo (1983) model.
[0, 1]
and
corresponds
Tlie parameter
gaps are closed quickly and microeconomic
Xc decreases, microeconomic flexibility declines.
two important components of our methodology:
the
employment gap and a
We
describe both ingredients in detail in what follows.
strategy to estimate the average (over
j and
In a nutshell,
t)
we
We
need to find a measure of
speeds of adjustment (the
Xc).
construct estimates of ej^, the
only unobserved element of the gap, by solving the optimization problem of a sector's representative firm,
as a function of observables such as labor productivity and a suitable proxy for the average market wage.
We
estimate Xc from (1), based
upon
the large cross-sectional size of our sample
heterogeneity ui the realizations of the gaps and the
(1997) for
US
\j/ycr's
and the well docimiented
(see, e.g., Caballero,
Engel and Haltiwanger
evidence).
"
2.1.2
DetaUs
Output and demand for a sector's representative firm are given by:
y
=
p
= d-^y,
a
+ ae + ^h,
-
•
(2)
(3)
'
where >>, p,
e, a, h,
d denote
output, price,
We
the price elasticity of demand.
ri is
employment, productivity, hours worked and demand shocks, and
let
=
y
(ri
-
l)/ri,
with
ri>l,0<a<l,0<P<l
<
and y
1/a.
All variables are in logs.
Firms are competitive
in the labor
market but pay wages that increase with the number of hours worked:
w = k° + \og{H'' + Q.).
This can be approximated by:
w = w° + n{h-h),
with w° and
//
determined by hf and Q, and h constant over time and interpreted below. In order to ensure
interior solutions,
we assume
A key assumption
changes employment
It
(4).
is
a/i
>
P and
>
/i
Py.
that the representative firm within each sector only faces adjustment costs
levels,
not when
employment, by solving the corresponding
In a fi'ictionless labor market the firm's
first
level.
its
current level of
order condition for hours.
employment
level also satisfies a simple static first order condi-
employment. Our fimctional forms then imply that the optimal choice of hours,
on the employment
it
changes the number of hours worked (beyond overtime payments).'^
it
follows that the sector's choice of hours in every period can be expressed in terms of
tion for
when
h,
does not depend
A patient calculation shows that
«
=
1
pQ
/
- lof
'
p.
\a//-Py
We denote the corresponding employment level by e and refer to it as the static employment target:
e
with
C a constant that depends on
//,
The following relation between
= C+-
1-ay
a, P and
the
[d
+ ya-w°\,
y.
employment gap and the hours gap then follows:
e-e = ^^—^{h-h).
1-ay
This
is
the expression used
by Caballero and Engel (1993).
(5)
It_ia
not useful in our case, since
we do
not have information on worked hours. Yet the argument used to derive (5) also can be used to express the
employment gap
in terms of the marginal labor productivity gap:
e-e =[y-w
1-ay
where v denotes marginal
For evidence on
this
productivity,
if
= fi/{^— Py)
see Sargent (1978) and Shapiro (1986).
is
),
decreasing in the elasticity of the marginal
wage
,
schedule with respect to average hours worked, //—I, and wP was defined in
term on the right hand side of ( 1 ) However,
tlie
.
walk (possibly with an exogenously time varying
have
that e*^, is equal to ejct plus a constant 5c(.
^)ct
where we have allowed
- sjci-i =
It
we assume
that
—an assumption
drift)
that e
o?
+ ya — w"
—
e
is
the
— ej^^
e*-^
random
follows a
consistent with the data^
— we
follows that
-T—zrr
{"^jct
for sector-specific differences in
- w'jct ) + Aeyrf + 5rt
y.
nominal terms. However, since these expressions are
in
if
Note
dynamic employment gap
difference between the static target e and realized employment, not the
related to
(4).
Note
that
(6)
both marginal product and wages are
in logs, their difference eliminates the aggregate
price level component.
We
estimate the marginal productivity of labor, vja, using output per worker multiplied
level labor share,
Two
industry-
assumed constant whithin income groups and over time.
natural candidates to proxy for
w°j^i
are the average (across sectors within a coimtry, at a given
point in time) of either observed wages or observed marginal productivities.
competitive labor market,
the latter
may
The former
is
consistent with a
be expected to be more robust in settings with long-term contracts
and multiple forms of compensation, where the salary
We
by an
may not represent the
actual marginal cost of labor.
''
performed estimations using both alternatives and found no discernible differences (see below). This
suggests that statistical power comes mainly from the cross-section dimension, that
magnitude of sector-specific shocks. In what follows
we
and approximate w° by the average marginal productivity, which leads
to:
mented and
large
s*ct-^Jci-\
where
=
{vjct
_
l-ay/
.
v.^ denotes the average, over j,
- v.ci + ^ejc, + del =
of Vj^, and
)
we
use
this
report the
Gapj^,
is,
fi^om the well docu-
more robust
alternative
+ 5ct,
(7)
convention for other variables as well. The
expression above ignores systematic variations in labor productivity across sectors within a country, for
may
example, because (imobserved) labor quality
differ systematically across sectors.
The presence of
such heterogeneity would tend to bias estimates of the speed of adjustment downward. To incorporate
possibiUty
we
subtract
from
[vjci
— v.d)
in (7) a
moving average of relative
Qjct^2^{vjcl-l~V.c,-\)
As
a robustness check, for our main specifications
moving average, without
^Pooling
Computing
'while
changes in our
correlation by country the
mean
value
is
also
computed
Qjct
results
(more on
this
using a three and four periods
when we check
robustness in
have assumed a simple competitive mai'ket for the base salary (salary for normal hours) within each
Xc).
is
—0.018.
0.011 with a standard deviation of 0.179.
procedure could easily accommodate other, more rent-sharing
some parameters, but not
where
{Vjc,-2-V.c,-2)]-
countries and sectors together, the first order autocorrelation of the measure of Ae^^., constructed below
all
tliis
we
significant
we
+
sectoral productivity, Qja,
this
like,
wage
setting
mechanisms (with
sector,
our
a suitable reinterpretation
of
.
Section 3.2). The resulting expression for the estimated employment-gap
e*jct-ejct-\
where ayy
is
=
T—— (vycr-0ycr-v.cr)+Aeyc?+5cr
is:
=
Gap^.„-|-5«,
constructed using the sample median of the labor share for sector
j
(8)
across year and
income
groups.
Rearranging
we
(8),
estimate
(j)
from
= ~ _ ^„ i^'^jct - Av-a) + Kg, + X)u + Ae*ct =
1 - aYy
^ejct
-(t^^ycr
+Kcr +e;cr,
(9)
1
where k
{Avjct
is
a country-year
dummy,
— Av.c,)/(1 - ajj). We
two consecutive
we
assume
is
the change in the desired level of
respectively.
full
estimate (9) using {Awjd-i —Aw.ct-i) as an instrument for (Avjd
we
and across income groups. The
Table
first
1
two
The remaining columns report
the estimation results for each of our three
set the value
of
(j)
at its full
sample estimate of 0.4 for
2.2).
all
Based on our
income groups and
results for the baseline
countries in our sample.
important to point out that our methodology has some advantages over standard partial adjustment
It is
.
— Av.^).^
sample, with and without two percent of extreme values for the independent variable,
job security groups (more on both of these measures in Section
case,
=
zjct
changes in sectoral labor composition are negligible between
that
reports the estimation results of (9) for the full sample of countries
columns use the
employment and
In order to avoid the simultaneity bias present in this equation (Av and Ae* are
years.
clearly correlated)
Ae^^,
estimations. First,
it
summarizes in a single variable
all
shocks faced by a sector. This feature allows us to
increase precision and to study the determinants of the speed of adjustment using interaction terms. Second,
and
related,
variables in
it
only requires data on nominal output and employment, two standard and well-measured
most
industrial surveys.
Most previous
studies
on adjustment
costs required measures of real
output or an exogenous measure of sector demand.^
2.1.3
Regressions
The central empirical question of the present study is how cross-country differences in job security regulation
affect the
speed of adjustment. Accordingly, from
(1)
and
(8)
it
follows that the basic equation
we
estimate
is:
Aejct
*We
lag the instniment to deal with the simultaneity
impact of measurement error
=
XctiG&Vj^t
+ ^ct),
'
problem and use the wage rather than productivity
(10)
to
reduce the (potential)
bias.
'Abraham and Houseman (1994), Harmnermesh (1993), and Nickel and Nunziata (2000)) evaluate the differential response of
employment to observed real output. A second option is to construct exogenous demand shocks. Although this approach overcomes
the real output concerns,
it
requii'es constructing
an adequate sectorial demand shock for every country.
papers by Burgess and Knetter (1998) and Burgess
et al. (2000),
which use the
real
exchange
A
rate as their
case in point
demand
ai'e
shock.
the
The
estimated effects of the real exchange rate on employment are usually marginally significant, and often of the opposite sign than
expected.
where Aey^
is
change in employment and
the log
Xct
We
denotes the speed of adjustment.
assume
that the
latter takes the form:
Xa
where JS^f
is
a
= Xy+X2^sf,
(11)
measure of effective ioh security regulation. In practice we observe job security regulation
(imperfectly), but not the rigor with
which
it is
We proxy the latter with a "rule
enforced.
of law" variable,
so that
JSf =
where a
in (1 1)
a constant and RLcr
is
A,3
= dkz
and
A,i
Gap^-^,
(12)
a standard measure of rule of law (see below). Substituting this expression
and the resulting expression for
Aeyrt =^
with
is
JS,,(l+flRL,,),
+ A,2
'kct
in (10), yields our
(Gap^^,
x
JSc/)
+ A,3
main estimating
(Gap^-^,
The main coefficients of interest
are
1.2
+ 5^ + e;cr,
x JS^ x RLc/)
denotes country xtime fixed effects (proportional to the
hct
equation:
(13)
defined above).
%ct
and A,3 which measure how the speed of adjustment varies across
,
countries depending on their labor market regulation (both dejure and de facto).
The Data
2.2
This section describes our sample and main variables. Additional variables are defined as
we
introduce
them
later in the text.
2.2.1
Job Security and Rule of Law
We use two measures ofjob security, or legal protection against dismissal:
by Djankov
et al.
(2003) for 60 countries world-wide (henceforth JSc) and the job security index constructed
by Heckman and Pages (2000)
measure
is
available for a larger
The HPrf measure has
Our main job
for
24 countries
OECD
and Latin America (henceforth HP^). The JSc
sample of countries and includes a broader range of job
by
The
rules
either party at
narrow
list
secxnrity variables.
the advantage of having time variation.
security index, JSc,
on values between
in
and
1:
(r)
is
the
sum of four
variables,
measured
groimds for dismissal protection PGc,
procedures PPc, {Hi) notice and severance payments PSc, and
tion PCc.
the job security index constructed
{iv)
{ii)
in 1997,
protection regarding dismissal
protection of employment in the constitu-
on grounds of dismissal range firom allowing the employment
any time (employment
at will) to
each of which takes
relation to
be teltninated
allowing the termination of contracts only under a very
of "fair" causes. Protective dismissal procedures require employers to obtain the authorization
of third parties (such as unions and judges) before terminating the employment contract. The third variable, notice
and severance payment,
is tlie
one closest
two components: mandatory severance payments
advance notice for dismissals
after
after
to the
HPcr measure, and
20 years of employment
20 years of employment {NStc
=
ba+iQ
is
(in
the normalized
sum of
months) and months of
+ SPct+io,
t
=
1997).
The four
components of JSc described above increase with the
The Heckman and Pages measure
on the
is
To quantify
costs of dismissal.
Our
and
firing costs,
and
is
of job
security.
narrower, including only those provisions that have a direct impact
the effects of this legislation, they construct an index that
The index includes both
the expected (at hiring) cost of a future dismissal.
legislation
level
measured
in units
We
RLc) and more
institutional variables as well as the countries in
efficient
statistics for
categories).'^
As
significant. Differences
by including
this
(1999), and interact
them
governments (higher GEc).
our sample and their corresponding income group
are reported in Table 2. Table 3 reports the sample correlations
and summary
at al.
do
expect labor market legislation to have a larger impact on adjustment costs in
countries with a stronger rule of law (higher
income
We
estimations also adjust for the level of enforcement of labor legislation.
with JSc and HPcr.'"
computes
advanced notice
of monthly wages.
measures of rule of law RLc and government efficiency GEc from Kaufrnann
The
the costs of
between our main cross-country variables
each of these measures for three income groups (based on World Bank per capita
expected, the correlation between the two measures of job security
can be explained mainly by the broader scope of the
positive and
is
Also as expected,
JSc; index.
rule of law and government efSciency increase with income levels. Note, however, that neither measure of
job security
income
is
income per
both JSa and HPc are highest for middle
countries.
employment and wage data come from the 2002
Our
output,
The
UNIDO
ISIC code (revision
2).
in sector
Statistics
Database.
Because our measures of job security and rule of law are time
measured in recent years, however, we
output and labor compensation are in current
regressions).
UNIDO Industrial
3-digit
database contains data for the period 1963-2000 for the 28 manufacturing sectors that corre-
to the 3 digit
invariant and
US
restrict
our sample to the period 1980-2000. Data on
dollars (inflation
Throughout the paper our main dependent variable
j of country c in period
A large number of countries
by
capita, since
Industrial Statistics
2.2.2
spond
positively correlated with
is
is
removed through time
Aeja, the log change in
effects in our
total
employment
t.
are included in the original dataset
— however ova sample
is
constrained
the cross-country availability of the independent variables measuring job security. In addition,
two percent of extreme employment changes
in
we
drop
each of the three income groups. For our main specification
the resulting sample includes 60 economies. Table 3
shows descriptive
statistics for the
dependent variable
by income group.
^"For rule of law and government eJBHciency
since this
'
and
is
closest to the
'income groups
Low
Income.
are:
Djankov
et al.
we
use the
earliest
(2003) measure, which
is
value available in
tlie
Kaufinann
et
al.
(1999) database: 1996,
for 1997.
l=High Income OECD, 2=High Income Non
OECD
and Upper Middle Income, 3=Lower Middle Income
Results
3
This section presents our main result, showing that effective job security has a significant negative effect on
the speed of adjustment of employment to shocks in the employment-gap.
It
also presents several robustness
exercises.
Main
3.1
results
Recall that our main estimating equation
Aejc,
Note
= XiGapj^, + X2{Ga.pj^xJSc)+h{Gzpj^,xJScX'RLc)+?ict+ejct-
we have dropped
that
rule of
law and job security
[Gapi^i
X
We
x RLc)
JScr
time subscripts from JSc and
in
RL^
as
we
we
also include the respective Gap,-^,
all
specifications that include the
x RLc
as a control variable.
of job security on the speed of adjustment, and set X2 and
effect
to zero.
This gives us an estimate of the average speed of adjustment and
Table
On
4.
(14)
only use time invariant measures of
our baseline estimation. Note also that in
interaction
by ignoring the
start
is:
average (across countries and periods)
we
find that
60%
reported in
is
equal
^,3
column
of the employment-gap
is
of
1
closed in
each period. Furthermore, our measure of the employment-gap and coimtryxyear fixed effects explain
60%
of the variance in log-employment growth.
The next
different Xi
three
by
columns present our main
sectors
and country income
which
results,
level. '-^
are repeated in
Column
columns 5
to 7 allowing for
2 (and 5) presents our estimate of ^2- This
coefficient has the right sign and is significant at conventional confidence levels.
Employment
adjusts
more
slowly to shocks in the employment-gap in countries with higher levels of ofiicial job security.
we
Next,
columns
3
allow for a distinction between effective and
and 4 (and, correspondingly, 6 and
official
job security.
Results are reported in
7) for different rules-enforcement criteria. In
6 the distinction between effective and ofiicial job security
is
columns
captured by the product of JSc and
3
and
DSRLc,
—
dummy variable for countries with strong rule of law (RLc > RLcreece where Greece is
the OECD country with the lowest RL score). The three panels in Figure
show the value of the job security
where DSRLc
is
a
1
index for countries in the high,
while
larger
A,3
has the right sign and
(downward)
medium and low income
is
highly significant. That
on the speed of adjustment
effect
measured by our rule-of-law dummy. The
In countries with strong rule of law,
percentile (0.23) reduces
effect, 0.006,
'-We allow
X by
0.22.
effect
moving
include an additional interaction between
same change
in JSc will
in countries with stricter
have a significantly
enforcement of laws, as
3 is large.
3 digit
three
in job security legislation has a considerable smaller
group of economies with weak rule of law. Employment
ISIC sector dummies (we also include sector fixed
by omitted
Gap^, and
the
firom the 20th percentile of job security (—0.19) to the 80th
in the
between Gap,„ and
control for the possibility that our results are driven
is,
Now X2 becomes insignificant,
of the estimated coefiicients reported in column
The same change
on the speed of adjustment
for an interaction
groups, respectively.
variables, correlated with our measures
income-group dummies.
of job
effects).
security.
For
We
also
this,
we
adjusts
more slowly to shocks
in the
employment-gap
Columns 4 and 7 address whether the negative
enforcement. To do so
effectiveness
we use
in countries with higher levels of effective job security.
on
coefficient
A,3
robust to other measures of legal
is
an alternative variable from the Kaufmann
(GE) - and construct a
dummy
et al.
(1999) dataset - government
variable for high effectiveness countries
Clearly, the results are very close to those reported in
columns
3
and
on the estimated speed of adjustment when governments
significant negative effect
(GEc >GEGreece)-
Job security legislation has a
7.
are effective
- a proxy
enforcement of existing labor regulation.
for
Finally, the last
column
in Table
4 uses an altemative measure ofjob security.
We repeat our specification
from column 7 (including sector and income dummies) using the Heckman-Pages (2000) measure of job
security.
size is
The
HP a
reduced by
data are only available for countries in the
half,
OECD
and Latin America so our sample
and most low income countries are dropped. The
flip side is that this
measure
time varying which potentially allows us to capture the effects of changes in the job security regulation.
reported in
column
8,
we
effect
As
ofHPc on the speed of adjustment.
Further robustness
3.2
We continue our robustness
tive
and significant
find a negative
is
by
exploration
assessing the impact of three broad econometric issues: alterna-
gap-measures, exclusion of potential (country) outliers, and misspecification due to endogeneity of the
gap measure.
3.2.1
Alternative gap-measures
Table 4 suggests that conditional on our measure of the employment-gap, our main findings are robust: job
security,
when
enforced, has a significant negative impact on the speed of adjustment to the employment-gap.
Table 5 tests the robustness of this result to altemative measures of the employment-gap. Colunms
relax the assumption of a
—
2 and 3 in Table 4
3
and 4 report the
({)
common
across
using the values of
results
(j)
all
countries.
They repeat our baseline
specifications
estimated per income-group reported in Table
of using values of
(f)
1.
1
and 2
—columns
In turn, columns
estimated across countries grouped by level of job security.'^
Next, columns 5 through 8 repeat our baseline specifications using a three and four period moving average
to estimate Qjct.
The
final
two columns
wages instead of average productivity
in Table
3.2.2
5,
(9
and 10) use an altemative specification for
(see equation 8) to build Gapj^,. In
our results remain qualitatively the same as in Table
all
w^^,,
based on average
of the specifications reported
4.
Exclusion of potential (country) outliers
Table 6 reports estimates of ^2 and
^,3
country from our sample at a time. In
using the specification from column 3 in Table 4 but dropping one
all
cases the estimated coefficient on I3
conventional confidence intervals.
'Countries are grouped into the upper, middle and lower thirds of job security.
is
negative and significant at
Boweva; it is
di^atajce in
Tsg)
also
iiie
eidier Hong
sppsKsA in tins table that exclndiiig
point ^limafes. For 4is reason,
BDW sscia&Bg these two
connjn^. In
we re-estimate
this case the vahie
howers: ihe ^^^^ resrslts lemain xmchanged. Table 7 rqiorts
Potential endogeneitT of the gap
3.2-3
Kong or Kenj' a makes
a snbstantial
our model from scratch (that
is,
from o
of 6 rises from 0.40 to 0.42. Qualitatively,
these results.
measure
One concsn with our procednre is that lie
construction of the gap measure includes the change in emplo}'-
rassA. While this does not rq>reseiii a problem trnda" the nuU hj'pothesis of the model,
euQi in sa^?loyms3Xt and ^ji could introduce important biases.
'
anj'
measurement
We address this issue wiA two procedures.
IB^xe descaibing these procedures, we note that the standard solution of passing the Ae-component of
&e gap
dsSasA in
Passing
As
(8) to tiie left
hand side of the estimating equation (10) does not work
to tiie left suggests that the coefficient
diown in Appss&Sax A,
Secfi(Hi2.1).
this
on the
turn to the
fra:
two procedures. Hie
first
fbs contraiqKHaneous gap measure.
valid insfrnmait if it
is
(co
=
A,/(l
— X.). As
in the notation of
productivity, 6/a.
Given
tiiat
Gap vj
data play flie role of p^odtlcti^'ily data
The second procedure
when
re-writes the
AGapja
=
Table 8 reports the values of the average
procedures. For comparison purposes,
for
tiie
this error is serially correlated. In
7.
flie first
and hence
we
It is
estimate of relative sectoral
eiflier
wage
hand side of (8).
side:'^
(l-7:r:)AGapj^_i-^ejct.
(15)
estimated witii these two alternative, and
row reproduces the
first
column
much less
as instruments. Finally,
Row 3
apparent from the table that the estimates of average
the (modified; gap or the change in
employment
is
is zero.
precise,
The second row
in Table 4.
conclude that the bias due to a potentially endogenous gap
the Cah.'i>-C2se, for evsrj' observation
our speci-
in a standard dynamic panel formulation that removes die
IV procedure based on using lagged wages
the estimate from the dynamic panel
right ballpark,
6zyc^_i. Uniortunatety, the latter is not a
calculating the v and 9 terms on the right
model
in-
ozp -^tseja can be rewritten as
construct an alternative measure of the ex-post gap letting
contempotaaeavs employment change from the right hand
tiie result
—
we use a moving average to construct the
To avoid this problem, we
model (small
approach.
procedure maintain'; our baseline specification, but
congrated with measurement error and
fication this could be tiie case because
-''in
be equal to
departures from a partial adjustment
when estimating ?. using this
^j^-\ -T-t^j^, a nataiai. instrumait is the lag of the ex-post gap,
shows
-will
By contrast, in the case of a Calvo-tj'pe a(^tistment (a = 1), the corresponding coefficient will,
on avaage, be negative}^ More important, even small
stnnxsnts
gap
holds only in the case of a partial adjustment model
ralnes of o) introduce significant biases
Next we
resulting
in our context
7^
reports
are in the
not significant.
The former happens
wheo adjtislmait laies place, die latter whai it does not It follows that the covaiiance of A^ and the (modified) gap will be eqtial to
minus tiae product of the mean of bofli variables. Since fliese means have the same sign^the estimated coefficient wiU be negative.
See Appaidis A fer a fbnnal
derivation.
''To estimate ftis equation
we follow Anderson
and Hsiao
(1 982)
and use twice and toee-times lagged values of AGap ^ as
instrtmientsforfteRHS-i^riable. Similar results are obtained ifwe follow Arellano and
10
Bond
(1991).
Gauging the Costs of Effective Labor Protection
4
By
impairing worker movements from less to more productive units, effective labor protection reduces ag-
gregate output and slows
this effect.
Any
down economic
growth. In this section
we develop
a simple
such exercise requires strong assumptions and our approach
is
framework to quantify
no exception. Nonetheless,
our findings suggest that the costs of the microeconomic inflexibility caused by effective protection
moving from
In countries with strong rule of law,
the 20th to the 80th percentile of job security lowers an-
nual productivity growth somewhere between 0.9 and 1.2 percent The
low rule of law has a negligible impact on
is large.
same movement
for countries with
TFR
Consider a continuum of establishments, indexed by
that adjust labor in response to productivity
i,
shocks, while their share of the economy's capital remains fixed over time.
Their production fiinctions
exhibit constant returns to (aggregate) capital, K,, and decreasing returns to labor:
= BuKrLl
(16)
<a<
that
Y,
where Bu denotes
and
plant-level productivity
can be decomposed into the product of a
common and
Alog^rt
where the
V;
are
i.i.d.
^{/ja o^ ) and the
,
the aggregate shocks) iA^(0, oj).
Vj, 's
are
=
i.i.d.
price-elasticity
bit
is ri
>
1
.
The Bu 's follow geometric random walks,
=v,
+ v[j,
(across productive units, over time and with respect to
We set /ja = 0, since we
of demand
.
an idiosyncratic component:
and idiosyncratic shocks, not in Jensen-inequality-type
The
1
are interested in the interaction
effects associated
Aggregate labor
is
between
rigidities
with aggregate shocks.
assumed constant and
set equal to one.
We
define aggregate productivity, ^, as:
,
A,
so that aggregate output,
Y,
= J Yi,di,
=
do that period.'^
The parameter
(17)
satisfies
Y,
Units adjust with probability
JBuL^di,
=
A,K,.
in every period, independent of their history
A.^
that captures
microeconomic
flexibility is Xc-
and of what other units
Higher values of
^e
"kc
associated with a faster reallocation of workers in response to productivity shocks.
Standard calculations show that the growth rate of output, gy,
gY
'''More precisely, whether unit
;
=
sA-?>,
units
(18)
time t is determined by a Bernoulli random variable ^;, with probability of success
and over time. This corresponds to the case CO = 1 in Section 2.1.
adjtists at
where the ^a's are independent across
satisfies:
II
Xc,
where s denotes the savings
Now
rate
(assumed exogenous) and 8 the depreciation
compare two economies
that differ only in their degree
Tedious but straightforward calculations relegated to Appendix
gY,2-gY,\
-
B show
1
rate for capital.
of microeconomic
flexibility, Xc,i
<
^c,2-
that:
1
e,
(gr,i+5)
(19)
with
^ 07(2-07)
,
2(1-07)2°'
7=(r|-l)/n,a2 = aJ + a2.i^
We
choose parameters to apply (19) as follows: The mark-up
is set at
average rate of growth per worker in our sample for the 1980-1990 period,
8
= 5/6), gyj to the
0.7%, a = 27%,'^ a = 2/3, and
20%
(so that 7
= 6%.
Table 9 reports estimated speeds of adjustment across countries. Table 10 reports the average aimual
productivity costs of the deviation between each quintile and the bottom quintile in estimated speed of
These numbers are
adjustinent.
large.
More
important, they imply that
moving from
the 20th to the 80th
percentile in job security, in countries with strong rule of law, reduces annual productivity growth
The same change
in
job security legislation has a
much
smaller effect on
TFP
by 0.86%.
growth, 0.02%, in the group
of economies with weak rule of law.
We are fully aware of the many caveats that such ceteris-paribus
the table
is
to
numbers
are rouglily consistent with
of productivity growth on the estimated ^c's for
Xc
is
1 1
and Figure
2).
Focusing on the
around 0.05 and significant
of Table 4 implies that
falls
by
as
raise,
but the point of
provide an alternative metric of the potential significance of observed levels of effective labor
protection. Moreover, these
Table
comparison can
much
as
1
.
8%
the economies in our sample from 1980 to
results controlling for
at the 1 percent level.
when moving from
1
all
what we obtain when miming a regression
income groups (column
3), the coefficient
Combining these estimates with those
the 20th to the 80th percentile in job security, annual
in countries with
1990 (see
in
on
Column
3
TFP growth
high rule of law. In contrast, a similar improvement in labor
regulation in countries with low rule law reduces
TFP growth by only 0.03%.
Concluding Remarks
5
Many
papers have shown that, in theory, job security regulation depresses firm level hiring and firing de-
cisions.
17
Job security provisions increase the cost of reducing employment and therefore lead to fewer dis-
There also
is
a
(static)
jump
in the level of aggregate productivity
when X
increases, given by:
A2-A1
«l
See Appendix
B
for the
proof
'^'This is Are average across the five countries
considered in Caballero
12
et al.
(2004).
missals
when
fiiTns are
faced with negative shocks.
Conversely,
when
optimal employment response takes into account the fact that workers
the
employment response
smaller.
is
The
overall effect
However, conclusive empirical evidence on the
is
faced with a positive shock, the
may have to be fired
in the future,
and
a reduction of the speed of adjustment to shocks.
effects
of job security regulation has been elusive. One
important reason for this deficit has been the lack of information on employment regulation for a sufficiently
large
paper
cal
number of economies
we have
that
can be integrated
to cross sectional data
on employment outcomes. In
developed a simple empirical methodology that has allowed us to
gap by exploiting:
(a) the
recent publication of two cross-country surveys on
(Heckman and Pages (2000) and Djankov
production available in the
fill
UNIDO
et al.
dataset.
(2003)) and, (b) the
homogeneous
some of the
this
empiri-
employment regulations
data on
employment and
Another important reason for the lack of empirical success
differences in the degree of regulation enforcement across countries.
is
We address this problem by interacting
the measures of employment regulation with different proxies for law-enforcement.
Using a dynamic labor demand specification we estimate the
60 countries for the period fi-om 1980
to 1998.
We
effects of job security across a
consistently find a relatively lower speed of adjustment
of employment in countries with high legal protection against dismissal, especially
likely to
sample of
be enforced.
13
when such protection
is
.
References
[1]
Abraham, K. and
ity:
"Does Employment Protection
In:
R.M. Blank,
Inhibit
Labor Market Flexibil-
editor. Protection Versus
Economic
There A Tradeoff?. Chicago: University of Chicago Press.
Journal of Econometrics, 18, 67—82.
M. and
Arellano,
S.R.
and an Application
[4]
(1994).
Anderson, T.W. and C. Hsiao (1982). "Formulation and Estimation of Dynamic Models Using Panel
Data,''
[3]
Houseman
Lessons From Germany, France and Belgium."
Flexibility: Is
[2]
S.
to
Bond
(1991).
"Some
Specification Tests for Panel Data: Montecarlo Evidence
Employment Equations," Review of Economic
Studies, 58, 277-298.
Barro, R.J. and J.W. Lee (1996). "International Measures of Schooling Years and Schooling Quality,"
American Economic Review, Papers and Proceedings, 86, 218-223.
[5]
Burgess.
and M. Knetter (1998). "An International Comparison of Eployment Adjustment
S.
to
Ex-
changeRateFhictuations." Review of International Economics. 6(1): 151-163.
[6]
Burgess,
S.,
M. Knetter and C. Michelacci (2000). "Employment and Output Adjustment in the OECD:
A Disaggregated Analysis of the Role of Job
[7]
Security Provisions."
Economica
67: 419-435.
Caballero R. and E. Engel (1993). "Microeconomic Adjustment Hazards and Aggregate Dynamics."
Quarterly Journal of Economics. 108(2): 359-83.
[8]
Caballero R., E. Engel, and
J.
Hahiwanger (1997), "Aggregate Employment Dynamics: Building
Microeconomic Evidence," American Economic Review
[9]
87(1),
1 1
fi-om
5-137.
Caballero R., E. Engel, and A. Micco (2004), "Microeconomic Flexibility in Latin America,"
NBER
Working Paper No. 10398.
[10]
Caballero, R. and
M. Hammour
(2000). "Creative Destruction and Development: Institutions, Crises,
and Restructuring," Annual World Bank Conference on Development Economics 2000, 213 -24 1
[1 1]
Calvo,
G
(1983). "Staggered Prices in a Utility
Maximizing Framework." Journal of Monetary Eco-
nomics. 12: 383-98.
[12] Djankov, S., R.
Labor."
La
Porta,
F
NBER Working Paper No.
[13] Foster L.,
J.
Heckman,
J.
J.
Botero (2003). "The Regulation of
9756.
Haltiwanger and C.J. Krizan (1998). "Aggregate Productivity Growth: Lessons from
Microeconomic Evidence,"
[14]
Lopez-de-Silanes, A. Shleifer and
NBER Working Papers No.
6803.
and C. Pages (2000). "The Cost of Job Security Regulation: Evidence fi-om Latin Ameri-
can Labor Markets." Inter-American Development Bank Working Paper 430.
14
[15]
Hamermesh, D.
(1993).
[16] Kaufinann, D., A.
Labor Demand, Princeton: Princeton University Press.
Kraay and
P.
Zoido-Lobaton (1999). "Governance Matters". World Bank Policy
Research Department Working Paper No. 2196.
[17]
Mankiw, G.N. (1995). "The Grov/th of Nations," Brookings Papers on Economic Activity, 275-326.
[18] Nickell, S. (1986).
"Dynamic Models of Labour Demand",
Handbook ofLabor Economics,
[19] Nickell, S.
in
O. Ashenfelter and R. Layard (eds.)
Elsevier.
and L. Nunziata (2000). "Employment Patterns in
OECD Countries." Center for Economic
Performance Dscussion Paper 448.
[20]
Sargent, T. (1978). "Estimation of Dynamic Labor
Demand under Rational
Expectations,"
J.
ofPoliti-
cal Economy, 86, 1009-44.
[21]
[22]
Shapiro,
M.D.
(1986).
"The Dynamic Demand
for Capital
and Labor," Quarterly Journal of Eco-
5X3-42.
nomics,
\{i\,
UNIDO
(2002). Industrial Statistics
Database 2002
15
(3-digit level
oflSIC code (Revision
2)).
.
Appendices
A
Endogeneity of the
The model
Gap Measure
the one described in Section 2.1. Ignoring country differences and fixing t}^
is
Ae;
=
where xj
<
CO
<
1
and the
\j/
are
Aej
=
with Uj
{\\ij
- X)xj
Removing Ae ixom
are
=
and
when
1
,
same
\|/
=
1
sign,
measure
the gap
and Aej
we have that when
will be equal to
the
(20)
mean X and
variance (i>k{l
— X),
= hcj + Uj,
=
we
is
when
obtain an imbiased and consistent estimator for
equivalent to replacing xy
\\fj
= 0.
= xj — Aej
by zj
Since the two values
estimating the regression coefficient via
minus the product of the average of Aey and
it
(21)
satisfying the properties of an error term in a standard regression setting. Thus, if we
observe the Aej and xj and estimate (21),
that zj
= \|/yXy,
independent of the x„ with
i.i.d.,
that
follows fi-om (20) that:
It
.
— ej,,_i
el
we have
\\fj
in (21). It
is
obvious
takes in the Calvo case (co
OLS the
average of the
tlie
X?^
follows that the regression coefficient will be negative (or zero
=
1)
covariance in the numerator
Zj.
Since both averages have
if
both averages are equal to
zero).
we
Since
(0=1 may seem somewhat
are considering sectoral data, the case
The following
extreme.
proposition shows that even for small departures from the partial adjustment case, the bias
is
lOcely to
be
significant.
Proposition
1
Consider the setting described above. Denote by
Aej
Denote by
ju
and d^
the (theoretical)
,•
Proof From Acj
= \\ijXj and zj =
Cov(Ae,z)
{\
(3
the
OLS estimate of the cofficient
'The
latter is justified
spending
to the
is
by the
mean and variance of the xj
's.
Then:
(l-co)o2-co;?
s
— ^j)xj
it
.
.^^^
follows that:
= - Y^i{l - \]/i)xf -
fact that
the essence
gap measure defined
most of our
in
= const + ^zj + error
(
- Xy,x,-
f
- X(1 - V;)^/
J
^"This,
This, in a nutshell,
[3
identification
comes from
th estimation procedure described
of the
in (8).
16
>
J
cross-sectional variation
in detail in
Section 2 of
tlie
maio
te.xt,
with xj corre-
and
-X(i-v,)^/
Taking expectations over the
\|/,,
conditional on the
A
The
result
now
^
E[y(l-v)](G^
follows that plim^^^^p
CO
=
1. It
A''
tend to infinity leads
to:
+ /?)-^l-l);/
= (l-co)Ml-X),
=
E[(l-v|/)2]
when
letting
follows from the expression above and the fact that:
E[\(/(1-V)]
It
and
Xi,
is
decreasing in
also follows that plim;^_^„P
[l-(l-co)^](l-^).
varying from X/{1
co,
decreasing in
is
— X) when co =
to
-l/?/(a^
+ Xf?)
so that:
(l-to)A,
<
pliinAr_,„oP
\ju\,
I
l-(l-co)A,'
B
Gauging the Costs
In this appendix
we
tions in Section
4
derive
we
(1
9).
From
( 1
8)
and
9)
(1
it
follows that
it
suffices to
show that under the assump'
have:
I
1_
e,
where
we have
dropped the subindex c from the X and
e
ay(2
=
- ay)
(o?
2(l-aY)2
with
7=
The
large
(ri
-
+ o3),
intuition is easier if we consider the following, equivalent, problem.
and fixed number of firms (no entry or
t
exit).
is Pi^t
Production by firm
— Y-^—
Alog^,V
That
is,
we
ignore hours in
tlie
/n
'
,
17
during period
consists
t is Yj^i
of a very
= Ai^tLf,^^
where Ai^t denotes productivity shocks, assumed
=^a^ = vf+Vi„
production ftmction.
i
The economy
'
I
follow a geometric random walk, so that
^'
(24)
1)/ti.
while (inverse) demand for good / in period
to
(23)
A,2
A,]
with vf
i.i.d.
^(0,a3) and
vf_,
^(0,a^). Hence
i.i.d.
Aa,,,
follows a 9l{Q,(5\), with
a^
=
o^
+ ay. We
assume the wage remains constant throughout.
In
what follows lower case
letters
denote the logarithm of upper case variables. Similarly, *-vaiiabIes
denote the frictionless counterpart of the non-starred variable.
Solving the firm's maximization problem in the absence of adjustment costs leads
to:
^^h
J_
= 1-^,^^'."
-ay"
(25)
^^tr
=
(26)
1
and hence
Denote by Y* aggregate production in period
1=1/(1- ay), Taking expectations
(over
noting that both terms being multiplied on the
7;
Averaging over
all
were no
if there
t
frictions. It
then follows from (26) that:
= e'^''%_,,
Y*,
with
r^^«'>
/
(27)
for a particular realization
r.h.s. are,
of vj^) on both sides of (27) and
by assumption, independent (random walk),
= e^^+2^"f^/7;_,,
yields
(28)
possible realizations of v^ (these fluctuations are not the ones
we
are interested in for the
calculation at hand) leads to
F;
and therefore for
A:
=
= e2^
^r7;_j,
1,2,3,...:
Y*
^ g2'"
^ty;_^,
(29)
Denote:
• Y,^,-k: aggregate
7
that
corresponding to period
•
From
Yi^t,t-k'-
would
t
— k.
This
is
t
the average
if firms
it
follows
that:
had the
Y for units
the corresponding level of production of firm
the expressions derived above
and therefore
attain in period
/
in
r.
frictionless optimal levels
that last adjusted k periods ago.
of labor
.
Taking expectations (with respect to idiosyncratic and aggregate shocks) on both sides of the
sion (here
we
use that
Aa,_, is
latter expres-
independent of }^*_,) yields
which combined with (29) leads
to:
A derivation similar to the one above, leads to:
Y.
,
J-i,t,l-k
which combined with (29)
—
—
pAfl,v+Aa,v-l+--+Aa;,r-*:+iy*
*=
-'/-i!
gives:
7,,,_i
= e-^r;,
(30)
with e deirned in (24).
Assuming Calvo-type adjustment with
probability
"k,
we decompose
aggregate production into the
sum
of the contributions of cohorts:
Y,
= XY; + 1(1 - X)F,,,_i +%{\- XfYt^,-! +
.
.
Substituting (30) in the expression above yields:
"
^'
- ^^T-^TTZe^A
- (l~?L)e-
(31)
1
It
follows that the production gap, defined as:
Prod.
is
Gap
Y* — Y
= ^—-^,
equal to:
A first-order Taylor expansion then shows that, when
Prod.
Subtracting this gap evaluated at
corresponds to (A2
—A\)/A\
A,]
in the
from
main
its
Gap
|9|
<<
~ ^^~
1:
...
^
9.
K
(33)
value evaluated at X2, and noting that this gap difference
text, yields (23)
19
and therefore concludes the proof
Figure
1
Inc
Job Security and Rule of Law in Countries with High,
:
3==1
Inc
Medium and Low Income
3==2
PRT
.382183
B#>=
PAN
VEN
FIN
DEU
ESP
ARG
TWN
FRA
CHL
NOR
Gl !C
K3R
ITA
BEL
-„
TUR
ZAF
ISR
IFi45t>lK
USA
NZL
^
URY
HKG
-.327817
-1.92079
lnc_3==3
^
^
SGP
MYS
PER
ECU
.382183]°^
^1
COL
yogOL
PHL
jQ[,
IDfl'™
lKATHA
TUN
SEN
^IGA
BFA
KEN PA1?^H^A°
JAM
MAR
-.327817
-1.92079
ZMB
1.26311
Rule of
20
Law
1.26311
Figure
2:
Productivity
Growth and Speed of Adjustment
coef =. 05275722, se = 01 829659,
.
t
= 2.88
04HKG
KOR
THA
o
0)
PRT
O
.
MWI
00
0)
TUR
j^f^
j:^MK
0-
KEN
USA
BEL
u>
FRA
Q.
"
LL
h
15
3
C
C
CHl'ECU
ZMB
VEN
ZWE
PER
<
MEX
PAN
ARG
041
1
1
.3
Speed
of Adj. fcontr.by sector)
21
Table
Specification:
(2)
(1)
1:
ESTIMATING ^
(3)
Zjci
Observations
Income Group
Job Sec. Group
Extreme obs. of instrument
Standard errors reported
instrumental variable.
in
in
Employment
(6)
(7)
(8)
(ha)
-0.280
-0.394
-0.558
-0.355
-0.387
-0.363
-1.168
-0.352
(0.044)
(0.068)
(0.135)
(0.119)
(0.116)
(0.091)
(.357)
(0.103)
22,810
22,008
8,311
6,378
7,319
7,730
6,883
7,036
All
All
1
2
3
All
All
AH
All
All
All
All
All
1
2
3
Yes
No
No
No
No
No
No
No
in parentheses. All estimates are significaiit at the
As
(5)
(4)
Change
described in the main text,
zjct represents the
1%
level.
All regressions use lagged
Awj^.,
— Aw.c;
as,
log-change of the nominal marginal productivity of labor
each sector, minus the country average, divided by one minus the estimated labor share. All regressions includes a country-year
Income groups are 1 High Income OECD, 2; High Income Non OECD and Upper Middle Income, and
Lower Middle Income and Low Income. Job Security Groups correspond to the highest, middle an lowest third of the measure
Djankov et al. (2003).
fixed effect (k^^ in (9)).
;
22
3:
in
Table
2:
SAMPLE COVERAGE AND Main Variables
Job Security
WDI
code
AUS
Inc
Group
Djankov
ct, al
AUT
-0.19
-0.15
BEL
-0.11
CAN
DEU
-0.16
DNK
-0.21
0.17
InsdlutiDns
HP
-0.71
-0.65
-0.70
-1.64
-1.56
0.17
1.29
FIN
0.24
FRA
QDR
GRC
-0.02
-0.13
^0.04
IRL
ITA
JPN
-0.21
-0.09
-0.82
-1.09
-1.00
-1.05
-1.40
NLD
NOR
NZL
PRT
0.04
-0.03
-0.29
RL
Rule of Law
Gov.
Eff.
1.03
0.95
1.13
0,92
081
0.92
1.04
0.92
1.02
41
0.64
1.22
089
081
0.78
1.09
1.05
-0.01
-0,06
0.92
0-S2
0.79
0.09
0.05
-1.84
-1.53
-1.55
0.76
0.46
1.09
1.25
1.23
1.13
-2.21
1.22
1,25
0.37
2.05
0.53
0,24
SWE
0.06
0,97
-0.25
-0.50
-2.43
1.17
USA
095
1,01
-0.48
-1.00
-0.37
-0.82
ARC
BRA
CHL
HKG
ISR
0.11
0.56
0.36
0.61
-0.02
-0.32
0.21
•0.17
KOR
-0.07
1.14
MEX
MYS
0.38
0.73
PAN
SGP
0.34
TUR
-0.22
-0.13
TWN
O.OI
1
0.44
0.32
1
1
0.86
0.81
1
1
0.36
0.42
1
1
0.02
-0.15
-0.85
-0.S6
-0.24
1
1.37
I
1.54
1
0.05
0.18
-0.50
-1.19
1.26
1.41
-0.73
-0.69
0.21
0.49
-0.17
-1.32
-0.40
URY
VEN
-0.30
-0.20
-0.26
0.31
4.29
ZAF
-0.17
-1.38
-0.42
BFA
-0.10
BOL
COL
ECU
0.24
2.32
0.29
1,17
0.34
097
EGY
0.13
GllA
-0.17
IDN
IND
JAM
JOR
0.22
KEN
LKA
-0.16
MAR
MDG
MOZ
-0.22
MWI
NGA
PAK
PER
PHL
SEN
TIU
TUN
ZMB
ZWE
-1.46
-1.37
-1.19
-1.13
-0.53
-0.86
-1.09
-0.77
-0.95
-0.56
0.10
-0.14
-0.20
-0.44
-1.4S
-0.48
-0.57
-1.55
-1.92
-0.94
-1.89
0.09
0.23
0.38
0.11
-0.07
-0.15
0.37
-1. 16
-1.08
-0.86
-0.92
-0.29
-0.69
-1,08
-0.97
2.25
0.24
-0.04
0.10
0.05
-0.33
-0.13
23
Hish Gov. Eff
0,81
1.02
1.17
ESP
-0.14
Strong;
-1.38
-1.12
-0.61
•.
-1.29
-0.99
-0.78
-0.55
-0.79
-1.06
-0.54
-1.13
-0,93
-0,73
-1,39
-1-23
-1.32
-1.68
-1.02
-0
87
-0..54
-1.04
-0.32
-0.24
-1.44
-0.86
1
1
1
.
Table
Inc.
3:
BASELINE Sample Statistics*
Employment Growth (Yearly
Mean
Group
Obs.
Avge.):
Min
-0.24
-0.43
-0.78
-0.78
1
8,607
-0.01
0.06
2
6,063
0.00
0.11
3
7,063
0.02
0.16
Total
21,733
0.00
0.11
Inc.
Job Security from Djankov
Group
Countries
Mean
-0.05
20
1
1980-2000
SD
et al.
Max
0.26
0.42
0.96
0.96
(2003): JS
SD
Min
0.18
-0.29
-0.32
-0.33
-0.33
2
15
-0.01
0.25
3
25
0.05
0.21
Total
60
0.00
0.21
Job Security from Heckman and Pages (2001):
Countries
Mean
Group
SD
Min
-2.43
-0.87
19
1.15
1
-0.20
1.14
1.30
2
9
-0.44
5
1.26
1.13
3
-2.43
Total
33
0.00
1.54
Max
0.37
0.38
0.38
0.38
HP
Max
Inc.
Rule of Law from Kauftnann
Inc.
Group
Countries
et al. (1999):
2.05
4.29
2.32
4.29
RL
Mean
SD
Min
Max
1
20
0.88
0.37
-0.01
1.23
2
15
0.72
3
25
60
-1.38
-1.92
-1.92
-0.29
Total
-0.16
-1.03
-0.18
Governmem
Inc.
?
Group
Effectiveness from
Countries
0.42
0.96
Kaufmann
SD
Min
-0.06
-1.32
-1.68
-1.68
1
20
0.80
0.37
15
0.76
3
25
Total
60
-0.16
-0.95
-0.17
0.36
0.90
Correlation Country
JS
HP
RL
GE
*Income groups
dle
GE
Max
1.25
1.41
•
-0.24
1.41
Means
HP
RL
GE
1.00,
.TS
OECD
1.26
(1999):
Mean
2
come Non
et al.
1.26
0.66
1.00
-0.36
-0.77
-0.77
-0.35
are:
l=Higli Income
1.00
0.97
OECD, 2=High
1.00
In-
and Upper Middle Income, 3=Lower Mid-
Income and Low Income.
24
Table 4:
(1)
(2)
Estimation Results
(3)
(4)
Change
Gap(Xi)
(6)
(7)
0.600
0.603
0.607
0.611
(0.009)*"
(0.008)***
(0.012)*"
(0.012)**'
-0.080
-0.015
-0.025
-0,126
-0.027
-0,038
(0.037)'*
(0.051)
(0.051)
(0.041)"*
(0.052)
(0.051)
GapxJS(^2):
GapxJSxDSRL(X3)
-0.514
-0.314
(0.068)"*
(0.070)***
GapxJSxDHGE(X3)
Gap X HP
(5)
(8)
Log-Employment
in
-0.515
-0.326
(0.068)***
(0.071)"*
-0.022
(X2)
(0.007)*"
Conti-ols
Gap X Dummy
Higli
RL
-0.076
0.086
(0.015)*'*
(0.023)"*
Gap X Dummy High GE
Observations
R-squared
Gap-Income
Interaction
Gap-Sector Interaction
Efficiency
Heckman and Pages
dummies
(in
0.045
(0.023)*
21,733
21,733
21,733
21,733
21,733
21,733
21,733
12,012
0.60
0.60
0.60
0.60
0.61
0.61
0.61
0.62
No
No
No
No
No
No
No
No
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
* significant at 10%; ** significant at
(2003) and
-0.091
(0.015)***
5%; ***
.
significant at 1%. Robust standard errors in parentheses. JS
(2000) job security measures, respectively.
both cases the threshold
regression has country-year fixed effects.
is
DSRL
given by Greece, see the main
Gaps are estimated using
a constant
estimated values of Gap.
25
(|)
and
text), respective!}',
=
0.40.
and
HP
stand for
DHGE stand for strong Rule of Law
using the
Sample excludes
Kaufmann
the upper
tlie
Djankov
et al.
and high Government
etal. (1999) indices.
and lower
1%
Each
of Ae and of the
_i
£
o
o
<
u
w
a,
-I
a
i
I
00
w
>
I I
=i>
[3
[5
<d
o
H
D
to
£
Q,
e
w 3
O 2
CJ
C3
O
o
w
H
P
CQ
O
e-
M
.^>
> i
j=)
_o
Pi
f2
£
6
3
Q
X
O
26
erf
O
Table
6:
EXCLUDING ONE COUNTRY AT A TIME
^2
Country
Coeff.
ARG
AUS
AUT
BEL
BFA
BOL
BRA
CAN
CHL
COL
DEU
DNK
ECU
EGY
ESP
FIN
FRA
GBR
GHA
GRC
HKG
IDN
IND
IRL
ISR
ITA
JAM
JOR
JPN
KEN
^2
^3
Dev.
Country
Coeff.
-0.51
0.07
0.07
KOR
LKA
-0.02
-0.02
-0.02
0.05
-0.52
-0.52
-0.52
-0.50
-0.52
-0.52
-0.52
-0.53
-0.51
-0.52
-0.52
-0.50
0.05
-0.51
0.07
0.05
-0.53
0.07
0.05
0.07
0.05
-0.54
-0.51
-0.51
-0.48
-0.51
-0.37
-0.51
-0.54
-0.54
-0.52
0.05
-0.51
0.07
0.05
-0.51
0.05
0.05
Dev.
Coeff.
-0.01
0.05
-0.02
-0.02
-0.02
-0.03
0.05
0.00
0.05
-0.01
0.05
-0.02
-0.02
-0.02
-0.02
-0.02
-0.03
-0.02
-0.02
-0.02
-0.02
-0.02
-0.05
-0.02
-0.02
-0.02
0.05
St.
0.05
0.05
0.05
0.05
0.05
0.05
0.05
0.05
0.05
0.05
0.05
0.05
0.05
0.01
0.05
-0.02
-0.02
-0.02
-0.02
-0.04
-0.02
-0.15
0.05
0.05
St.
MAR
MDG
MEX
MOZ
0.07
0.07
0.07
0.07
^3
Dev.
Coeff.
0.05
0.07
0.05
-0.52
-0.51
-0.51
-0.02
0.05
-0.51
0.07
0.00
0.05
0.02
0.05
-0.53
-0.55
-0.52
-0.46
-0.53
-0.51
-0.51
-0.53
-0.55
-0.52
-0.59
-0.50
-0.54
-0.53
-0.52
-0.53
-0.51
-0.51
-0.50
-0.49
-0.50
-0.53
-0.53
-0.51
-0.51
-0.55
St.
0.05
0.07
MWI
-0.01
0.05
0.07
NYS
-0.02
0.05
0.07
0.00
0.05
-0.02
0.05
0.07
NGA
NLD
NOR
0.05
0.07
NZL
-0.02
-0.02
0.07
PAK
PAN
0.02
0.05
-0.01
0.05
0.07
PER
PHL
PRT
SEN
SGP
0.07
0.07
0.07
0.05
0.06
0.05
-0.03
0.05
-0.02
0.05
0.00
0.05
-0.02
-0.02
-0.01
-0.02
-0.03
-0.02
-0.02
-0.02
0.05
0.00
0.05
0.07
SWE
THA
TUN
TUR
0.07
TWN
0.07
0.07
URY
USA
VEN
-0.49
0.07
ZAF
-0.02
0.05
-0.52
-0.38
0.07
ZMB
ZWE
-0.02
0.05
0.03
0.05
This table reports the estimated coefficients for
one countiy (the one indicated for each
set
0.07
0.07
0.07
0.07
"k^
and
^,3,
of coefficients)
for the specification in
at at time.
27
0.05
0.05
0.05
0.05
0.05
0.05
0.05
Column
St.
Dev.
0.07
0.07
0.07
0.07
0.07
0.07
0.07
0.07
0.07
0.07
0.07
0.07
0.07
0.07
0.07
0.07
0.07
0.07
0.07
0.07
0.07
0.07
0.07
0.07
0.07
0.07
0.07
0.07
3 of Table 4, leaving out
Table
ESTIMATION RESULTS EXCLUDING
7:
(2)
(1)
(3)
(4)
Change
Gap(Xi)
GapxJS
Hong Kong and Kenya
in
(5)
(V)
(6)
0.615
0.620
0.649
0.652
(0.009)***
(0.009)***
(0.012)*"
(0.012)***
-0.105
-0.156
-0.163
-0.204
-0.171
-0.183
(0.039)***
(0.051)*"
(0.05 1)***
(0.042)***
(0.052)'**
(0.052)***
(X-.):
GapxJSxDSRL(^3)
-0.231
-0.062
(0.062)***
(0.072)
Gapx]SxDHGE(>.3)
(8)
Log-Employment
-0.227
-0.071
(0.070)***
(0.072)
GapxHP(^2)
-0.021
(0.007)***
Contivls
Gap X Dummy High RL
Gap X Dummy High
R-squared
Interaction
Gap-Sector Interaction
* significant at
(2003) and
10%, **
Heckman
0.065
(0.023)*'*
GE
Observations
Gap-Income
-0.121
(0.015)'"
-0.136
0.023
(0.015)***
(0.024)
20,881
20,881
20,881
20,881
20,881
20,881
20,881
12,003
0.61
0.61
0.61
0.61
0.62
0,62
0.62
0.62
No
No
No
No
No
No
No
No
Yes
Yes
Yes
Yes
Yes
Yes
significant at
1%. Robust standard
significant at
5%; ***
and Pages (2000) job security measures, respectively.
Law and Government Efficiency dummies,
Gaps are estimated using
a constant
(|)
=
respectively, using the
0.42.
Sample excludes
DSRL
Kaufmann
the upper and lower
28
errors in parentheses. JS
and
et al.
Yes
DHGE stand for high
and
HP
stand for
Yes
tire
Djankov
(above Greece, see main
text)
et al.
Rule of
(1999) indices. Each regression has country-year fixed effects.
1%
of Ae and of the estimated values of Gap.
Table
8:
IV ESTIMATION
Average speed of adjustment
Estimation
Baseline
Method
Model (Column
1
in
Gap instrumented with wage
Table 4
data
Standard dynamic panel formulation
See section 3.2.3 for
details.
29
Point Estimate
Robust Standard Error
0.600
0.009
0.570
0.065
0.543
0.07S
Table
Country
9:
^c,I
Estimating Country-Specific Speeds of Adjustment
St.
Ki
Dev.
St.
Country
Dev.
Kj
St.
Dev.
Ki
St.
Dev.
KOR
LKA
0.719
0.028
0.696
0.052
0.744
0.041
0.729
0.056
0.572
0.060
0.569
0.071
0.688
0.067
0.666
0.075
0.467
0.042
0.451
0.058
0.064
MAR
MDG
MEX
MOZ
0.414
0.111
0.370
0.122
0.346
0.078
MWI
0.499
0.095
0.437
0.094
0.038
0.547
0.055
NYS
0.750
0.032
0.723
0.053
0.631
0.045
0.618
0.061
NGA
0.782
0.094
0.754
0.102
COL
0.624
0.030
0.591
0.052
0.467
0.060
0.414
0.073
DEU
0.463
0.048
0.458
0.062
0.472
0.037
0.468
0.054
DNK
0.500
0.058
0.488
0.070
NLD
NOR
NZL
0.483
0.065
0.443
0.076
ECU
EGY
0.645
0.044
0.624
0.059
0.771
0.054
0.740
0.069
0.694
0.052
0.671
0.065
PAK
PAN
0.575
0.049
0.561
0.064
ESP
FIN
0.488
0.033
0.469
0.052
PER
0.379
0.038
0.376
0.056
0.445
0.032
0.428
0.051
PHL
0.664
0.036
0.652
0.055
FRA
0.292
0.032
0.278
0.052
0.369
0.029
0.362
0.050
GBR
GHA
GRC
HKG
0.577
0.037
0.565
0.054
0.716
0.080
0.707
0.090
0.502
0.064
0.498
0.075
PRT
SEN
SGP
0.631
0.039
0.605
0.057
0.552
0.032
0.536
0.052
0.578
0.037
0.555
0.054
0.837
0.029
0.821
0.050
0.490
0.143
0.450
0.142
IDN
IND
0.676
0.046
0.645
0.061
0.633
0.099
0.605
0.099
0.746
0.051
0.736
0.067
SWE
THA
TUN
TUR
0.506
0.031
0.475
0.051
IRL
ISR
0.516
0.031
0.488
0.051
TWN
0.406
0.037
0.379
0.056
0.606
0.040
0.592
0.057
0.584
0.034
0.575
0.053
ITA
0.364
0.047
0.341
0.062
0.544
0.040
0.542
0.056
JAM
0.563
0.411
0.510
0.364
URY
USA
VEN
0.519
0.037
0.515
0.056
JOR
0.697
0.041
0.681
0.057
ZAF
0.581
0.044
0.548
0.059
JPN
0.470
0.035
0.463
0.054
0.482
0.095
0.457
0.106
KEN
0.224
0.038
0.201
0.055
ZMB
ZWE
0.774
0.051
0.749
0.064
Istquintile
0.346
2nd
quintile
0.482
0.457
3rd quintile
0.546
."•0.527
4th quintile
0.619
0.600
5th quintile
0.743
0.723
ARG
0.380
0.060
0.364
0.071
AUS
AUT
BEL
BFA
BOL
0,558
0.048
0.537
0.063
0.521
0.040
0.504
0.056
0.160
0.046
0.158
0.062
0.327
0.066
0.309
0.076
0.562
0.049
0.545
BRA
CAN
CHL
0.385
0.067
0.565
0.325
This table reports estimated coefficients for X
Xc.2 includes sectoral controls,
while Xc
i
at the
cotmtry level.
does not.
30
A common
value of
())
=
0.40
is
used throughout.
Table 10:
PRODUCTIVITY GROWTH AND SPEED OF ADJUSTMENT
Change
Reported: change
in
in X^-Quintile
Change
in
Annual Growth Rate
Isttolud
0.88%
2nd
3rd
0.29%
3rd to 4th
0.23%
4th to 5th
0.28%
1st to 5th
1.68%
to
I
annual growth rates associated with moving from the average adjustment speed of one quintile
next Quintiles for X^ from Table
9,
equation (19)). Parameter values: y
column with
=
5/6), gyj
sectoral controls. Calculation
= 0.007, c =
0.27
31
a=
based on model described
2/3, and 5
=
0.06.
in Section
to the
4 (see
Table 11:
PRODUCTIVITY GROWTH AND Speed of Adjustment
(1)
(2)
(4)
(3)
Annual Average Change
Speed of adjustment:
Constant:
Observations
R-squared
Income Fixed
Effects
Estimation via
10%, **
* significant at
size determined
TFP growth =
by
Per-capita
0.040
0.031
0.053
0.026
0.026
0.046
(0.019)*
(0.018)'**
(0.019)
(0.019)
(0.019)**
-0.016
-0.013
-0.016
-0.012
-0.012
-0.014
(0.010)
(0.010)
(0.009)*
(0.011)
(0.010)
(0.009)
42
42
42
42
42
42
O.IOS
0.066
0.294
0.043
0.041
0.230
No
OLS
No
OLS
OLS
No
No
Yes
WLS
WLS
WLS
growth
comes from World Penn Tables
weights inversely proportional
5%; **"
of TFP data.
GDP
5.6
-
significant at
TFP growth
7!^
1%. Robust standard errors
at the
growth
and Barro and Lee (1996).
to the standard deviation
in
Yes
parentheses. Sample-
country level calculated following Marjciw (1995):
0.3 Per-capita capital
-
0.5 Avge. Schooling growth.
WLS
of estimated
32
2427
(6)
(5)
TFP 1980-1990
(0.018)**
significant at
availability
in
II
^^'s.
refers to
weighted
OLS refers
The data
least squaies,
with
to ordinary least squares.
Date Due
Lib-26-67
MIT LIBRARIES
3 9080 02618 0353
V
i
liill
imi
"'
ill'
II
,1
'
r
ilBiifc
iiiil>
I
,liii,iilm,ili!'
'
m
'
'
<
kl
iiii
mm\w
m\WAm>h
mm\kk4
^^m.
v\
i
m
Ji)!t!iU,.!iil,iii.„!
imvM
.lM'',i!,
ptf|ip|fjfi,iiiriiiiiMii'i;iiif;i;ii,
!'!
m
i>ii''i
1
wm]\\
Villi
i'
ir
'
It
i\''
lill'Mtlh!,
I
';
,
1
!'
i';ii',v,iii
idgMlMMMIMMl
IHl
>'