Content uploaded by Gilles Mourre
Author content
All content in this area was uploaded by Gilles Mourre
Content may be subject to copyright.
WORKING PAPER SERIES
NO. 358 / MAY 2004
DID THE PATTERN
OF AGGREGATE
EMPLOYMENT
GROWTH CHANGE
IN THE EURO AREA
IN THE LATE 1990s?
by Gilles Mourre
In 2004 all
publications
will carry
a motif taken
from the
€100 banknote.
WORKING PAPER SERIES
NO. 358 / MAY 2004
DID THE PATTERN
OF AGGREGATE
EMPLOYMENT
GROWTH CHANGE
IN THE EURO AREA
IN THE LATE 1990s? 1
by Gilles Mourre 2
1 The opinions expressed in this paper are those of the author and do not necessarily reflect the views of the European Central Bank.
All the errors and omissions are my own. I thank Gerard Korteweg, Neale Kennedy, Geoff Kenny, Franck Sédillot, Marie Diron,Mark
Stocker, Jarkko Turunen,Reiner Martin, Nadine Leiner Killinger, Ramon Gomez Salvador, Julian Morgan, Julian Messina,Francesco
Mongelli, Ana Lamo and an anonymous referee for helpful comments and discussions. I am grateful to Franck Sédillot and Marie
Diron who provided me with some econometric routines.
2 The paper was written when the author was working at the European Central Bank. E-mail address: gilles.mourre@cec.eu.int.
This paper can be downloaded without charge from
http://www.ecb.int or from the Social Science Research Network
electronic library at http://ssrn.com/abstract_id=533027.
© European Central Bank, 2004
Address
Kaiserstrasse 29
60311 Frankfurt am Main, Germany
Postal address
Postfach 16 03 19
60066 Frankfurt am Main, Germany
Telephone
+49 69 1344 0
Internet
http://www.ecb.int
Fax
+49 69 1344 6000
Telex
411 144 ecb d
All rights reserved.
Reproduction for educational and non-
commercial purposes is permitted provided
that the source is acknowledged.
The views expressed in this paper do not
necessarily reflect those of the European
Central Bank.
The statement of purpose for the ECB
Working Paper Series is available from the
ECB website, http://www.ecb.int.
ISSN 1561-0810 (print)
ISSN 1725-2806 (online)
3
ECB
Working Paper Series No. 358
May 2004
CONTENTS
Abstract 4
Non-technical summary 5
1. Introduction 7
2. Estimation of a standard employment equation 8
2.1 Theoretical framework 8
2.2 Data used
10
2.3 Estimation method
10
3. Is there any evidence of a structural change in
recent years?
12
3.1 Evidence of a break in the standard
employment equation for the euro area
12
3.1.1 Quality of the dynamic simulation
and the forecasting performance
when allowing for a break
13
3.1.2 The role of traditional determinants,
when allowing for a break
14
3.2 Robustness of the break while considering
hours worked or employment in full-time
equivalents
15
3.3 Taking account of heterogeneity across
countries
16
4. What factors may account for a change in
aggregate employment pattern in recent years?
17
4.1 Changes in the sectoral composition of
euro area employment
17
4.2 The importance of labour market
institutions
18
4.3 The role of active labour market policies
21
4.4 The role of structural changes in explaining
cross-country differences in recent
employment performance
22
Conclusion
24
References
25
Figures and tables
28
Annex
42
European Central Bank working paper series 45
Abstract
The paper examines whether the pattern of growth in euro area employment seen in the period 1997-
2001 differed from that recorded in the past and what could be the reasons for that. First, a standard
employment equation is estimated for the euro area as a whole. This shows that the lagged impact of both
output growth and real labour cost growth, together with a productivity trend and employment “inertia”,
can account for most of the employment developments between 1970 and the early 1990s. Conversely,
these traditional determinants can only explain part of the employment development seen in recent years
(1997-2001). Second, the paper shows sound evidence of a structural break in the aggregate employment
equation in the late 1990s. Third, the paper provides some tentative explanations for this change in
aggregate employment developments, using in particular country panels of institutional variables and of
active labour market policies but also cross-sectional analyses. Among the relevant factors likely to have
contributed to rising aggregate employment in recent years are changes in the sectoral composition of
euro area employment, the strong development of part-time jobs, lower labour tax rates and possibly less
stringent employment protection legislation and greater subsidies to private employment.
JEL Classification numbers: C2, E24, H50, J23.
Key words: Euro area; Aggregate employment; Demand for labour; Labour market institutions; Active
labour market policies.
4
ECB
Working Paper Series No. 358
May 2004
Non-technical summary
The ratio of employment growth to real GDP growth indicates that real GDP growth was more job
intensive in the euro area over recent years, compared with the late 1980s. The aim of this paper is to
explain the high employment growth observed in 1997-2001 by means of a standard labour demand
equation and to see whether the pattern of aggregate employment in the euro area has changed in recent
years. Compared to the late 1980s, the strong growth in employment observed between 1997 and 2001 is
partly explained by a buoyant, albeit somewhat lower, economic growth and by much lower labour cost
growth.
However, the standard employment equation failed to fully account for the good employment
performance observed in the euro area in 1997-2001. The introduction of a break from 1997 onwards
turns out to be statistically very significant, improves greatly the quality of the dynamic simulation and
increases the stability of the equation. The forecasting performance of the equation also improves
strongly. This break may be interpreted as the additional employment growth recorded between 1997 and
2001 which cannot be explained by traditional determinants. Although the choice of the starting date of
the break is somewhat arbitrary, its statistical significance is maximised when it starts in 1997.
When the employment equation is re-estimated with employment measured in terms of full-time
equivalents and hours worked instead of the number of people employed, the break is still significant,
although of a lesser magnitude. This indicates that developments in part-time employment have
contributed to the strong employment performance in the 1990s but cannot fully account for the break in
labour demand in the euro area.
Taking account of heterogeneity across OECD countries, panel data estimates show that most euro
area countries (representing almost two thirds of euro area employment) have experienced a positive
break in their aggregate labour demand since 1997. However, five euro area countries, including
Germany, did not record any significant change in their employment equation in the late 1990s. No
positive break is significant for countries outside the euro area. Panel data estimates broadly confirm the
timing of the break, starting in around 1997, although the precise dating varies slightly across countries.
In addition to part-time developments, three possible causes for the change in the aggregate labour
demand are examined: changes in the sectoral composition of euro area employment, developments in
labour market institutions and the impact of active labour market policies. A simple accounting exercise
indicates that the average annual growth rate of employment in the period 1997-2001 would have been
0.2 percentage point lower if the sectoral composition of employment had remained the same as in 1986-
1991. Therefore, the higher weight of fast growing and job-intensive sectors, such as market-related
services, in the late 1990s compared to the past, is likely to have contributed to fostering employment
growth in recent years.
Albeit difficult to show clearly, labour market reforms and changes in institutions in many euro area
labour markets may also have played a role in the recent good employment performance in the euro area.
Panel data estimates suggest that total taxes on labour negatively affect employment growth. More mixed
5
ECB
Working Paper Series No. 358
May 2004
evidence suggests that employment protection legislation may reduce employment growth, whilst the
level of bargaining coordination is found to be positively correlated with employment. As a result, the
decline in total labour tax rates, in particular on low-wage employees, is likely to have positively
influenced job creation.
Active labour market policies might also contribute to explaining the good employment performance
recorded in the recent period. However, the results coming from tentative panel data estimates appear
very mixed and display a low level of statistical significance. The aggregate expenditure devoted to
public employment services and labour market training are not significant. While subsidies to private
employment may be positively related to the employment rate, the impact of direct job creation in the
public or non-profit sectors is clearly insignificant. The effect of measures for youth employment does not
come out clearly.
The role of structural changes may be highlighted further by relating the cross-country differences
observed in the employment pattern since 1997 to changes recorded in the sectoral composition of
employment, institutions and active labour market policies in the second half of the 1990s. A clear
negative relationship emerges between the tax wedge and the presence of a positive break in recent
employment performance. Although the other variables are less tightly linked to the presence of positive
break in employment pattern, the cross-country analysis confirms that the impact of changes in the
sectoral composition of employment is correlated to employment performance. The strong decline in
employment protection legislation in some countries may also explain partly their good employment
performance. Moreover, part-time employment rate and subsidies to regular employment in the private
sector may have helped improve employment performance in the late 1990s in some countries.
Conversely, other institutions such as unionisation, benefit replacement rate, benefit duration or most
active labour market policies (public employment services, labour market training and direct job creation
in the public sector) do not display any obvious link with the employment performance in euro area
countries in the late 1990s. These cross-country results broadly support the general findings arising from
panel data analysis.
Overall, part-time employment developments, changes in the sectoral composition of employment
and decreasing labour tax rates are good candidates to explain at least partly the break in aggregate
employment equation seen in the late 1990s.
6
ECB
Working Paper Series No. 358
May 2004
1. Introduction
Until recently, the emphasis in the economic literature to account for the improvement in labour
markets in Europe in 1997-2001 was mainly put on the decline in structural unemployment. It has been
argued that overly rigid institutional structures have prevented the necessary adjustment to changes in the
economic environment, thereby leading to higher or more persistent unemployment (Scarpetta, 1996;
Nickell 1997; Layard and Nickell, 1998; Morgan and Mourougane, 2001). Blanchard and Wolfers (2000)
attribute the rise in unemployment in Europe to the interaction of institutions with adverse
macroeconomic shocks.
More recently, several studies have dealt with employment growth directly to shed some light on the
strong improvement of labour market performance in many European countries in the late 1990s. From a
descriptive point of view, Duchêne and Jacquot (1999) investigate whether a break occurred in the trend
growth rate of labour productivity per person employed during the first half of the 1990s in the main
OECD countries and whether such a break could be accounted for by changes in relative factor costs or in
the number of hours worked. Some other studies have focused on specific aspects to explain the
improvement in net employment creation, such as wage discipline in EMU (Pichelmann, 2001) or the
change in employment composition (ECB, 2002a). Some more comprehensive studies have attempted to
survey all the changes capable of accounting for the higher job intensity in Europe (European
Commission, 2000; Decressin et al., 2001; Garibaldi and Mauro, 2002). The geographic focus varies
across these studies (EU countries, large euro area countries or OECD countries). This paper continues in
this vein by analysing the determinants of employment for the euro area as a whole, while not neglecting
heterogeneity across countries. From a methodological point of view, the paper follows that of Fagan,
Henry and Mestre (2001), who estimated an aggregate dynamic employment equation for the euro area
with an error correction mechanism. While Fagan et al. ran their equation up to 1997, this article focuses
on the pattern of employment in recent years.
The employment rate in the euro area, at almost 64% in 2001, was considerably lower than in the
United States (nearly 75%). The period 1997-2001 however saw a protracted period of sustained
employment growth, which led to a fall in unemployment despite a strong increase in the labour force.3
Total employment has grown at an average year-on-year rate of 1.5% from 1997 to 2001, compared with
a decline of 0.2% between 1990 and 1997. This corresponds to an increase of around 7 million in the
number of persons employed, whereas earlier in the 1990s, by comparison, employment fell by over 1
million. This strong employment growth is also noticeable when compared with the previous period of
strong growth, registered in the late 1980s. Since the late 1960s, as seen in Table 1, the average growth
rate recorded in the late 1990s is only comparable to that recorded in the second half of the 1980s (1.4%).
However, the ratio of employment growth to real GDP growth indicates that real GDP growth was more
job intensive in the recent period, at 0.6, compared with 0.4 in the late 1980s. Likewise, the ratio of
employment growth to real GDP growth became higher in the late 1990s in the euro area than that in the
3 See, for example, “Labour force developments in the euro area since the 1980s”, Véronique Genre and Ramón Gómez-
Salvador, July 2002, ECB Occasional Paper No. 4.
7
ECB
Working Paper Series No. 358
May 2004
US and the UK. By contrast, this ratio was lower than in the US during past periods of expansion and also
lower than that in the UK in the late 1980s.
What accounts for this development? This paper argues that the traditional determinants (GDP
growth, labour cost developments, trend productivity) do not fully explain the strong employment growth
recorded in the euro area in the recent period. Sound econometric evidence, based mainly on time-series
analysis, but also on panel data analysis, suggests that the recent employment performance is related to a
structural change in aggregate employment behaviour in the euro area. Looking further, in order to
explain the factors underlying this change, a panel of time-varying institutions and measures of active
labour market policies is used. This latter methodological approach can be associated with the branch of
the literature initiated by Scarpetta (1996) and extended by Belot and van Ours (2000) and Nickell et al.
(2001). These articles used cross-sectional or pooled time series data on indicators of labour market
performance and labour market institutions to account for unemployment differentials across countries.
However, the results found using institutional variables are often unclear or not robust, partly due to
measurement problems.
The rest of the paper is structured as follows. Section 2 presents a standard employment equation,
estimated for the euro area as a whole. Section 3 shows the existence of a structural break in the
aggregated employment equation in the late 1990s. Section 4 provides some tentative explanations for
this change, using in particular country panels of institutional variables and of active labour market
policies as well as cross-sectional analysis.
2. Estimation of a standard employment equation
2.1 Theoretical framework
A CES production function with two production factors and constant returns to scale, proposed by
Arrow et al. (1961), provides a simple and standard analytical framework to highlight the effect of the
main determinants of labour demand:
1
1
1
)1()(
−
−
−
−+=
σ
σ
σ
σ
σ
σ
αα
ttt
KLaY
with Y standing for output, L for labour, K for capital, a for labour productivity
4
IRUWKHODERXULQWHQVLW\
of the method of production and for the elasticity of substitution between effective labour (aL) and
capital. Then, the first order condition of firm’s profit maximisation leads to equate the marginal labour
productivity to real compensation per employee w/p. This leads to the following expression:
σσσ
σ
α
111
tt
t
YLa
p
w
−
−
=
4
a represents the labour efficiency. It can also be seen as the degree of labour-augmenting technical progress (i.e. Harrod-neutral
technical progress).
8
ECB
Working Paper Series No. 358
May 2004
We suppose conventionally that labour productivity growth is constant and positive with log at = .t
DQG !0, reflecting trend technological progress. After rearranging and writing in logarithms, we end up
with:
log Lt = log <W ORJZS W ORJ
The employment level depends on total output, a labour productivity trend and real labour costs. The
elasticity of substitution is conventionally between zero and unity (imperfect substitution between
production factors), the elasticity of employment to real labour costs is negative and lower than 1 in
absolute value and the coefficient of productivity trend is negative as well. If the elasticity of substitution
is equal to unity, the production function becomes a Cobb-Douglas function and labour demand has the
following form: log Lt = log Yt - log(w/p) + log ZLWKDXQLWHODVWLFLW\RIUHDOODERXUFRVWVDQGQRWLPH
trend.
There are a number of other possible equivalent ways to specify the long run condition for
employment in this framework. For instance, instead of the profit maximisation problem, Fagan et al.
(2001) used the employment level induced by the inverted (Cobb-Douglas) production function, which
depends upon real GDP, total capital stock and trend total factor productivity. Alternatively, following the
cost-minimisation problem subject to a given capital stock, employment becomes a function of real GDP,
technical progress and relative factor prices. However, the choice of the profit-maximisation approach
stems from the fact that real wage statistics are more reliable than capital stock data or capital cost data
and available on a quarterly basis. Moreover, in this specification, employment only depends on output
and labour market variables (labour costs and trend labour productivity).
,Q WKLV VHWWLQJ UHDO ODERXU FRVW HODVWLFLW\ JLYHV D PHDVXUH RI WKH HODVWLFLW\ RI VXEVWLWXWLRQ ,Q
economic terms, this parameter means that a growth of 1% in the relative cost of labour to capital will
OHDG WR D JURZWK RI LQ WKH UDWLR RI FDSLWDO WR ODERXU 0RUH IRUPDOO\ ZH KDYH
d(K/L)/d(w/r)*r/w*L/K, where r is the cost of capital. The interpretation of the time trend should also be
discussed. The absolute value of its (negative) coefficient depends negatively on the elasticity of
VXEVWLWXWLRQ DQGSRVLWLYHO\RQWUHQGODERXUSURGXFWLYLW\PLUURULQJ WHFKQRORJ\ GHYHORSPHQWV DVVXPHG
to be constant over time. The constant, ORJ ZKLFKWXUQVLQWR ORJ ZLWK EHLQJWKHPDUNXSRYHU
costs in the case of imperfect competition (see Morgan, 2001), depends positively on the elasticity of
VXEVWLWXWLRQ WKHODERXULQWHQVLW\RIWKHPHWKRGRISURGXFWLRQ DQGWKHPDUNHWSRZHURIILUPV
In this setting, the elasticity of employment to output is unity. Calling into question this result would
mean to allow for the interaction between the level of labour productivity and the level of output. For
instance, if we set log at = W ORJ < ZLWK the elasticity becomes less than unity. Such an
interaction is difficult to explain. One could suppose that the level of output reflects the level of
knowledge in the economy, as argued in some endogenous growth models. But no sound evidence has
been provided on this. Therefore, in the subsequent section, we will assume an elasticity of employment
to output equal to unity and use panel data analysis to test it.
9
ECB
Working Paper Series No. 358
May 2004
2.2 Data used
The euro area data for the period 1985-2002 are built from the aggregation of quarterly ESA95
National Account series. Before 1985, they have been back-cast using data of the ECB area-wide model5.
Employment series refers to total employment (employees plus self-employed) in terms of persons
employed. An alternative could have been to reason in terms of hours worked or full-time equivalents.
This choice was motivated by two reasons. First, part-time employment data, needed to compute full-time
equivalents, are missing before 1991 on a quarterly basis (Eurostat quarterly labour indicator) and are not
available on an annual basis before 1983 (European Community Labour Force Survey)6. Second,
compared with the number of employed persons, the estimates of hours worked are more fragile and the
concept itself more uncertain (hours effectively worked or hours usually worked in the reference period).
The number of persons employed is more consistent with the measures of unemployment and the labour
force. However, tentative estimates based on hours worked and full-time equivalent employment will be
provided in this paper as a robustness check.
Labour cost data refer to total compensation per employee deflated by euro area GDP deflator at
market prices. This series encompasses total labour costs, i.e. direct (wages) and indirect (social security
contribution) remuneration. However, unlike employment data, compensation per employee covers only
employees’ compensation, excluding remunerations of the self-employed, which are unavailable. This
could slightly bias the estimation of the employment equation, because it is implicitly assumed that
average compensation received by the self-employed has grown at the same pace as compensation per
employee.
2.3 Estimation method
The explanatory variables being non-stationary, we have chosen an error-correction-model
specification, allowing for distinguishing the short-term dynamics from the long-term determinants
(corresponding to the cointegration relation). The use of quarterly data starting in 1970 yields a relatively
long time series dimension, allowing for the precise analysis of the dynamics, which generally requires
data with reasonable frequency and implies a large loss of degrees of freedom. Moreover, the relatively
long period chosen covers at least three full economic cycles, which will help to better distinguish the
cyclical behaviour of labour demand from possible structural changes.
From an economic point of view, the use of an error-correction model is justified by the existence of
costs of adjustment, which induce a slow response to shocks to labour demand (e.g. changes in GDP or
labour costs), as pointed out by the large literature on dynamic labour demand, e.g. Nickell (1986). As
explained by Hamermesh and Pfann (1996), these adjustment costs are of two kinds. First, the net costs
5 Data are downloadable from the ECB website. See ECB working paper no. 42, “An area-wide model (AWM) for the euro
area” by G. Fagan, J. Henry and R. Mestre, January 2001.
6 Other sources, for instance National Accounts, provide quarterly full-time equivalent series, which are unfortunately
available for a short time period.
10
ECB
Working Paper Series No. 358
May 2004
are those of changing the numbers of employees in the firms, for instance the loss of efficiency due to the
internal reorganisation of work. Gross costs of adjusting labour demand are those related to the flows of
workers entering or leaving the firm, such as search and recruitment costs, slow adjustment of capital
stock, the cost of training and job protection legislation (mandatory notice of layoffs, severance payment,
cost of legal disputes, etc).
In order to disentangle the long-term equilibrium relation between variables and the dynamics, we
will follow the two-step procedure of Engel and Granger (see Hamilton, 1994). We will estimate first the
long-run level of labour demand and identify a cointegration relationship between variables. Second, an
equation in first differences will be estimated to capture dynamics. In order to avoid endogeneity, the
contemporaneous quarterly change in GDP and real labour costs is omitted. An estimate using
instrumental variables approach, where the contemporaneous change in GDP is instrumented by lagged
changes in GDP, confirms that this current term is not significant. Conversely, it turns out to be highly
significant with a standard OLS approach, suggesting that the contemporary correlation between GDP
and employment growth mainly captures the reverse causality, i.e. the current impact of employment to
activity.
With E, Y and w/p standing respectively for total employment, real GDP and real labour costs, euro
area labour demand can be modelled by the following equation, where , and are estimated separately
by OLS.
∑∑∑
=== −−−−−− ++
+−++−−∆+∆+∆=∆ I
i
I
it
I
itttitiitiitit et
p
w
YEd
p
w
cYbEaE
110 111 )1()ln(lnln)ln(lnlnln
εγβα
However, this labour demand equation is estimated with actual employment data, which by definition
satisfy the equilibrium condition between labour demand and the labour supply. Therefore, labour supply
variables may have explained a part of the employment developments. Due to lack of data or data
limitation, it appears quite difficult to control for labour supply variables in a macroeconomic equation.
For instance, institutional data, constructed by Nickell and Nunziata (2001) and available at the country
level, cannot be aggregated at the euro level, given the strong methodological differences in the
construction of those series across countries. Moreover, the low number of observations limits the
relevance of using them in a time-series approach. Another relevant supply-side dimension, the structure
of population by educational attainment, cannot be taken into account, because of the lack of a long-time
series. However, some demographic variables may be used to control for a part of the labour supply
effects. The working age population7 appears a natural control variable positively related to employment
growth, as it represents the potential labour force. Moreover, recent studies, such as Korenman and
Neumark (2000) or Jimeno and Rodriguez-Palenzuela (2002) point to the importance of the age structure
of the working-age population and in particular of the relative size of the youth population to explain
aggregate employment and unemployment rates. Young people are most affected by labour market
7 The working-age population is defined as those aged 15-64 (OECD usual definition).
11
ECB
Working Paper Series No. 358
May 2004
institutions which impose some kind of wage floor (like minimum wages, collective bargaining,
employment protection legislation, unemployment benefits, etc.), which translates into a low youth
employment rate compared to that of prime–age people. The increase in the relative size of the young
population is supposed to decrease employment. Thus, an equation controlling for these shifts in the size
and structure of the working-age population is estimated. However, as shown in equations (1) to (4) in
Table 2, the effect of both demographic variables turns out to be clearly insignificant. This suggests that,
given the lack of specific labour-supply-related data, actual employment developments in the euro area
seem to be captured reasonably well by the standard labour demand equation8. This is highlighted by
equations (1) and (2) reported in Table 3 (see also the Annex for further details regarding the estimation
results).
3. Is there any evidence of a structural change in recent years?
This section presents some evidence pointing to a structural change in the employment behaviour in
the late 1990s. First, a break is introduced in the euro area employment equation, in which the number of
person employed measures employment. The role of the break and the traditional determinants is
carefully assessed. Second, the robustness of the break is tested by using other measures of employment
(full-time equivalent, hours worked), which permits to evaluate the importance of part-time employment
developments. Lastly, the question of cross-country heterogeneity is addressed.
3.1 Evidence of a break in the standard employment equation for the euro area
Some evidence points to a possible change in the pattern of euro area employment in recent years.
Although the overall stability of the equation is not rejected, some instability is visible at the end of the
period when performing a recursive estimate of the coefficient. This may explain why the Error
Correction Mechanism (ECM) term is not highly significant (see equations 1 and 2 in Table 3). Another
piece of evidence is the poor performance of the dynamic simulation at the end of the period, which
clearly overestimates employment in the early 1990s and underestimates it in the late 1990s (Figure 1).
Dynamic contributions computed on the basis of equation 1 in Table 3 and shown in Figure 2 suggest that
residuals have substantially contributed to employment growth in the period 1997-2001, explaining 0.7
p.p. of total employment growth each year on average. This is confirmed by the instability displayed by
equations (1) and (2) in Table 3 from around 1997, when estimating its coefficients recursively.
8 An alternative to the single equation approach would have been to estimate a system of two equations comprising a labour
demand equation and a wage equation. However, such an approach would face a serious problem of identification, as all the
terms of the long-run labour demand equation are also included in the long-run wage equation (see Morgan and Mourougane
2001). Indeed, by construction, the wage equation mixes labour demand aspects (firms’ willingness to pay wages) and labour
supply effects (employees’ bargaining power). Moreover, this approach is still affected by the problem of lack of macroeconomic
data on labour supply variables (skills, institutions, etc).
12
ECB
Working Paper Series No. 358
May 2004
3.1.1 Quality of the dynamic simulation and the forecasting performance when allowing for a break
In order to identify possible changes in employment pattern, we reestimate equation (2) (one-step
ECM procedures) in Table 3 by allowing for breaks in all variables. Then we sequentially remove the
least significant break until only statistically significant breaks remain. As shown in detail in Table 4, this
sequential procedure leads to retaining only one break, in the intercept. According to various criteria
(adjusted R square, t-value of the break, RMSE of recursive out-of-sample forecast), the break appears
most relevant when starting in 1997. This is in line with the result yielded by the dynamic contribution of
the residuals from the traditional equation and the recursive estimates of the coefficient. The break still
appears highly significant and its magnitude is unchanged, when controlling for some observable labour-
supply effects mentioned earlier (working age population and the relative size of the youth population).
This emerges when comparing equations (3) and (4) in Table 2 with equation (4) in Table 3.
The break in intercept can be interpreted as the additional employment growth recorded between
1997 and 2001 which cannot be explained by traditional determinants. It corresponds to an upward shift
in the long term relationship in levels, which translates into a higher but temporary employment growth
rate until the new long-term level is reached, unlike a break in the trend, which would imply a permanent
change in the growth rate. According to the theoretical model presented earlier in section 2, the increase
in the intercept may be interpreted either as an increase in the mark-up , which seems unlikely given the
increase in competition induced by the continuing integration in the Single market9, or a rise in the
ODERXULQWHQVLW\RIWKHPHWKRGRISURGXFWLRQ
When including a break, the adjusted R² from the two-step ECM estimation increases from 0.61 to
0.65 (see equations 1 and 3 in Table 3) and the dynamic simulations derived from either one-step or two-
step estimation appear to be very close to the actual series and much better than that given by equation 1
without a break (see Figure 3). The introduction of the break makes the error correction mechanism very
significant, which was not the case without a break. Each coefficient of the equation appears very stable.
The long-term elasticity of real labour costs (estimated in one step) is slightly lower than that estimated in
one step without the break. The strong elasticity of employment to real labour costs in equation 1 could
have artificially captured the structural changes, which occurred in a period of moderate wage
developments recorded since 1997.
Another illustration of the inability of traditional determinants to fully explain employment growth in
the recent period is provided by the results of the out-of-sample dynamic performance. Over 3 million
jobs created in the euro area since 1999 are not explained by the employment equation estimated between
1970Q1 and 1999Q1 (see Figure 4). In other words, 0.7 p.p. of the annual employment growth between
1999Q2 and 2002Q2 does not stem from the traditional determinants. More formally, the root mean
squared errors (RMSEs) of the out-of-sample forecasts are one third lower when allowing for a break.
9 However, the rising share of services, more protected in general from the international competition than industry, in the
whole economy might have contributed to raising the aggregate mark-up, offsetting the effect of enhanced competition coming
from the integration of product markets within the European Union.
13
ECB
Working Paper Series No. 358
May 2004
3.1.2 The role of traditional determinants, when allowing for a break
To illustrate the role of traditional determinants of employment over time, the dynamic contributions
to employment growth10 are computed from equation (4) with a break (see Table 3). The results regarding
the contribution of GDP or real labour costs are robust, whatever the equation considered (1), (2), (3) or
(4) and are well reflected in Figure 2. For instance, they are not much affected by the inclusion of a break
in the equation. Conversely, the reaction lags are very sensitive to the specification of the equation and,
particularly, to the inclusion of a break.
The positive employment performance in the euro area in 1997-2001 resulted to a large extent from
the robust economic growth in the second half of the 1990s (see Figure 2). Over the period 1980-2000,
employment seems to have been largely driven by GDP growth. For instance, the poor employment
performance in the early 1990s is clearly related to weak activity growth. Employment equations
(including all the lags up to five quarters) also allow for computing the adjustment lag of employment to
GDP over different sub-periods. As shown in Table 5, in the period 1970-2002, employment growth is
found to react to GDP with a mean lag of around eight quarters, when we estimate the equation with all
lags. The mean lag seems to have decreased in the late 1980s-early 1990s and increased in the recent
period. However, when we use an alternative equation, retaining the significant lags only, these results are
reversed and the mean lag of GDP between 1985 and 2002 shortens somewhat down to 6 quarters, which
leads us to interpret the mean lag with considerable caution. Conversely, the median (50% of the long-
term effect) reaction lag to GDP seems to have been fairly stable over the past thirty years, at around 4.5
quarters. On the whole, reaction lags tend not to signal any noticeable change in the adjustment process
of employment to GDP.
As shown by Figure 2, the deceleration of real labour cost growth between 1992 and 1996 set the
conditions for dynamic employment growth. By contrast with the late 1980s, real labour cost
developments remained moderate during the last upturn of 1997-2000, contributing partly to the
historically strong employment expansion. If real labour cost developments had been the same as those
recorded in the late 1980s, annual employment growth would have been 0.3-0.4 percentage point lower
than actually seen since mid-1997. Although most of the slowdown in employment growth is attributable
to the economic downturn, the slight increase in real labour costs since mid-2000 seems to have adversely
affected employment growth in 2001 and the first half of 2002. Likewise, the poor employment
performance in the early 1990s, mainly related to the slowdown in activity, was worsened by the strong
labour cost increase. The slow employment growth recorded in the late 1970s despite buoyant economic
growth was also likely linked to the substantial increase in real labour costs seen in this period. The
various economic equations estimated here suggest that the mean lag of employment to real labour cost
developments between 1985 and early 2002 was around 5 quarters and the median lag between 3 and 5
10 These contributions are computed other things remaining equal, i.e. supposing that the exogenous variables in the model
are not interdependent. For instance, any rise in output would lead to higher employment and then lower unemployment, and
thus, higher real wages, according to the Philips Curve. The rise in real labour costs would partly offset the initial effect of higher
output on employment.
14
ECB
Working Paper Series No. 358
May 2004
quarters (see Table 6). Moreover, it seems that the mean and median reaction lags of employment to
labour cost have not changed significantly since 1997.
3.2 Robustness of the break while considering hours worked or employment in full-time equivalents
Evidence of a structural break in employment behaviour suggests that employment developments
could also have been affected by labour market reforms or structural changes (in addition to the indirect
effect passing through labour cost moderation, such as the social security contribution cuts). A natural
candidate to explain the lower trend productivity in the late 1990s is the rising share of part-time in total
employment, meaning that an increasing proportion of those employed is working less. This development
has caused a reduction of hours worked per person of 0.25 p.p. per year on average between 1997 and
2001, according to Labour Force Survey data. As this effect combined with possible measurement
effects11 may explain around 0.3 p.p. out of 0.7 p.p. unexplained by traditional determinants12, other
structural changes or labour market reforms should have played a role in explaining the remaining 0.4 p.p.
in the change in employment behaviour. Moreover, the timing of part-time employment developments
indicates that they are unlikely to account for a break in employment in the late 1990s. Indeed, the
positive contribution of part-time jobs to total employment growth declined from the late 1990s, as shown
by the reduced difference between employment growth measured in number of persons and in full-time
equivalents (see Figure 5). The difference between employment growth measured in number of persons
and in full-time equivalents, which was around 0.4 p.p. on average between 1991 and 1998, fell to around
0.1 p.p. between 1999 and 200113. Looking further back, part-time employment developments and their
contribution to total employment growth were broadly similar to those recorded in the late 1980s, when
economic growth was much less job-intensive.
To underpin this result, the equation with break (equation 3 in Table 3) is re-estimated with
employment in terms of full-time equivalents (see Table 7). While the error correction mechanism
appears much less significant, the break in intercept is still significant at 5%. The break is of a lower
magnitude than in the equation estimated with employment in terms of persons: around 0.3 p.p. annual
employment growth in full-time equivalents has not been explained by traditional determinants between
1997 and 2001. Of course, this lower part of unexplained employment growth after 1997 reflects the
effect of part-time on recent employment growth. This result is broadly consistent with those found by
Garibaldi and Mauro (2002): increases in part-time employment in the services sector, where most part-
11 The employment performance may also have been affected by the change in employment definition in Germany
(measurement effect). The inclusion of low-paid part-time jobs in the new employment definition in Germany might have
increased euro area employment growth by around +0.1 p.p. year-on-year in 1997-2001. Indeed, these low-paid part-time jobs
were not included in employment data in the past, while this category of workers grew at a very fast pace. However, the
magnitude of this effect should be considered with considerable caution.
12 Measured as the average contribution of the residual in the equation without break over the period 1997-2001 (column 1
in Table 3).
13 While the decrease in the contribution of part-time work to net job creation was, of course, mainly accounted for by the
slower increase in the part-time employment rate (by around 2 p.p.), the relative increase in hours worked in part-time in 1999-
2001 played an additional role (by around 0.5 p.p.). The latter effect is, however, relatively weak and seems to go in the same
direction as the development in the part-time rate.
15
ECB
Working Paper Series No. 358
May 2004
time jobs were created, have been associated with increases in the overall number of jobs but most likely
also with partial crowding out of full-time jobs.
The use of full-time equivalents improves the measure of “effective” labour but does not take into
account developments in usual hours worked by full-time workers as well as changes in the number of
working days. Thus, the equation with break has also been re-estimated with employment in terms of
total hours worked (see Table 7). The series was built by Korteweg and Vijselaar (2002), using OECD
data on total hours worked in the economy, completed by Labour Force Surveys data on usual weekly
working hours. A positive break from 1997 onwards is significant at the 1% level and corresponds to a
0.4 p.p. unexplained annual growth in hours worked in the period 1997-2001. This latter result should be
taken with caution, given the fragility of working time measurements, but confirms the break in
employment behaviour at the end of the period.
3.3 Taking account of heterogeneity across countries
In order to infer that employment behaviour has changed in recent years, it is necessary to check if
the break for the euro area as a whole is broadly based across countries or if this only reflects specific
features in a very limited number of countries. In addition, the findings presented for the euro area as a
whole might be slightly affected by an aggregation bias, due for instance to changes in country weights
over time. For this purpose, fixed-effects regressions are run with a macro-panel of 21 OECD countries.
A break in employment equation, modelled as a dummy for the period 1997-2001, is tested for the EU
countries and euro area countries (see Table 8). As shown in columns 1 and 2, the break for these two
groups of countries appears fairly low and insignificant. The break for countries outside the euro area and
the EU turns out to be negative and clearly insignificant.
As shown in Table 9, regressions allowing for a break for each euro area country seem to indicate
that a group of countries (Belgium, France, Ireland, Italy, the Netherlands and Spain) have recorded a
stronger employment growth that is not fully explained by classical determinants. Indeed, the break in
these countries appears significant in all regressions (except for Ireland). It should also be noted that the
significance of the break is not strongly affected by the choice of its starting date. Although its exact
dating is somehow arbitrary, we make the break start from 1997 onwards in order to be consistent with
results for the euro area shown in section 3.1. This choice is broadly supported by country-by-country
estimates14. The break turns out to be particularly significant for Spain and France. Additional growth
recorded in these countries since 1997 varies from 0.9 percentage point in Belgium to 3.6 percentage
points in Spain. Conversely, the second group of countries (Austria, Finland, Germany, Greece and
Portugal) has not experienced any clear change in their employment pattern. In other words, in these
countries, employment growth was mostly explained by the traditional determinants in the recent years.
14 In addition to the panel approach, we also re-estimated equations (1) and (2) presented in Table 9 country by country so as to
allow for different starting dates for the break. The results for equation 2 are presented between brackets. The statistical
significance of the break is maximised when it starts in 1998 (1998) for Belgium, in 1995 (1995) for Spain, in 1998 (1998)
for France, in 1994 (1996) for Ireland, in 1998 (1999) for Italy and in 1997 (1995) for the Netherlands. However, the
magnitude and the significance of the break are not dramatically affected when we make it start in 1997 for all countries.
16
ECB
Working Paper Series No. 358
May 2004
For most OECD countries outside the euro area, the break is insignificant, with the exception of Japan,
where it is negative. The regression in column 4 of Table 8 summarises these results by testing the break
globally for both groups. In the first group, accounting for around 59% of the total euro area employment,
more than 0.7 p.p. employment growth has not been explained by traditional determinants since 1997, this
break being statistically significant. The break is negative but clearly insignificant in the second group of
countries15. These results confirm that most euro area countries have experienced positive structural
changes but not all of them.
4. What factors may account for a change in aggregate employment pattern in recent
years?
This section enters a very difficult area, trying to explain the change in employment pattern, shown
empirically in the previous section. Three aspects are investigated in this section: changes in the sectoral
composition of the euro area employment, developments in labour market institutions and the impact of
active labour market policies.
4.1 Changes in the sectoral composition of euro area employment
Compositional effect may have played a part in explaining development in aggregate employment
growth, as suggested by Marimon and Zilibotti (1998). A simple accounting exercise indicates that the
average annual growth rate of employment between 1997-2001 would have been around 0.2 percentage
point lower if the sectoral composition of employment had remained the same as in 1986-1991 (see Table
10). Indeed, the share of sectors with high employment growth (i.e. market-related services, such as trade,
repairs and financial and business services) was much higher at the start of the economic expansion of the
late 1990s than at the beginning of the boom of the late 1980s. Those sectors are characterised by a strong
economic growth, high employment intensity or both. The strong employment growth in market-related
services is broadly attributable to a very strong value-added growth. Job intensity of growth, measured by
the ratio of employment to value-added growth, appears to have been very high (1.4) in financial, real
estate renting and business services in 1997-2001, while it was higher in trade, repairs, hotels and
restaurant, transport and communication than in industry excluding construction.
Symmetrically, sectors with low or negative employment growth (such as agriculture and industry
excluding construction) had a lower weight in total employment in the late 1990s than in the previous
decade. Given that, as a first approximation, total employment growth can be computed as the sum of
sectors’ employment growth weighted by the share of each sector in total employment, the movements in
the sectors’ share might affect total growth even though there is no change in sectoral growth. Another
way to consider the compositional effect is to notice that employment growth in all sectors (except
15 A similar equation has been estimated for the EU (see column 3), leading to the same conclusion.
17
ECB
Working Paper Series No. 358
May 2004
agriculture) was lower in the period 1996-2001 than in 1986-1991, while aggregate employment growth
was broadly similar, as seen in Table 10.
4.2 The importance of labour market institutions
It was found earlier that part-time development contributed to higher employment growth in the
1990s but cannot fully explain the break in the late 1990s. The evidence of the effect of structural reforms
on macroeconomic labour market variables appears patchy in the literature, mainly due to the difficulty to
quantify and study labour market reforms at a macro-level. Moreover, most of the labour market
institution data used in this section are not available for the late 1990s. Thus, the goal of this section will
remain modest, attempting to collect first quantitative evidence by introducing labour market variables in
employment equations. While past studies mainly focused on unemployment, this section emphasises the
effect of labour market institutions on net employment creation.
Table 11 shows the panel data estimates of employment models when including annual data on
labour market institutions as collected from various sources by the OECD and Nickell and Nunziata
(2001)16. Two sets of institutions should be distinguished: those influencing both labour demand and
labour supply (job protection legislation17, total taxes on labour, unionisation18 and wage bargaining co-
ordination) and those mostly affecting labour supply (benefit replacement rates and benefit durations).
Various estimations have been carried out. Regression (1) uses the first group of institutions, while
regressions (2) to (8) also integrated the second group of institutions. Following the approach of Belot and
van Ours (2000), interactions between institutions are taken into account in regressions (4) and (6). Such
interactions are used as a robustness check and take into account the fact that similar reforms could have
different effects in different countries and comprehensive reforms are more effective than piecemeal
labour market policy19.
The equations estimated follow two slightly different specifications. Equations (1) to (4) correspond
to the traditional employment equation, estimated in section 2 for the euro area as a whole. They use total
employment (in logarithm) as the dependent variable and include also the GDP level and real labour
costs. Thus, the following general specification is estimated, where E denotes employment (either in log
or in rate), Y real GDP and w/p real labour costs (compensation per employee deflated by the GDP
deflator), the i and t are country- and time- indices and k kinds of institutions Xk are taken into account
and interact with each other:
16 For the most recent observations, see also S. Nickell, L. Nunziata and W. Ochel "Unemployment in the OECD since the
1960s. What do we know?” Bank of England, May 2002.
17 Indeed, EPL can raise insider power and therefore lower effective labour supply by reducing the wages expected by
outsiders.
18 Called also union density. This is the percentage of reported union members among wage and salaried employees.
19 From an econometric point of view, this is referred to as semi-poolable time series.
∑∑∑
=≠
++++++= K
k
K
kti
J
kj tijtikkjtikk
ti
tititi XXbXaYE 1
p
w
lnlnln
εγβαα
18
ECB
Working Paper Series No. 358
May 2004
Equations (5) and (6) refer to the specification used by Nickell et al. (2001), in which the dependent
variable is the employment rate (i.e. the ratio of total employment to population aged 15-64) and GDP
and real labour costs are replaced by country trends.
All the equations are estimated over a sample of euro area countries, as the test for poolability
suggests that the effect of explanatory variables is very different between euro area countries and the
other OECD countries. Indeed, a Chow test strongly rejects the hypothesis of common slopes across these
two groups. A possible and tentative explanation may be that the effect of institutions is stronger in
countries facing significant rigidities in their labour market, which seems to be the case for most euro area
countries. Therefore, the equations shown in Table 11 are estimated on a panel of euro area countries
only. However, the results in terms of sign and significance are not very different when including other
OECD countries in the sample. Moreover, as shown by Nunziata for the institutions (2001), there is clear
evidence of non stationarity for many of the variables used in the equations. We therefore need to test for
cointegration so as to check the absence of spurious regressions. The Madala-Wu test consists in testing
for unit roots in the residuals of the equations20. The test clearly rejected the hypothesis of residuals
having a unit root.
Several findings should be emphasised. First, the total labour tax rates21 (called also tax wedge, i.e.
employees’ and employers’ social security contributions and personal income tax as percentage of total
labour costs) appears to be significant in all equations. Its coefficient is always negative and relatively
stable, at around -0.15. Given the autoregressive term, this means that, other things being equal, a decline
of 10 percentage points in the total labour tax rate would lead in the long run to a rise of around 11% in
the level of employment (models 1 to 3) or to 7 percentage point increase in the employment rate (model
5), which represents a fairly strong effect. Moreover, total labour taxes have an additional adverse effect
when combined with a high union density.
Second, evidence appears mixed for employment protection legislation (EPL). It is found to be
negatively correlated to employment in all equations without interactions. However, it only appears
significant in models (1) to (3). When adding interaction between institutions, the effect of employment
protection legislation on its own becomes negative. Nonetheless, it has a negative impact on employment
when combined with the level of bargaining co-ordination (model 4) or unemployment benefit duration
(model 6). The latter institutions are likely to raise the bargaining power of insiders and then the
equilibrium wage, which lowers effective labour supply by reducing the prospect of the “outsiders” of
being hired. EPL may exacerbate this phenomenon of labour supply segmentation by limiting further the
ability of outsiders to compete with insiders.
20 As the test relies on the assumption of no cross-country correlation, we control for cross-country correlation by means of
time dummies in the equation.
21 This indicator is an average macroeconomic measure computed from national account data. This average tax wedge can
also be seen as the difference between the after-tax disposable labour income received by wage earners and total labour costs
borne by employers.
19
ECB
Working Paper Series No. 358
May 2004
Third, the level of bargaining coordination is found to be positively correlated with employment.
While it appears strongly significant (at a level of 1%) in models (4) and (6), it is not significant in the
other equations. Model (3) permits to reject the hypothesis of the convexity22 in the effect of bargaining
coordination, as modelled by the square of bargaining co-ordination variable. This is supported by the
economic literature (e.g. Nickell et al., 2001), which found that highly co-ordinated bargaining offset the
adverse effects of unionisation on employment.
Fourth, apart from some interactions with the tax wedge or EPL, unionisation, the unemployment
benefits replacement ratio and unemployment benefit duration are found not significant in general (or
displaying an unexpected sign in equation 6). Although the unemployment benefits replacement ratio
seems to adversely affect the employment rate, it has no impact on total employment and the effect is not
robust to the introduction of interactions.
Five, equations (4) and (6) confirm that institutions play a role, not only in isolation but also
interacting between each other. However, the significance of these interactions does not appear robust to
the specification chosen (logarithm of total employment versus employment rate), except for the joint
negative effect of total labour taxes and unionisation.
The interpretation of the findings requires much caution, as some results are not robust across the
various models estimated. Indeed, the small number of time varying observations for institutions as well
as high collinearity among institutional variables does not permit to identify precisely the impact of
individual institutions. Moreover, the role of some institutional variables such as taxation and
employment protection in determining employment has extensively been discussed in both the theoretical
and empirical literature and appears not to be clear cut. However, Daveri and Tabellini (2000) show that
higher taxes lead to higher unemployment and lower output growth. An increase in labour taxation is
likely to raise total labour costs, leading to lower employment growth. The econometric results shown in
Table 11 go in this direction. In some countries, labour tax rates are unevenly distributed among wage
earners, being particularly high for low wage earners, youth or low-skilled workers, therefore reducing
further their employability. Hiring the low skilled is all the more costly for employers as employment
protection limits the possibility of firing workers who turn out to display low productivity.
The impact of employment protection legislation on employment appears ambiguous. Bentolila and
Bertola (1990) argue that both job creation and destruction will decrease as a result of an increase in
labour adjustment costs but the resulting effect on total employment in the long run is uncertain. Bertola
(1992) suggests also that individual sectors may be affected differently by job protection, which
complicates the analysis at the aggregate level. However, Caballero and Hammour (1998) have
highlighted that a rise in firing costs may lead firms to substitute capital for labour in the medium run,
resulting in a lower job intensity of economic growth. Empirically, the evidence is mixed. Using cross-
sectional data, Nickell (1997) and Nickell et al. (2001) do not find a significant effect of employment
22 This corresponds to the hypothesis of “U-shaped curve”, presented by Calmfor and Driffil (1988): very decentralised or,
at the other extreme, very centralised wage bargaining structure would lead to a better outcome in terms of unemployment and
employment than the intermediate case of negotiation by branch.
20
ECB
Working Paper Series No. 358
May 2004
protection legislation on unemployment rates and employment rates across countries, whereas Blanchard
and Wolfers (2000) argue that higher employment protection leads to a larger effect of adverse
macroeconomic shocks on unemployment. Exploiting time-series dimension of the data, Lazear (1990)
and Scarpetta (1996) show a positive relationship between firing costs and unemployment, while Morgan
(2001) finds that employment security slows the dynamic adjustment of employment but does not
increase the number of persons employed.
4.3 The role of active labour market policies
In addition to the institutions mentioned above, active labour market policies (ALMP) may have
played a role in explaining the good employment performance in the late 1990s. Table 12 presents panel
data estimates using OECD data (database on labour market programmes). In order to take into account
that active labour market measures are likely to impact employment growth gradually and to correct for
endogeneity problems23, ALMP are computed as the share of ALMP expenditures in GDP lagged by two
years. The results are based on a euro area panel, but for most models they are found broadly similar
between euro area countries and other OECD countries, according to a Chow test on common slopes.
The findings are mixed in the sense that none of the ALMP expenditures appears statistically very
significant. However, it should also be noted that coefficient signs are consistent across models (1), (2),
(3) and (4), except for public employment training. Model (1) shows that expenditures devoted to public
employment services, labour market training and subsidised employment are not significant (at a 5%
level). In model (2), subsidies to employment have been broken down into subsidies to regular
employment in the private sector and direct job creation in the public sector. The former, which comes
down to lowering taxation rates and reducing labour costs, is positively correlated with employment
growth, although not statistically significant. Direct job creation in the public or non-profit sector does not
seem to affect future employment growth with a very low coefficient and t-statistic. In a recent study,
Algan et al. (2002) argue that job creation in the public sector crowds out private sector employment and
can even eventually lead to a decline in total employment. The attraction for public activities (positively
depending on the size of rent in the public sector and the degree of substitutability of public and private
jobs) exerts upward wage pressure in the private sector, reducing employment in this sector. Moreover,
direct job creation might contribute to increase taxes, which have distorting effects on economic activity.
In model (3), expenditures for youth have been included but are not significant at all. Their sign is not the
one expected. Model (4) is a re-estimation of model (2) using the employment rate instead of the natural
logarithm of total employment. Subsidies to regular employment in the private sector turn out to be
significant at 10% level, while the t-statistic for direct job creation is close to zero.
All in all, the results based on aggregate data are not very robust and display a low level of statistical
significance. Scarpetta (1996) confirms that some ALMP, such as job assistance, training programs and
23 It is indeed difficult to identify the causal relationship between employment and ALMP. Sluggishness in the labour
market induces mechanically an increase in ALMP, as more people become eligible. On the other hand, a high level of ALMP
may improve the employment prospects of the unemployed and increase employment.
21
ECB
Working Paper Series No. 358
May 2004
financial assistance for firm creation can stimulate employment. Nevertheless, Layard, Nickell and
Jackman (1991) emphasise that the composition of spending is as important as the level. Moreover, as
pointed out by Decressin et al (2001), ALMP tend to be ineffective when they are not focused on well-
defined beneficiaries. For example, broadly based employment subsidies may have little effect relative to
the level of expenditures because of dead-weight losses or substitution effects detrimental to non-
subsidised employment. Using country evidence, Decressin et al. (2001) conclude that the increased
employment intensity of growth is unlikely to have primarily been caused by ALMP. However, the
increase in subsidies to regular employment in the euro area private sector, which doubled as a percentage
of GDP from 1994 to 2000, might have contributed to the positive employment performance. This
argument is close to that stated earlier about the reduction in labour taxes in the euro area. Employment in
public administration increased relatively slowly in the late 1990s compared with that in other sectors,
which may also have supported employment creation in the private sector (see Algan et al., 2002).
4.4 The role of structural changes in explaining cross-country differences in recent employment
performance
The role of structural changes may be highlighted further by relating the cross-country differences
observed in the employment pattern since 1997 to changes recorded in the sectoral composition of
employment, institutions and active labour market policies in the second half of the 1990s. As mentioned
earlier in section 3.3, some countries (Austria, Germany, Greece, Finland and Portugal) do not seem to
have experienced any significant change in their aggregate employment pattern, while the others benefit
from higher than expected employment in the recent period.
Looking at the observed cross-country differences and as shown by Figures 6a, 6b and 6c, a clear
negative relationship emerges between labour tax rates and the presence of a positive break in recent
employment performance, confirming the panel results presented in section 4.2. In particular, countries
with higher than expected employment in the late 1990s experienced a decline (Ireland, Netherlands,
Spain) or at least no movement in their labour tax rate (Belgium and France), while most of the countries
which saw no significant change in their employment growth in the late 1990s faced an increase in their
tax wedge. One should also notice that the countries experiencing a rise in the tax wedge did not face
lower than expected employment owing perhaps to an offsetting effect of the concomitant loosening in
their employment protection legislation. Looking deeper into tax reforms, while there have been across-
the-board tax-cutting measures in most euro area countries in recent years, some particular attention has
been paid to reducing tax pressure at the lower end and in the middle of the income distribution (see also
ECB 2002b). The strong decline in tax rates on low-wage earners recorded in the late 1990s (by 3 p.p.),
which was mainly related to cuts in employers’ social security contribution, is indeed a natural candidate
to account for the strong employment performance in this period.
Although the other variables are less tightly linked to the presence of a positive break in employment
pattern, some interesting results have been found. The cross-country analysis confirms that the impact of
changes in the sectoral composition of employment is positively correlated to employment performance,
22
ECB
Working Paper Series No. 358
May 2004
but to a much lesser extent than tax wedge. In particular, it may have contributed to explaining the break
in employment seen in particular in France, Italy, Ireland and Spain (see Figure 6a). The strong decline in
employment protection legislation in Belgium, Italy and Spain may also partly explain the good
employment performance recorded in these countries. In line with the results found in section 4.3, Table
6c suggests that subsidies to regular employment in the private sector may also have helped in improving
employment performance in the late 1990s in Belgium, France, Italy and the Netherlands. Conversely,
Austria, Germany, Greece and Portugal may have suffered from a decline in the rate of subsidies to
private sector employment. Lastly, Belgium, Ireland, the Netherlands and, to a lower extent, Italy may
have benefited from the strong rise in the part-time employment rate, whereas Greece and Finland
suffered from relatively weak developments in part-time employment. However, the role of part-time job
developments in accounting for cross-country differences in employment performance does not appear
predominant as Germany, and to a lesser extent Portugal and Austria, also experienced a significant rise
in the part-time employment rate.
Conversely, other institution such as the share of temporary jobs, union density, benefit replacement rate,
benefit duration or most ALMP (public employment services, labour market training and direct job
creation in the public sector) do not display any evident clear link with the employment performance in
euro area countries in the late 1990s. This is again broadly in line with panel data findings reported
earlier. However and more tentatively, it is plausible that Germany and Portugal may have suffered from
the strong concomitant increase in the replacement rate and the duration of unemployment benefit. At the
other extreme, Spain which experienced the strongest break in the employment performance may have
taken advantage of the decline in benefit replacement rate and union density in addition to that in
employment protection legislation and tax wedges.
To summarise, tentative evidence seems to point to the positive impact of structural changes on
employment creation. However, the timing of the structural break (from 1997-1998 onwards according to
panel data and time series estimates) is important. The IMF (1999) argues that it is not a coincidence that
positive effects of structural reforms appear in economic upturns, even though the reforms were
implemented earlier. This argument is similar to that developed by Blanchard and Wolfers (2000),
according to whom a labour market outcome results from the interaction of both macroeconomic shocks
and institutions. The increase in “potential employment”, induced by structural changes and reflected by
higher job intensity, will raise potential output, which will require a corresponding increase in effective
demand so that reforms could translate into effective increases in output and employment. Furthermore,
following a rationale close to Rowthorn’s (1999), an increase in capital stock, which is mainly driven by
the cycle, may also be required for the positive effects of structural reforms to actually lead to create new
jobs. An alternative (and not mutually exclusive) explanation for the timing of the impact of labour
market reforms may be that many of them have been taken in the mid-1990s and may have materialised
gradually over the late 1990s (see ECB 2002a).
23
ECB
Working Paper Series No. 358
May 2004
Conclusion
The present paper aims to better understand the recent employment growth in the euro area.
Econometric estimations of labour demand show that employment growth inertia coupled with a
productivity trend and the lagged impact of both economic growth and real labour costs can largely
explain employment developments between 1970 and the mid-1990s. Moreover, relatively low increases
in real labour costs in the late 1990s compared to the 1980s certainly contributed to the good employment
performance recorded in recent years. However, employment equations estimated for the period 1970-
1996 explain only partly the strong employment growth observed between 1997 and 2001. The inclusion
of a break from 1997 onwards improves substantially the fit of the dynamic simulation. The significance
of the break seems robust, whatever the measure of employment used (employment per head, full-time
equivalents or hours worked). Moreover, most euro area countries (but not all) appear to have
experienced a break in the late 1990s.
Compositional effects, linked to the higher share of fast growing and job-intensive sectors such as
market related services in total employment in the late 1990s, is likely to have slightly raised aggregate
employment growth in recent years. Albeit difficult to show clearly, the break in employment would also
suggest that labour market reforms and/or structural changes might have played a role in the good
employment performance in the euro area24. The strong development of part-time jobs in the 1990s should
have played a positive part and lower labour tax rates should have contributed to the good employment
performance since 1997 in the euro area. More tentatively, the relaxation of job protection legislation may
have facilitated employment creation in the late 1990s. Furthermore, some active labour market policies,
such as subsidies to private employment, might also have played a positive role, although the results do
not appear very significant or robust. Conversely, most ALMP are found clearly insignificant in
explaining employment developments. It should be borne in mind that data limitations, particularly for
labour market institutions (poor time series dimension and unavailability of very recent data) and active
labour market policies (highly aggregated data), as well as the lack of robustness of some results call for
considerable caution in explaining the break in employment. The results presented in the last section on
the impact of institutions and active labour market policies illustrate the difficulty of highlighting the
effect of structural reforms at a macroeconomic level, as confirmed by numerous studies.
24 It should be noted that the compositional effects and the impact of institutions are not mutually exclusive, as structural reforms
may also boost service sector growth.
24
ECB
Working Paper Series No. 358
May 2004
References
Algan Y., P. Cahuc and A. Zylberberg (2002), “Public employment. Does it increase unemployment?” ”,
Economic Policy, April.
Arrow, K., H. Chenery, B. Minhas and R. Solow (1961), “Capital-labour substitution and economic
efficiency”, Review of Economics and Satistics, vol 43, pp. 225-250.
Baltagi, B. (1995), Econometrics of panel data, Wiley, West Sussex.
Belot, M. and J. van Ours (2000) “Does the recent success of some OECD countries in lowering their
unemployment rates lie in the clever design of their labour market reforms?”, CEPR Discussion Paper
No. 2492.
Bentolila, S. and Bertola, G. (1990) “Firing costs and labour demand: how bad is eurosclerosis?”, Review
of Economic Studies, 57, 381-420.
Bertola, G. (1992) “Labour Turnover Costs and Average Labour Demand,” Journal of Labor Economics,
10, 389-411.
Bertola, G., F.D. Blau and L.M. Kahm (2002) “Comparative analysis of labour market outcomes; Lessons
for the US from the international long run evidence”, in A. Krueger and R. Solow, eds. The Roaring
Nineties: Can full employment be sustained?, New York: Russell Sage.
Blanchard, O. and J. Wolfers (2000) “The Role of Shocks and Institutions in the Rise of European
Unemployment: the Aggregate Evidence”, Economic Journal Vol. 78(2), pp.182-187.
Brown, R.L., J. Durbin and J.M. Evans (1975) “Techniques for testing the constancy of regression
relationships over time”, Journal of the Royal statistical society, nÛSS
Caballero, R. and M. Hammour (1998). “Jobless growth: Appropriability, factor substitution and
unemployment”, Carnegie-Rochester Conference Series on Public Policy, 48, 51-99
Calmfors, L. and J. Driffill (1988) “Bargaining Structure, Corporatism and Macroeconomic
Performance”, Economic Policy, April.
Daveri, F. and G. Tabellini (2000) “Unemployment, Growth and Taxation in Industrial Countries”,
Economic Policy, April.
Decressin, J., M. Estevão, P. Gerson and C. Klingen (20001). “Job-rich growth in Europe”. Chapter 3 in
“Selected Euro-Area Countries: Rules-Based Fiscal Policy and Job-Rich Growth in France, Germany,
Italy and Spain”, Country Report No. 01/203, IMF.
Duchêne, S. and A. Jacquot (1999), “Une croissance plus riche en emplois depuis le début de la décennie?
Une analyse en comparaison internationale”, Document de travail de l’INSEE, G9901, Février.
ECB (2002a), “The composition of employment growth in the euro area in recent years”, ECB Monthly
Bulletin, November.
25
ECB
Working Paper Series No. 358
May 2004
ECB (2002b), “Labour market mismatches in euro area countries”, report prepared by the Monetary
Policy Committee of the European System of Central Banks, March.
Escriban A. and G. A. Pfann (1998), "Non linear error correction, asymmetric adjustment costs and
cointegration" Economic Modelling, 15, 197-216
European Commision (2000), “Performance of the European union Labour Market”, European Economy,
No 4.
Fagan, G., J. Henry and R. Mestre, “An Area-Wide Model (AWM) for the Euro Area”, ECB working
paper series, No. 42, January.
Garibaldi P. and P. Mauro (2002), “Employment growth. Accounting for the facts”, Economic Policy,
April.
Hamermesh D. and G.A. Pfann (1996), “Adjustment costs in factor demand”, Journal of Economic
Literature Vol. XXXIV, pp. 1264-1292.
Hamilton, J.D. (1994), Time series analysis, Princeton University Press.
IMF (1999), “Chronic unemployment in the Euro Area: Causes and Cures”, World Economic Outlook,
May.
Jackman, R., C. Pissarides and S. Savouri (1990), “Unemployment Policies”, Economic Policy, October.
Jackman, R., S. Roper (1987), “Structural Unemployment”, Oxford Bulletin of Economics and Statistics,
49, 1.
Korenman, S. and D. Neumark (2000), “Cohort crowding and youth labor markets: a cross-national
analysis.” In D. Blanchflower and R. Freeman, eds. Youth employment and joblessness in advanced
countries, NBER, Chicago University Press.
Korteweg, G. and F.W. Vijselaar (2002), “A Tale of Eight Legs on a Fly“, paper prepared for the annual
conference of the International Association for Official Statistics, available at
http://www.statistics.gov.uk/IAOSlondon2002/
Layard, R., Nickell, S. and Jackman, R. (1991), “Unemployment, macroeconomic performance and the
labour market”, Oxford University Press, Oxford.
Lazear, E. P. (1990) “Job Security Provisions and Employment”, Quarterly Journal of Economics 105.
Marimon, R. and M. Zilibotti (1998), “Actual versus virtual employment in Europe: Is Spain different?”,
European Economic Review, 42(1).
Mermet, E. (2001), “Wage Formation in Europe”, ETUI, Brussels.
Morgan, J. (2001), “The employment security and the demand for labour in Europe”, Applied Economics,
33, 1763-1774.
Morgan, J. and A. Mourougane (2001), “What can changes in structural factors tell us about
unemployment in Europe”, ECB working paper series, No. 81, October.
26
ECB
Working Paper Series No. 358
May 2004
Nickell S. (1986), “Dynamic models of labour demand”, in Handbook of Labour Economics, volume 1,
Edited by O. Ashenfelter and R. Layard, Elsevier Science Publishers, 473-522.
Nickell, S. (1997), Unemployment and labour market rigidities: Europe versus North America, Journal of
Economic Perspectives 11(3), 55-74.
Nickell, S. (1998), Unemployment: question and some answers, The Economic Journal, 108 (May), 802-
816.
Nickell, S and L. Nunziata (2000), "Employment patterns in OECD countries," CEP Discussion Papers
448, Centre for Economic Performance, LSE
Nickell, S, L. Nunziata, W. Ochel and G. Quintini (2001), "The Beveridge Curve, Unemployment and
Wages in the OECD from the 1960s to the 1990s - Preliminary Version," CEP Discussion Papers 502,
Centre for Economic Performance, LSE.
Nickell, S and L. Nunziata (2001). Labour Market Institutions Database. Centre for Economic
Performance, LSE.
Nickell, S, L. Nunziata and W. Ochel (2002), "Unemployment in the OECD since the 1960s. What do we
know?” Bank of England, May.
Nunziata, L (2001). Institutions and Wage Determination: a Multi-Country Approach. Nuffield College
Working Papers in Economics, (2001-W29), December.
OECD (1994), The Jobs Study, Paris.
OECD (1997), Employment Outlook, Paris.
OECD (1999), Employment Outlook, Paris.
Pichelmann, K. (2001), “Wage discipline in EMU. A Feature of the Early Years. Only?”, paper prepared
for the 3rd ECB Labour Market Workshop, 10-11 December 2001. “How are wages determined in
Europe?”.
Rowthorn, R. (1999), “Unemployment. Capital-Labor Substitution and Economic Growth”, IMF Working
Paper series No. 99/43, March 1, 1999
Scarpetta, S. (1996), “Assessing the role of labour market policies and institutional settings on
unemployment: A cross-country study”, OECD Economic Studies 2(26), 43-82.
27
ECB
Working Paper Series No. 358
May 2004
FIGURES AND TABLES
Table 1
Employment, activity and job intensity during economic expansions
(Average annual change %, unless otherwise indicated)
Period of economic expansion
Euro area 1969-1973 1976-1980 1986-1990 1997-2000
total employment 0.8 0.4 1.4 1.6
GDP 5.3 3.3 3.3 2.8
ratio employment growth / GDP
g
rowth 0.2 0.1 0.4 0.6
United Kingdom
total employment 0.3 0.1 1.9 1.4
GDP 2.6 1.8 3.3 2.9
ratio employment growth / GDP
g
rowth 0.1 0.1 0.6 0.5
US
total employment 1.6 3.0 2.1 1.6
GDP 2.9 3.7 3.2 4.2
ratio employment growth / GDP
g
rowth 0.5 0.8 0.6 0.4
Sources: European Commission, Eurostat, OECD and author’s calculations.
28
ECB
Working Paper Series No. 358
May 2004
Table 2
OLS Estimates of employment equations, controlling for some labour supply effects (age structure and
working age population).
Equation (1)
without break
(one step-
estimation)(1)
Equation (2)
Without break
(one step-
estimation)(1)
Equation (3)
without break
(one step-
estimation)(1)
Equation (4)
without break
(one step-
estimation)(1)
Estimation period 1970Q1-2002Q2 1970Q1-2002Q2 1970Q1-2002Q2 1970Q1-2002Q2
Coefficients
(
t-statistics
)
¨OQ(t-1 0.39
(
4.1
)
0.41
(
2.5
)
0.32
(
3.3
)
0.30
(
3.1
)
¨OQ<t-1 0.06
(
1.84
)
0.07
(
1.9
)
0.06
(
1.8
)
0.06
(
2.0
)
¨ln Y t-2 0.06
(
1.59
)
0.07
1.7
)
0.06
(
1.6
)
0.06
(
1.8
)
¨ln (w/p) t-5 -0.047
(
-1.24
)
-0.038
(
-0.9
)
-0.028
(
-0.76
)
-0.020
(
-0.56
)
Error correction mechanism -0.0456
(
-1.82
)
-0.030
(
-2.5
)
-0.062
(
-2.4
)
-0.061
(
-2.6
)
ln (w/p) t-1
(long-term relationship) -0.021
(
-1.8
)
-0.017
(
-2.8
)
-0.027
(
-2.1
)
-0.026
(
-2.4
)
Time trend t-1
(long-term relationship) -0.0000
(
-1.48
)
0.0000
(
1.5
)
-0.0002
(
-3.17
)
-0.0002
(
-2.0
)
Intercept
(long-term relationship)
ln (Working age population) (2)
(long-term relationship) -0.021
(
-1.036
)
0.014
(
0.67
)
ln (age structures) (3)
(long-term relationship) 0.00062
(
0.17
)
0.0053
(
1.59
)
Dummy 1975Q2 -0.0035
(
3.2
)
-0.0035
(
-2.6
)
-0.003
(
-3.1
)
-0.003
(
-2.8
)
Dummy 1984Q1 -0.0052
(
-9.1
)
-0.0056
(
-8.7
)
-0.005
(
-9.4
)
-0.005
(
-10.9
)
Dummy 1992Q3 -0.00426
(
-7.2
)
-0.0044
(
-7.3
)
-0.004
(
-7.8
)
-0.004
(
-8.1
)
Intercept 0.096
(
0.47
)
2.1
(
3.2
)
-0.36
(
-1.7
)
0.22
(
-2.9
)
Break (1997Q1-2002Q2) 0.0026
(
4.7
)
0.0027
(
4.8
)
Main statistics
R20.652 0.656 0.681 0.687
Adjusted R20.618 0.622 0.648 0.653
Durbin Watson 2.05 2.03 2.0 1.99
(1) As some heteroskedasticity has been detected., the t-statistics presented in this column are computed with the White heteroskedasticity-
consistent standard errors.
(2) The working-age population is defined as those aged 15-64 (OECD usual definition).
(3) The relative size of youth/prime age population, defined as the size of population aged 15-24 over the population aged 25-54.
29
ECB
Working Paper Series No. 358
May 2004
Table 3
OLS Estimates of employment equations with and without break
Equation (1)
without break
(two step-
estimation)
Equation (2)
Without break (1)
(one step-
estimation) (1)
Equation (3)
With break (1)
(two step-
estimation) (1)
Equation (4)
with break (1)
(one step-
estimation) (1)
Estimation period 1970Q1-2002Q2 1970Q1-2002Q2 1970Q1-2002Q2 1970Q1-2002Q2
Coefficients
(
t-statistics
)
¨OQ(t-1 0.498
(
7.11
)
0.409
(4.74)
0.319
(
4.69
)
0.319
(
3.34
)
¨OQ<t-1 0.0561
(
1.81
)
0.069
(
2.08
)
0.055
(
2.08
)
0.057
(
1.71
)
¨ln Y t-2 0.0476
(
1.48
)
0.068
(
1.90
)
0.055
(
1.898
)
0.053
(
1.54
)
¨ln (w/p) t-5 -0.0636
(
-2.55
)
-0.99
(-1.90)
-0.0392
(
-1.22
)
-0.033
(
-0.88
)
Error correction mechanism -0.027
(
-1.40
)
-0.031
(-1.67)
-0.065
(
-2.93
)
-0.067
(
-2.91
)
ln (w/p) t-1
(long-term relationship) -0.448 -0.550
(-1.59)
-0.448 -0.407
(
-2.3
)
Time trend t-1
(long-term relationship) -0.003 -0.0022
(-1.30)
-0.003 -0.00212
(
-2.95
)
Intercept
(long-term relationship) -3.30 -3.30
Dummy 1975Q2 -0.0042
(
-2.24
)
-0.0033
(
-2.9
)
-0.0036
(
-4.73
)
-0.0034
(
-3.2
)
Dummy 1984Q1 -0.0061
(
-3.42
)
-0.0056
(
-12.8
)
-0.0052
(
-13.21
)
-0.0053
(
-12.0
)
Dummy 1992Q3 -0.004
(
-2.08
)
-0.004
(
-7.64
)
-0.004
(
-8.63
)
-0.0041
(
-7.6
)
Intercept 0.0005
(
1.15
)
-0.113
(
-
1.69
)
0.0004
(
1.21
)
-0.21
(
-2.72
)
Break (1997Q1-2002Q2) 0.0021
(
4.53
)
0.0021
(
3.78
)
Main statistics
R20.632 0.654 0.680 0.681
Adjusted R20.607 0.621 0.654 0.649
Diagnostic tests
Durbin Watson 2.13 2.05 1.99 1.99
LM (1) 0.16
(
0.20
)
0.61
(0.43)
0.02
(
0.88
)
0.016
(
0.90
)
LM (4) 7.79
(
0.10
)
7.13
(0.13)
2.77
(
0.60
)
2.82
(
0.59
)
ARCH(1) 0.38
(
0.54
)
1.54
(0.21)
3.10
(
0.08
)
3.22
(
0.07
)
WHITE 17.14
(
0.19
)
37.46
(0.003)
25.6
(
0.03
)
39.3
(
0.003
)
Normality 3.88
(
0.14
)
0.63
(0.73)
0.77
(
0.68
)
0.76
(
0.68
)
RESET(1) 5.74
(
0.02
)
7.76
(0.005)
3.32
(
0.07
)
3.00
(
0.09
)
CHOW(3) 14.57
(
0. 48
)
8.70
(0.894)
5.14
(
0.99
)
5.46
(
0.99
)
(1) As some heteroskedasticity has been detected., the t-statistics presented in this column are computed with the White heteroskedasticity-
consistent standard errors.
(2) Asymptotic tests are presented.
(3) Predictive failure test over the period 1999Q1-2002Q2.
30
ECB
Working Paper Series No. 358
May 2004
Figure 1
Total employment: dynamic simulation without break
observed versus fitted
level in thds
Figure 2
Dynamic contributions to the annual growth rate of total employment
(from equation 1 in Table 3, without break)
-2
-1
0
1
2
3
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
GDP Real labour cost Technology trend Residuals Employment
110000
115000
120000
125000
130000
135000
1971Q3
1972Q3
1973Q3
1974Q3
1975Q3
1976Q3
1977Q3
1978Q3
1979Q3
1980Q3
1981Q3
1982Q3
1983Q3
1984Q3
1985Q3
1986Q3
1987Q3
1988Q3
1989Q3
1990Q3
1991Q3
1992Q3
1993Q3
1994Q3
1995Q3
1996Q3
1997Q3
1998Q3
1999Q3
2000Q3
2001Q3
actual employment fitted (estimated over 1971- 1996) fitted (estimated over 1971-2002)
31
ECB
Working Paper Series No. 358
May 2004
Table 4
Sequential selection of different possible breaks (from 1997 onwards) in the standard employment equation
Breaks in the equation The least significant break
(removed in the following steps) P-value of t-
statistics Adjusted R-
squared
Step 1
(allowin
g
for breaks
in all variables)
¨OQ(t-1 ¨OQ<t-1 ¨OQ<t-2 ¨ln(w/p)t-5
ECM ln(w/p)t-1 TIME-TREND
INTERCEPT
¨OQ(t-1 0.76 0.636
Step 2 ¨OQ<t-1 ¨OQ<t-2 ¨ln(w/p)t-5 ECM
lnw/pt-5 ln(w/p)t-1 TIME-TREND
INTERCEPT
Ln(w/p)t-1 0.68 0.639
Step 3 ¨OQ<t-1 ¨OQ<t-2 ¨ln(w/p)t-5 ECM
TIME-TREND INTERCEPT
ECM
(lnEt-1-lnYt-1)0.76 0.636
Step 4 ¨OQ<t-1 ¨OQ<t-2 ¨ln(w/p)t-5
TIME-TREND INTERCEPT
TIME-TREND 0.13 0.646
Step 5 ¨OQ<t-1 ¨OQ<t-2 ¨ln(w/p)t-5
INTERCEPT
¨ln(w/p)t-5 0.18 0.645
Step 6 ¨OQ<t-1 ¨OQ<t-2 INTE RCEPT ¨OQ<t-2 0.07 0.647
Step 7 ¨OQ<t-1 INTERCEPT ¨OQ<t-1 0.31 0.647
Final step INTERCEPT INTERCEPT 0.002 0.649
Note: This table is based on the re-estimation of equation 1 of Table 3, but using a one-step ECM estimation procedures and
allowing for breaks in all variables (Step 1). The least significant break is removed sequentially in the following steps. The
results are obtained with OLS regressions with standard errors corrected for possible heteroskedasticity and autocorrelation
(Newey-West method).
Figure 3
Dynamic simulation with a break since 1997
observed versus fitted ; level in thds
110000
115000
120000
125000
130000
135000
1971Q3
1972Q3
1973Q3
1974Q3
1975Q3
1976Q3
1977Q3
1978Q3
1979Q3
1980Q3
1981Q3
1982Q3
1983Q3
1984Q3
1985Q3
1986Q3
1987Q3
1988Q3
1989Q3
1990Q3
1991Q3
1992Q3
1993Q3
1994Q3
1995Q3
1996Q3
1997Q3
1998Q3
1999Q3
2000Q3
2001Q3
actual employment fitted two steps fitted one step
32
ECB
Working Paper Series No. 358
May 2004
Figure 4
Forecasting performance between 1999Q2 and 2002Q2
127000
129000
131000
133000
1999Q1
1999Q2
1999Q3
1999Q4
2000Q1
2000Q2
2000Q3
2000Q4
2001Q1
2001Q2
2001Q3
2001Q4
2002Q1
2002Q2
actual employment forecast without break forecast with break
Table 5
Lagged reaction of employment to GDP growth (in quarters)
period of estimation 1970Q1 - 1996Q4 1970Q1 – 2002Q2 1985Q1 – 1996Q4 1985Q1 – 2002Q2
50%* lag 4.6 (4.3) 5.0 (4.6) 4.0 (6.9) 4.4 (4.9)
80%* lag 12.0 (11.3) 11.3 (10.9) 6.9 (13.5) 10.3 (7.4)
mean lag 7.8 (7.4) 7.5 (8.1) 5.9 (8.8) 8.6 (5.6)
Note: lagged reactions given by an equation including all lags of endogenous and exogenous variables up to 5 quarters, even those non-
significant. Between brackets, lagged reactions given by an equation including the significant lags only (see equation 1 in Table 3).
* Number of quarters needed to reach 50% (80%) of the long-term effect.
Table 6
Mean lag of employment to real labour costs (in quarters)
period of estimation 1970Q1 1996Q4 1970Q1 2002Q2 1985Q1 1996Q4 1985Q1 2002Q2
50%* lag 4.9 (4.8) 5.1 (5.3) ** (**)3.6 (4.9)
80%* lag 10.2 (10.1) 8.9 (12.7) ** (**)5.7 (7.4)
mean lag 7.7 (7.3) 7.2 (8.6) ** (**)5.0 (5.2)
Note: lagged reactions given by an equation including all lags of endogenous and exogenous variables up to 5 quarters, even those non-
significant. Between brackets, lagged reactions given by an equation including the significant lag only, see equation 1 in Table 3).
* Number of quarters needed to reach 50% (80%) of the long-term effects.
** Non interpretable: long term elasticity has a positive sign, which is contrary to the theory.
Figure 5
0.0
0.1
0.2
0.3
0.4
0.5
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
Constructed from labour force survey (Spring)
Constructed from quarterly eurostat data (annual average)
Contribution of part-time jobs to total employment growth (p.p)
33
ECB
Working Paper Series No. 358
May 2004
Table 7
Time series estimates with alternative measures of employment for euro area
Full-time equivalents Hours worked(1)
Estimation period 1983Q4-2002Q1 1981Q3-2001Q4
Coefficients
(t-statistics)
¨OQ(t-1 0.564
(7.26) 0.530
(4.33)
¨OQ<t-2 0.105
(2.73) 0.072
(1.53)
¨ln Y t-3 0.061
(1.74)
¨ln (w/p) t-2 -0.067
(-1.65)
¨ln (w/p) t-5 -0.071
(-2.01)
Error correction
mechanisms -0.033
(-1.37) -0.0719
(-2.77)
ln (w/p) t-1
(long-term relationship)
-0.320 -0.148
Time trend (-1)
(long-term relationship)
-0.0037 -0.005
Intercept
(long-term relationship)
-2.964 3.489
Dummy 1984Q1 -0.0051
(-3.51) -0.0035
(-7.58)
Dummy 1992Q3 -0.0049
(-3.16) -0.005
(-9.87)
Intercept -0.00019
(-0.52) 0.516
(2.76)
Break (1997Q1-2002Q2) 0.00083
(2.01) 0.00155
(2.69)
Main statistics
R20.796 0.717
Adjusted R20.770 0.690
Diagnostic tests(2)
Durbin Watson 2.04 2.04
LM (1) 0.08
(0.77) 0.09
(0.77)
LM (4) 7.73
(0.10) 3.46
(0.48)
ARCH(1) 0.35
(0.55) 8.64
(0.003)
WHITE 18.75
(0.13) 27.77
(0.004)
Normality 2.32
(0.31) 20.03
(0.0005)
RESET(1) 0.58
(0.45) 0.49
(0.48)
CHOW(3) 9.22
(0.82) 9.08
(0.77)
(1) As some heteroskedasticity has been detected., the t-statistics presented in this column are computed with the White heteroskedasticity-
consistent standard errors. Euro area data on total hours worked are coming from Korteweg and Vijselaar (2002).
(2) Asymptotic tests are presented, as the hypothesis of normal residuals is not always fulfilled (e.g. equation in hours worked).
(3) Predictive failure test over the period 1999-2001
34
ECB
Working Paper Series No. 358
May 2004
Table 8
Testing a break in employment equation from 1997 with a panel of 21 OECD countries, 1977-2001
Variables (1) (2) (3) (4)
GDP growth 0.568 0.567 0.563 0.565
(15.09) (15.05) (14.98) (15.03)
GDP growth (-1) 0.233 0.233 0.232 0.232
(5.62) (5.63) (5.60) (5.61)
GDP growth (-2) 0.114 0.112 0.110 0.110
(2.98) (2.94) (2.87) (2.89)
Real labour cost -0.252 -0.252 -0.249 -0.249
(-9.24) (-9.27) (-9.13) (-9.15)
Real labour cost (-1) -0.036 -0.036 -0.032 -0.032
(-1.32) (-1.33) (-1.17) (-1.17)
Break in euro area countries 0.304
(1.24)
Break in countries outside euro area 0.048 0.047
(0.19) (0.19)
Break in EU countries 0.320
(1.48)
Break in Non-EU -0.088 -0.090
(-0.30) (-0.30)
Break in EU countries with faster employment growth from 1997 0.745
(2.25)
Break in EU countries with unchanged employment growth from 1997 0.018
(0.06)
Break in euro area countries with faster employment growth from 1997 0.741
(2.24)
Break in euro area countries with unchanged employment growth from 1997 -0.194
(-0.55)
Number of observations 523 523 523 523
Number of countries 21 21 21 21
R squared 0.504 0.504 0.506 0.509
Poolability of euro area countries (0.000) (0.000) (0.000) (0.000)
Absolute value of t-statistics in parentheses
Data sources: OECD, economic outlook. Author’s calculations.
Note: The equations are estimated by fixed-effects (within) regression. GDP and w/p have been instrumented in order to overcome endogeneity
problems. The list of instruments is the contemporaneous export of goods and services and real labour costs lagged by two quarters. Euro area
countries with faster employment growth from 1997 onwards are Belgium, France, Ireland, Italy, Spain and Netherlands. Euro area countries
with unchanged employment growth from 1997 are Austria, Germany, Greece, Finland and Portugal. EU countries with faster employment
growth from 1997 onwards are Belgium, France, Ireland, Italy, Spain and Netherlands. EU countries with unchanged employment growth from
1997 are Austria, Denmark, Germany, Greece, Finland, and Portugal, Sweden and the UK. These groups were constituted on the basis of the
sign and significance of country break in preliminary regressions shown in Table 9. The poolability of restriction between the euro area
countries and the other OECD countries is rejected by the Chow test on common slopes. This might generate a bias in parameter estimates,
although the estimation may gain in efficiency when pooling. However, we keep pooling all OECD countries, as the purpose of the table is
primarily to test the significance of the break with different groups of countries.
35
ECB
Working Paper Series No. 358
May 2004
Table 9
Testing a break in employment equation for each country from 1997 with a panel of 21 OECD countries,
1977-2001
Country
(euro area countries in italics) (1) (2) (3)
Austria 0.533 0.375 0.375
(1.38) (1.07) (1.07)
Belgium 0.953 0.753 0.753
(2.17)** (1.82)* (1.82)*
Canada 0.676 0.320 0.320
(0.68) (0.39) (0.39)
Switzerland -0.236 -0.269 -0.269
(-0.32) (-0.56) (-0.56)
Germany -1.006 -0.848 -0.848
(-0.21) (-0.22) (-0.22)
Denmark 0.219 -0.099 -0.099
(0.28) (-0.17) (-0.17)
Spain 3.589 3.264 3.264
(2.94)*** (3.46)*** (3.46)***
Finland 2.009 1.218 1.218
(1.47) (1.39) (1.39)
France 1.240 1.189 1.189
(2.99)*** (3.71)*** (3.71)***
United Kingdom 0.851 0.674 0.674
(0.92) (1.12) (1.12)
Greece -0.376 -0.891 -0.891
(-0.33) (-0.97) (-0.97)
Ireland 2.987 1.554 1.554
(2.19)** (1.48) (1.48)
Italy 1.055 1.288 1.288
(1.88)* (3.34)*** (3.34)***
Japan -0.906 -0.469 -0.469
(-2.47)** (-2.03)** (-2.03)**
Netherlands 1.460 0.919 0.919
(2.21)** (2.08)** (2.08)**
Norway 0.376 0.219 0.219
(0.50) (0.43) (0.43)
New Zealand -0.401 -0.319 -0.319
(-0.33) (-0.31) (-0.31)
Portugal 1.156 0.906 0.906
(1.30) (1.35) (1.35)
Sweden 1.139 1.141 1.141
(1.11) (1.53) (1.53)
United States -0.495 -0.829 -0.829
(-0.62) (-1.25) (-1.25)
Observations 523 523 523
Number of countries 21 21 21
Absolute value of t statistics in parentheses.
* significant at 10%,; **significant at 5%; *** significant at 1%.
Note: The break is modelled by a break in the intercept (additional growth). The dependent variable is annual employment growth, while the
regressors are GDP growth and real compensation per employee. Various lag specifications have been used regressors lagged by 2 years
(equation 1) so as to avoid endogeneity problems; contemporaneous regressors plus their lagged values by 1 and 2 years (equation 2) as lags of
3 years and more turn insignificant; lagged regressors by 1 and 2 years (equation 3). The three equations are estimated by generalised least
squares with country fixed effects, allowing for heteroskedastic errors and common-across-group first order serial correlation.
36
ECB
Working Paper Series No. 358
May 2004
Table 10
Impact of sectoral composition on total employment growth
Industrial sectors Average annual growth rate
of employment Share in total employment Decomposition of employment growth
1997-2001
1986-1991 1997-2001 1986-1991 1997-2001 Value-added
growth Employment
intensity 1
Agriculture -3.7 -1.5 7.1 4.5 0.8 -1.8
Industry excluding construction 0.5 0.2 24.8 20.2 2.3 0.1
Construction 1.7 0.6 7.3 7.1 0.2 2.6
Trade, repairs, hotels and restaurant, transport
and communication 2.0 1.9 24.0 25.1 3.7 0.5
Financial, real estate renting and business
services 4.9 4.8 9.8 13.6 3.5 1.4
Public administration, education, health and
other services 2.4 1.4 27.0 29.5 1.3 1.0
Total 1.6 1.5 100 100 2.6 0.6
Total with the sectoral structure of 1986-
1990. 1.6 1.3
Data sources: ESA95 national account, Eurostat. OECD, STAN databases. Author’s calculations.
1 Ratio employment growth / value-added growth. In other words, this is the empirical elasticity of employment to value added.
37
ECB
Working Paper Series No. 358
May 2004
Table 11
Panel data models of employment including labour market institutions
Dependent variable: total employment (level in logarithm/employment rate)
(Euro area countries(1). 1960-1997)
(1) (2) (3) (4) (5) (6)
Dependent variables Employment
(in log) employment
(in log) employment
(in log) employment
(in log) employmen
t
rate (%) employment
rate (%)
Macroeconomic variables
Ln (Employment) (-1) 0.862 0.856 0.857 0.788
(19.55) (18.68) (18.71) (17.52)
Employment rate (-1) 0.854 0.771
(31.17) (23.71)
Ln (GDP) -0.003 0.008 0.009 0.003
(-0.77) (0.21) (0.23) (0.08)
Ln (Real compensation per employee) -0.05 -0.045 -0.041 -0.083
(-1.80) (-1.49) (-1.33) (-2.65)
Institutions
Employment protection -0.014 -0.013 -0.014 0.113 -0.003 0.010
(-2.16) (-1.93) (-2.06) (3.71) (-0.81) (0.92)
Total taxes on labour -0.137 -0.147 -0.147 -0.257 -0.101 -0.161
(-3.09) (-2.92) (-2.91) (-1.87) (-4.34) (-2.79)
Unionisation -0.036 -0.040 -0.035 0.101 -0.005 0.128
(-1.52) (1.58) (-1.34) (0.91) (0.42) (2.92)
Bargaining coordination 0.007 -0.024 0.079 0.003 0.027
(1.85) (-0.60) (3.28) (1.61) (2.85)
Bargaining coordination squared 0.007
(U-shape curve hypothesis) (0.78)
Unemployment benefits replacement ratio 0.005 0.008 0.112 -0.024 0.014
(0.19) (0.28) (1.32) (-2.06) (0.37)
Unemployment benefits duration 0.003 0.003 0.006 0.001 0.023
(0.35) (0.48) (0.25) (0.33) (2.38)
Interactions between institutions
Employment protection* Total taxes on labour -0.034 0.051
(-0.47) (1.96)
Employment protection* Bargaining coordination -0.050 -0.003
(-4.72) (-0.82)
Employment protection* Unionisation 0.063 -0.025
(1.85) (-1.68)
Employment protection* Unemployment benefits duration -0.001 -0.021
(-0.05) (-2.40)
Employment protection* Unemployment benefits replacement ratio -0.053 -0.005
(-0.96) (-0.24)
Total taxes on labour* Unionisation -0.087 -0.044
(-2.66) (-3.43)
Total taxes on labour* Bargaining coordination 0.074 -0.009
(2.20) (-0.56)
Unemployment benefits replacement ratio* Unemployment benefits duration -0.026 -0.014
(-0.50) (-0.54)
Intercept 2.379 2.268 2.242 3.393 0.149 0.153
(3.70) (3.41) (3.38) (5.18) (5.30) (4.90)
Country specific trends No No No No Yes Yes
Number of observations 289 289 289 289 330 330
Number of countries 9 9 9 9 10 10
Cointegration (Stationarity of residuals - Maddala-Wu test) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Poolability of euro area countries (2)
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Sources: OECD economic outlook, Nickell and Nunziata (2001). Author’s calculations.
(1) Countries included: Austria, Belgium, Finland, France, Germany, Ireland, Italy, Netherlands and Spain. Portugal is included in equation (5)
and (6),
(2) Chow test on common slopes between the euro area and other OECD countries (Australia, Canada, Japan, Norway, Sweden, Switzerland, UK,
US, Denmark and New Zealand).
Note: The equations are estimated by generalised least squares allowing for heteroskedastic errors and common-across-group first order serial
correlation. Each equation contains country dummies and time dummies (fixed effects). Nickell (1981) shows that the bias of dynamic (with
lagged dependent variable among the regressors) fixed effects models with first order serial correlation is o(1/T) and therefore becomes less
important as T grows. Moreover, Judson and Owen (1999) showed that the fixed effect estimator performs as well as many alternatives when
T=30 (see Nunziata, 2001). Employment, GDP and real labour cost are included with a lag of two years to tackle endogeneity problems (reverse
causality). The panel is unbalanced as some data are missing for the 1960s and 1970s.
38
ECB
Working Paper Series No. 358
May 2004
Table 12
Panel data models of employment including active labour market policies
Dependent variable: total employment (level in logarithm/employment rate)
(Unbalanced euro country panel (1), 1988-2001)
(1) (2) (3) (4)
Variables Employment (in
log) employment (in
log) employment (in
log) employment rate
(%)
Ln (Employment) (-1) 0.904
(17.19)
Employment rate (-1) 0.873 0.865 0.877
(12.84) (11.58) (11.35)
Ln (GDP) (-2) 0.068 0.090 0.073
(0.82) (0.97) (0.70)
Ln (Real compensation per employee) (-2) -0.050 -0.038 -0.011
(-0.83) (-0.48) (-0.13)
Public employment services (-2) -0.022 -0.059 -0.058 0.009
(-0.37) (-0.84) (-0.76) (0.29)
Labour market training (-2) (2) -0.049 -0.055 -0.053 -0.021
(-1.77) (-1.83) (-1.56) (-1.91)
Tobal subsidised employment (-2) (a+b) 0.010
(0.78)
Subsidies to regular employment in the private sector (-2) (a) 0.031 0.026 0.020
(1.10) (0.69) (1.76)
Direct job creation (public or non-profit) (-2) (b) 0.009 0.010 0.000
(0.50) (0.44) (0.01)
Youth measures (-2) (3) -0.026
(-0.47)
Constant 0.779 0.381 0.402 0.067
(0.73) (0.31) (0.27) (1.63)
Country specific trends No No No Yes
Number of observations 138 125 117 134
Number of countries 10 10 10 10
Cointegration (Stationarity of residuals - Maddala-Wu test)
(0.012) (0.000) (0.000) (0.259)
Poolability of euro area countries (4)
(0.502) (0.602) (0.107) (0.000)
Data sources: OECD economic outlook. Active labour market policies (ALMP) data stems from the OECD database on labour market
programmes. Author’s calculations.
(1) Countries included: Austria, Belgium, Finland, France, Germany, Ireland, Italy, Netherlands, Spain and Portugal.
(2) Training for employed and unemployed adults and those at risk
(3) Measures for unemployed and disadvantaged youth and support of apprenticeship and related forms of general youth training
(4) Chow test on common slopes between the euro area countries and other OECD countries (Australia, Canada, Japan, Norway, Sweden,
Switzerland, UK, US, Denmark and New Zealand).
Note: The equations are estimated by generalised least squares allowing for heteroskedastic errors and common-across-group first order serial
correlation. Each equation contains country dummies and time dummies (fixed effects). Nickell (1981) shows that the bias of dynamic (with
lagged dependent variable among the regressors) fixed effects models with first order serial correlation is o(1/T) and therefore becomes less
important as T grows. Moreover, Judson and Owen (1999) showed that the fixed effect estimator performs as well as many alternatives when
T=30 (see Nunziata, 2001). Employment, GDP and real labour cost are included with a lag of two years to tackle endogeneity problems (reverse
causality). In order to take into account that active labour market measures are likely to impact employment gradually and to correct for possible
endogeneity, ALMP (expressed as the share of ALMP expenditures in GDP) are estimated with a two year lag. The panel is unbalanced as some
data are missing for the 19960s and 1970s.
39
ECB
Working Paper Series No. 358
May 2004
Figure 6a
Relation between break in employment pattern and changes in employment structure
Change in the sectoral composition of employment from the late 1980s
y = 4.23x - 0.073
R
2
= 0.170
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.1 0.2 0.3 0.4 0.5
Contribution to annual e mployme nt growth 1997-2001
Break in employment growth
rate after 1997
AUT
FRA
SP IRE
GER
ITA
NLD
GRC FIN
BEL
Data sources: New Cronos, Eurostat. The break in employment growth since 1997 corresponds to the panel estimation (model 1) reported in
Table 9. Non significant breaks are set at zero.
Figure 6b
Relation between break in employment pattern and changes in institutions
Labour tax rate
y = -0.3052x + 0.972
R
2
= 0.4816
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
-6 -4 -2 0 2 4 6
Change in p.p. 1995-1999
Break in employm ent growth
rate after 1997
BEL
GRC
AUT
POR
SP
NLD
GER
FIN
FRA
IRE
Union density
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
-5 -4 -3 -2 -1 0
Change in p.p. 1995-1998
Break in employment growth
rate after 1997
AUT
POR
FRA
SP
IRE
GER
ITA
NLD
GR
C
FIN
Benefit replaceme nt ra te
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
-4-2024681012
Change in p.p. 1995-1999
Break in employm ent growth
rate after 1997
FIN
BEL
AUT POR
FRA
SP
NLD
GER
ITA
GRC
IRE
Benefit duration
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
-25 -20 -15 -10 -5 0 5 10 15 20 25
Change in p.p. 1995-1999
Break in employment growth
rate after 1997
FIN
BEL
GRC
AUT
POR
FRA
SP
NLD
GER
ITA
IRE
Employment protection legislation
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
-4 -3 -2 -1 0
Change in indicator betw een average 1988-1995 and 1998
(sourc e: Nickell, Nunz iata, Ochel-2002)
Break in emplo yment growth
rate after 1997
BEL
AUT
POR
SP
NLD
GER FIN
FRA
ITA
IRE
Part-time employment
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
-2.0 -1.0 0.0 1.0 2.0 3.0 4.0 5.0 6.0
Change in part-time e mployment rate (p.p.) 1996-2001
Break in emp loyment grow th
rate after 1997
FIN BEL
GERAUT
POR
FRA NLD
ITA
GRC
IRE
SP
40
ECB
Working Paper Series No. 358
May 2004
Figure 6b (continued)
Relation between break in employment pattern and changes in institutions
Temporary employment
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
-6 -4 -2 0 2 4 6 8 10 12
Change in tempor ary em ploym ent rate (p.p.) 1996-2001
Break in employment growth
rate after 1997
FIN BEL
GER
AUT
POR
FRA
NLD
ITA
IRE
SP
GRC
Data sources: OECD. Nickell and Nunziata (2001), Nickell, Nunziata and Ochel (2002). Labour Force Surveys (Eurostat). The break in
employment growth since 1997 corresponds to the panel estimation (model 1) reported in Table 9. The results for bargaining co-ordinations are
not displayed given the absence of any significant changes in most countries in the late 1990s.
Figure 6c
Relation between break in employment pattern and changes in active labour market policies
Public employment services
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
-0.14 -0.12 -0.1 -0.08 -0.06 -0.04 -0.02 0 0.02 0.04
Change in expe nditure as a % of GDP 1995-1999
Break in employm ent growth
rate after 1997
FIN
BEL
GRC AUT POR
FRA
SP
NLD
GER
ITA
Labour market training
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
-0.15 -0.1 -0.05 0 0.05 0.1
Change in expen diture as a % o f GDP 1995-1999
Break in employment growth
rate after 1997
FIN
BEL
GRC
AUT POR
FRA
SP
NLD
GER
ITA
Subsidies to regular employment in the private sector
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
-0.05 0 0.05 0.1 0.15 0.2 0.25 0.3
Change in expendit ure as a % of GDP 1995-1999
Break in employment growth
rate after 1997
FIN
FRA
GRC AUT
POR
ITA
SP
NLD
GER BEL
Direct job creation in the public sector
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
-0.3 -0.2 -0.1 0 0.1 0.2 0.3
Change in expen diture as a % of GDP 1995-1999
Break in employment growth
rate after 1997
FIN BEL
GER AUT
POR
FRA
SP
NLD
ITA
Data sources: OECD. The break in employment growth since 1997 corresponds to the panel estimation (model 1) reported in Table 9.
41
ECB
Working Paper Series No. 358
May 2004
ANNEX
Detailed results of the estimation of a standard employment equation for the euro area. (Addendum
to section 2.3)
The two-step estimation without break, called equation 1 in the rest of the text and reported in column 1
of Table 3, yields similar results to the one-step estimation (column 2 of Table 3). The residuals are
normal and there is no serial correlation. The null hypothesis of no serial correlation is not rejected by
Ljung-Box Q test. This is confirmed by the Breusch-Godfrey Lagrange multiplier test (with one and four
lags) at 5%. Heteroskedasticity is rejected with the White test ð DW 7KH HTXDWLRQ HVWLPDWLRQ
results from a “general-to-specific” approach. The chosen length of the lag distribution, i.e. 5, is that
associated with the lowest value of both the Akaike Information Criterion and the Schwarz Criterion. The
existence of the cointegration relation, captured by the ECM term, is established according to two
methods. First, we check that the residual of the long-run relationship is stationary with both Augmented
Dickey-Fuller and Phillips-Perron Tests. We use the asymptotic critical values tabulated by Phillips and
Ouliaris, which should be applied to residuals from spurious cointegrating regressions (Hamilton, 1994).
Second, we use the Johansen Test based on the estimation of a VECM. Both trace statistic and maximum
eigen value statistics confirm the existence of a cointegration relationship between GDP, real labour cost
and employment.
The equation contains time dummies, encompassing specific time effects. Even though the choice of
dummies and that of the specific quarters are somewhat arbitrary, their inclusion increases substantially
the fit of the equation, while having an economic meaning. The first dummy (1975Q2) may partly capture
adverse effects of the first oil shock in the mid-1970s, such as the deterioration in employers’ confidence
and the rise in economic uncertainty. The second dummy (1984Q1) might refer to the negative impact on
employers’ expectation of the tightening fiscal and monetary policy in France, implemented after the
short-lived and unsuccessful expansionary macroeconomic policy conducted in 1981-1982. The third
(1992Q3) is related to the strong and temporary increase in value added in the German building sector,
caused by public subsidies in the context of German Reunification. Those few quarterly dummies only
capture part of those effects, but correct for the strongest outliers. Conversely, dummies capturing
asymmetry of labour demand across the economic cycle (e.g. negative dummy when the level of GDP is
decreasing over four quarters) turn out not to be significant.
Equation 1 allows us to determine the impact of the traditional determinants on employment growth. The
long-term elasticity of real labour costs is 0.44 over the period 1971Q3-2002Q2 and 0.34 when estimated
over 1971Q3-1996Q4. As a robustness check, we obtained a broadly similar elasticity, 0.55 and 0.40 for
the two periods mentioned earlier, when estimating equation (1) in one step, allowing for joint estimation
of the long-run effects with dynamic effects. A Cobb-Douglas specification has been tested and clearly
rejected. In such a frame, the long-term elasticity of employment to real labour costs is (minus) unity and
there is no time trend. First, the error correction mechanism, which is equal to the logarithm of real unit
42
ECB
Working Paper Series No. 358
May 2004
labour costs with a Cobb-Douglas production function25, turns out to be clearly non stationary with the
Dickey-Fuller and Phillip-Perron unit root tests. Moreover, a simple panel data estimate of labour demand
indicates that the long-term elasticity26 of employment to real labour costs is -0.35 and constraining it to -
1 is largely rejected by a Wald test (see Table 13 below). The long-term elasticity of employment to GDP
is estimated to be 0.97 and constraining it to unity is largely accepted. All in all, the results of the euro
area equation are confirmed using a panel data approach.
Table 13
Panel data estimates of employment in euro area countries
Dependent variable: annual employment growth
Explanatory variables expressed in annual growth
(12 euro area countries. 1977-2001)
employment (1) (2)
emp (-1) 0.03
(0.04)
gdp 0.78
(11.06) 0.78
(11.1)
gdp (-1) 0.16
(1.93) 0.15
(1.80)
gdp (-2) 0.03
(0.39) 0.03
(0.53)
w/p -0.35
(-8.3) -0.35
(-8.34)
w/p(-1) -0.04
(-0.87) -0.03
(-0.07)
intercept -1.86
(-2.7) -0.05
(-2.00)
Observations 262 262
R-squared (within) 0.62
R-squared (between) 0.43
R-squared (overall) 0.59
Unit long-term GDP elasticity:
c(gdp)+c(gdp/p(-1)+ c(gdp/p(-2))=1 0.08 0.08
Long-term real labour cost elasticity constrained to
minus unity: c(w/p)+c(w/p(-1))=-1 577.6 514.85
Arellano-Bond test on autocorrelation of order 2 N(0,1)=0.37
Sagan test
Global significance 326.18 419.6
Sources: OECD economic outlook. Author’s calculation.
Note: Equation (1) is estimated by fixed-effects (within) regression and contains time dummies. As a robustness check, equation (2) is estimated
with variables in first difference by instrumental variables using the dynamic panel estimator package (DPD) developed by Arellano and Bond
(1991), which derives a Generalised Method of Moments estimator. The Sargan test of over-identifying restrictions, which follows a chi-squared
with 426 degrees of freedom, is clearly accepted with a p-value of 0.99. This means that the instruments (in excess of the regressors) are valid.
The R squared is not reported for equation (2) as it is not comparable to that of equation (1) given equation (2) refers to first-differences.
Equations (1) and (2) contain time dummies. For equations 1 and 2, GDP and w/p have been instrumented in order to overcome endogeneity
problems. The list of instruments is the contemporaneous export of goods and services and contemporaneous investments. As the explanatory
variables are expressed in annual growth rates, the time trend becomes an intercept. The panel is unbalanced (data for Luxembourg and
Portugal are available from the early 1990s only). Regressions start in 1977, as some variables, especially the instruments, are not available
before for many countries.
25 The error correction mechanism may be rewritten as: ECM=lnE –lnY +ln(w/p)=ln((w/p)/(Y/E))=ln(ULC/p), where ULC
means nominal unit labour costs.
26 The long-term elasticity is the sum of the contemporaneous elasticity and that of all significant lags.
43
ECB
Working Paper Series No. 358
May 2004
Concerning the stability of the equation, evidence is mixed. The coefficients appear very stable up to
1997 but vary henceforth. The Chow forecast test with diverse break points does not allow for rejecting
the hypothesis of stability. The CUSUM test (Brown, Durbin, and Evans, 1975), based on the cumulative
sum of the recursive residuals, confirms that the equation is broadly stable over the estimation period. As
shown by Figure 1, dynamic simulations from equation (1) estimated over 1971Q3-1996Q4 appear
satisfactory in the 1970s, 1980s and early 1990, but underestimate the magnitude in employment variation
in the late 1990s. Conversely, the dynamic simulation obtained from equation (1) estimated over the full
period is fairly mediocre, especially from 1991, which might signal some instability at the end of the
period, as confirmed by the recursive estimates of the coefficient.
The equation appears unstable at the end of the estimation period. The recursive estimates of the
coefficients are very stable up to 1997 but vary henceforth. As shown by Figure 1, dynamic simulations
from equation (1) of Table 3 estimated over 1971Q3-1996Q4 appear satisfactory in the 1970s, 1980s and
early 1990, but underestimate the magnitude in employment variation in the late 1990s. Conversely, the
dynamic simulation obtained from equation (1) estimated over the full period is fairly mediocre,
especially from 1991, which seems to confirm the instability at the end of the period.
44
ECB
Working Paper Series No. 358
May 2004
302 “Deposit insurance, moral hazard and market monitoring” by R. Gropp and J. Vesala, February 2004.
303 “Fiscal policy events and interest rate swap spreads: evidence from the EU” by A. Afonso and
R. Strauch, February 2004.
304 “Equilibrium unemployment, job flows and inflation dynamics” by A. Trigari, February 2004.
305 “A structural common factor approach to core inflation estimation and forecasting”
by C. Morana, February 2004.
306 “A markup model of inflation for the euro area” by C. Bowdler and E. S. Jansen, February 2004.
307 “Budgetary forecasts in Europe - the track record of stability and convergence programmes”
by R. Strauch, M. Hallerberg and J. von Hagen, February 2004.
308 “International risk-sharing and the transmission of productivity shocks” by G. Corsetti, L. Dedola
and S. Leduc, February 2004.
309 “Monetary policy shocks in the euro area and global liquidity spillovers” by J. Sousa and A. Zaghini,
February 2004.
310 “International equity flows and returns: A quantitative equilibrium approach” by R. Albuquerque,
G. H. Bauer and M. Schneider, February 2004.
311 “Current account dynamics in OECD and EU acceding countries – an intertemporal approach”
by M. Bussière, M. Fratzscher and G. Müller, February 2004.
European Central Bank working paper series
For a complete list of Working Papers published by the ECB, please visit the ECB’s website
(http://www.ecb.int).
312 “Similarities and convergence in G-7 cycles” by F. Canova, M. Ciccarelli and E. Ortega,
February 2004.
313 “The high-yield segment of the corporate bond market: a diffusion modelling approach
for the United States, the United Kingdom and the euro area” by G. de Bondt and D. Marqués,
February 2004.
314 “Exchange rate risks and asset prices in a small open economy” by A. Derviz, March 2004.
315 “Option-implied asymmetries in bond market expectations around monetary policy actions of the ECB”
by S. Vähämaa, March 2004.
45
ECB
Working Paper Series No. 358
May 2004
321 “Frequency domain principal components estimation of fractionally cointegrated processes”
by C. Morana, March 2004.
322 “Modelling inflation in the euro area” by E. S. Jansen, March 2004.
323 “On the indeterminacy of New-Keynesian economics” by A. Beyer and R. E. A. Farmer, March 2004.
324 “Fundamentals and joint currency crises” by P. Hartmann, S. Straetmans and C. G. de Vries, March 2004.
325 “What are the spill-overs from fiscal shocks in Europe? An empirical analysis” by M. Giuliodori
and R. Beetsma, March 2004.
326 “The great depression and the Friedman-Schwartz hypothesis” by L. Christiano, R. Motto and
M. Rostagno, March 2004.
327 “Diversification in euro area stock markets: country versus industry” by G. A. Moerman, April 2004.
328 “Non-fundamental exchange rate volatility and welfare” by R. Straub and I. Tchakarov, April 2004.
329 “On the determinants of euro area FDI to the United States: the knowledge-capital-Tobin's Q framework,
by R. A. De Santis, R. Anderton and A. Hijzen, April 2004.
330 “The demand for euro area currencies: past, present and future” by B. Fischer, P. Köhler and F. Seitz, April 2004.
331 “How frequently do prices change? evidence based on the micro data underlying the Belgian CPI” by
L. Aucremanne and E. Dhyne, April 2004.
332 “Stylised features of price setting behaviour in Portugal: 1992-2001” by M. Dias, D. Dias
and P. D. Neves, April 2004.
316 “Cooperation in international banking supervision” by C. Holthausen and T. Rønde, March 2004.
317 “Fiscal policy and inflation volatility” by P. C. Rother, March 2004.
318 “Gross job flows and institutions in Europe” by R. Gómez-Salvador, J. Messina and G. Vallanti, March 2004.
319 “Risk sharing through financial markets with endogenous enforcement of trades” by T. V. Köppl, March 2004.
320 “Institutions and service employment: a panel study for OECD countries” by J. Messina, March 2004.
333 “The pricing behaviour of Italian firms: New survey
evidence on price stickiness” by
S. Fabiani, A. Gattulli and R. Sabbatini, April 2004.
334 “Is inflation persistence intrinsic in industrial economies?” by A. T. Levin and J. M. Piger, April 2004.
335 “Has eura-area inflation persistence changed over time?” by G. O’Reilly and K. Whelan, April 2004.
336 “The great inflation of the 1970s” by F. Collard and H. Dellas, April 2004.
337 “The decline of activist stabilization policy: Natural rate misperceptions, learning and expectations” by
A. Orphanides and J. C. Williams, April 2004.
46
ECB
Working Paper Series No. 358
May 2004
338 “The optimal degree of discretion in monetary policy”
by S. Athey, A. Atkeson and P. J. Kehoe, April 2004.
339 “Understanding the effects of government spending on consumption” by J. Galí, J. D. López-Salido
and J. Vallés, April 2004.
340 “Indeterminacy with inflation-forecast-based rules in a two-bloc model” by N. Batini, P.Levine
and J. Pearlman, April 2004.
341 “Benefits and spillovers of greater competition in Europe: A macroeconomic assessment” by T. Bayoumi,
D. Laxton and P. Pesenti, April 2004.
342 “Equal size, equal role? Interest rate interdependence
between the euro area and the United States” by
M. Ehrmann and M. Fratzscher, April 2004.
343 “Monetary discretion, pricing complementarity and
dynamic multiple equilibria” by R. G. King
and A. L. Wolman, April 2004.
344 “Ramsey monetary policy and international relative prices” by E. Faia and T. Monacelli, April 2004.
345 “Optimal monetary and fiscal policy: A linear-quadratic approach” by P. Benigno and M. Woodford, April 2004.
346 “Perpetual youth and endogenous labour supply: a problem and a possible solution” by G. Ascari and
N. Rankin, April 2004.
347 “Firms’ investment decisions in response to demand and price uncertainty” by C. Fuss
and P. Vermeulen, April 2004.
348 “Financial openness and growth: Short-run gain, long-run pain?” by M. Fratzscher and M. Bussiere, April 2004.
349 “Estimating the rank of the spectral density matrix” by G. Camba-Mendez and G. Kapetanios, April 2004.
350 “Exchange-rate policy and the zero bound on nominal interest rates” by G. Camba-Mendez
and G. Kapetanios, April 2004.
351 “Interest rate determination in the interbank market” by V. Gaspar, G. P. Quirós and
H. R. Mendizábal, April 2004.
352 “Forecasting inflation with thick models and neural networks” by P. McNelis and
P. McAdam, April 2004.
353 “Towards the estimation of equilibrium exchange rates for CEE acceding countries: methodological
issues and a panel cointegration perspective” by F. Maeso-Fernandez, C. Osbat and B. Schnatz, April 2004.
354 “Taking stock: monetary policy transmission to equity markets” by M. Ehrmann
and M. Fratzscher, May 2004.
355 “Production interdependence and welfare” by K. X. D. Huang and Z. Liu, May 2004.
47
ECB
Working Paper Series No. 358
May 2004
356 “Developing a euro area accounting matrix: issues and applications” by T. Jellema, S. Keuning,
P. McAdam and R. Mink, May 2004.
357 “Seasonal adjustment and the detection of business cycle phases” by A. M. Mir, and
D. R. Osborn, May 2004.
358 “Did the pattern of aggregate employment growth change in the euro area in the
late 1990s?” by G. Mourre, May 2004.
48
ECB
Working Paper Series No. 358
May 2004
