Content uploaded by Jose Enrique Garcilazo
Author content
All content in this area was uploaded by Jose Enrique Garcilazo on Aug 05, 2014
Content may be subject to copyright.
BNL Quarterly Review, no. 228, March 2004.
Unemployment, inequality and
the policy of Europe: 1984-2000
JAMES K. GALBRAITH and ENRIQUE GARCILAZO
1. Introduction
Unemployment happens to individuals. But the unemployment rate is a
matter of place. And places are nested inside larger places. The local has
properties the nation may not share. The nation has characteristics that may
not apply to the continent. In an integrated economy, the forces that oper-
ate on unemployment rates may extend over many horizons, from the near
neighborhood to the entire world.
Yet the literature on unemployment in Europe tends to concentrate
on national characteristics and national unemployment rates. The predisposi-
tion is to blame unemployment on labor market ‘rigidities’ – and then to
search for particular culprits, generally in the fields of national unemploy-
ment insurance, job protections and wage compression. Periodic move-
ments to reform national labor markets sweep aside the careful qualifica-
tions found in empirical work such as Nickell (1997) and Blanchard and
Wolfers (1999), and presuppose that greater wage flexibility is the estab-
lished cure for European unemployment. Neither local conditions nor the
influence of economic policy at the continental level play important roles in
the policy debate.
In a recent paper, Baker et al. (2003) provide a comprehensive review
of the national-institutions approach to explaining European unemploy-
ment. They find only one robust result, namely that coordinated collective
––––––––––
The University of Texas at Austin, The Lyndon B. Johnson School of Public Affairs,
Austin (USA); e-mails: Galbraith@mail.utexas.edu; garcil@uts.cc.utexas.edu.
* We thank Ken Flamm, David Howell, Richard Freeman and two referees for very
useful advice. We thank also the Ford Foundation and the Carnegie Scholars Program for
financial support.
BNL Quarterly Review
4
bargaining and (perhaps) union density are associated with less unemploy-
ment in Europe. Of course, this interesting finding is inconsistent with the
rigidities framework. So far as macroeconomic policy is concerned, while a
handful of lonely voices argue that interest rates and growth rates dominate
the determination of unemployment in Europe, these too tend to root the
relevant decision-making at the national level (e.g. Palley 2004). Meanwhile
the higher policy discussion accepts that European policy – especially
monetary policy – mainly influences the price level, leaving unemployment
to be governed by market forces and national institutions.
In this paper, we try a different approach. Instead of the nation, our
smallest unit of analysis is the region. Data are generally available for up to
159 regional entities across Europe, embedded within 13 countries. We
specify just four regional ‘labor market’ variables that, we find, account
significantly for the variation in regional unemployment rates. Then the
panel structure permits us to measure national fixed effects, and so to
identify those countries with characteristics that affect unemployment rates
after controlling for regional conditions. Next, the panel structure permits
us to identify time effects, whose pattern gives a picture of the influence of
transnational forces, such as the integration of Europe and the effect of
European macro and monetary policies. In this way we allow the data to
separate for us the influences of factors operating at the regional, national,
and international or continental levels.
We identify two regional factors that influence the demand for labor.
First is the strength of economic growth at any given time – an obvious
determinant of construction and investment jobs, and a consequence of the
local effects of macroeconomic policies and regional fiscal assistance. The
second is a measure, which we constructed, of the average wage rate of the
region relative to the average for Europe as a whole. Our thinking is that
regions with higher average wages should tend to have stronger tax bases,
more public employment and also more open (and therefore taxed) em-
ployment in services.
On the supply side, we also identify two factors. The first is the rela-
tive size of the population of very young workers – an obvious measure of
the difficult-to-employ. The second is a measure of the inequality of the
wage structure. To acquire this measure, we construct, for the first time, a
panel of European inequalities at the regional level, comparable both across
countries and through time.
Our hypothesis that regional pay inequalities should be placed on the
supply side of the labor market is an innovation. It is more conventional to
treat local wage rates as the product of supply and demand, while begging the
Unemployment, inequality and the policy of Europe: 1984-2000 5
question of whether these forces operate at the regional, national or higher
levels. Instead, in this analysis we take the regional wage structure as a
datum facing individual workers. We consider that this datum affects how
long they choose to search for employment. The greater the differential
between high and low-paid jobs in the local setting, the longer a rational
person will hold out for one of the better jobs, accepting unemployment if
necessary.
This theoretical position is well-known in neoclassical development
economics, going back to a classic article by Harris and Todaro (1970),
which treats the urban-rural pay differential as part of an incentive to
migrate from the countryside to the cities, despite the presence of urban
unemployment. The general concept, that inequality creates an incentive to
search, has not been applied to Europe or to any developed-country setting
so far as we know. But there is no compelling reason why it should not be.
In practice, we find that pay inequality is a strong determinant especially of
cross-sectional variation in European unemployment, and the positive
coefficient is consistent with the Harris-Todaro conjecture.
Once regional conditions have been accounted for, our fixed-effects
model finds few significant differences in unemployment among larger
countries. The only substantial large-country fixed effects are for the UK (a
negative shift) and Spain (a positive shift). However, large negative shifts
are found for a number of smaller countries, which have much lower
unemployment rates than our model would otherwise predict. The coun-
tries for which this is true are widely separated and appear to have little in
common apart from the fact that they are small. We will present some
hypotheses below that may help account for this phenomenon.
Finally, we replicate the estimates for sub-populations, including men,
women and very young workers. We find significant differences in the
unemployment experiences of different sub-populations: the very young as
against older workers, and men as against women. As a broad rule, it ap-
pears that the less migratory a population, the higher its unemployment rate
and the larger the effect of local labor market conditions on unemploy-
ment.
The time effects are striking for all population groups. They show a
sharp rise in unemployment common to all regions beginning in 1993. This
is an interesting break-point in view of the introduction of the Maastricht
Treaty on European Union at the start of that year. The effect continues
through the 1990s, and suggests that a substantial part of European excess
unemployment – generally between two and three percentage points –
BNL Quarterly Review
6
reflects policy conducted at the European level since the Union. In this regard,
the monetary policy of the European Central Bank and the convergence
criteria for the euro come to mind as leading suspects.
2. Theoretical considerations
Our hypothesis is that unemployment at the local level is governed princi-
pally by four factors: two each on the demand and on the supply sides. On
the demand side, the growth rate of effective demand and activity strongly
conditions the availability of jobs; in periods of strong growth construction
and investment jobs are notably abundant.
But so too does relative income. Richer places offer more employ-
ment of all kinds, whether in the public sector (because they have more tax
revenue) or in the private services sectors (because they have more discre-
tionary private income). In poor regions surplus labor is more likely to
work, if at all, in the cash economy and to report itself as unemployed.
On the supply side, labor force demography clearly matters. Young
people are hard to employ and to keep employed. So much is uncontrover-
sial.
Our other argument is that regions with more equal pay structures will,
other things equal, experience less unemployment. Since this is contrary to
the standard view, it deserves a full explanation.1
A half century ago Simon Kuznets (1955) argued that inequality
would rise in the early stages of economic development and transition to
industrial growth. New urban centers were places of concentrated income
and wealth. It was the differential between incomes in these places and those
in the countryside that would become significant as cities grew, and only
decline later as the proportion of the population remaining in the country-
––––––––––
1 One might suppose the causation to run the other way: that regional pay inequality
would be simply a positive function of local unemployment rates. But while this is possible,
two considerations suggest that it is not predominantly the case. First, unemployment rates
vary much more than inequality measures over time. The effect of inequality on unemploy-
ment is therefore mainly cross-sectional (places with higher inequality experience higher
unemployment on a chronic basis). Second, part of the greater inequality observed in a
regional pay structure is due to the scarcity of decently-paid middle-range jobs, and not
exclusively to larger pay differentials per se, though in practice both may contribute. There is
no compelling reason in neoclassical theory why higher unemployment rates should produce
a gap in employment in the middle of the pay scale, as opposed to the bottom of it.
Unemployment, inequality and the policy of Europe: 1984-2000 7
side shrank. This was the most significant single factor behind Kuznets’
inverted-U curve.
In 1970 John Harris and Michael Todaro offered a model capturing
these characteristics, in a paper aimed mainly at development economists.
In their model, workers migrate from a low-marginal-product rural sector
to cities where minimum wages are imposed, and accept a high probability
of sustained unemployment in exchange for a low probability of getting
one of those jobs and enjoying the resulting rise in income. The equilibrium
condition is that the expected value of the gain be just equal to cost in-
curred in leaving rural employment, and this condition entails substantial
equilibrium unemployment. From this, a positive relationship between
urban/rural pay inequality and equilibrium unemployment emerges.
While Harris and Todaro focused on East Africa, consider how their
argument might apply in modern Europe. Modern advanced societies have
an elite group of knowledge and finance workers, a core of manufacturing
workers, and a large reservoir of workers in the services. Access to knowl-
edge and finance jobs is restricted by cartels and credentialing. The same is
not true for manufacturing workers, who nevertheless enjoy wage premi-
ums due to industry-specific labor rents. Services workers with few skills
enjoy few such advantages, and the pay in the services sector is largely set
by social minimums, which are governed in substantial part by political
decision-makers. Services workers are like the earlier generation of farm
workers in many relevant economic respects, and they may be considered a
reserve army of the underemployed.
So long as the differential between service wages and manufacturing
wages is fairly small, or if it is possible to search for better jobs while work-
ing, services workers will not abandon current employment to seek for
better. But on the other hand, if there are large differentials and obstacles
to on-the-job search, they will do so. In that event, measured unemploy-
ment will rise. As in Harris and Todaro, equilibrium local unemployment is
a positive function of local pay inequalities.
Supply and demand at the regional level do not exhaust the possible
sources of variation in unemployment. Labor market policies, and to some
extent the rules for measuring who is unemployed and who is not, are set at
the national level. These factors may be expected to introduce some variation
in unemployment rates between countries.
Our analysis does not attempt to sort out the particular institutional
factors behind differences in national unemployment rates, once local
conditions have been controlled for. Rather we seek to establish how much
BNL Quarterly Review
8
of the observed differences in unemployment can be attributed to national
differences, and for which countries these differences are important. The
introduction of country fixed effects permits this measurement to be
carried out easily.
Finally, the factors that work on the continental (or, indeed, global)
level need to be considered. Where a rise or decline in unemployment is
common across the full spectrum of regions of Europe, it is reasonable to
attribute it to policies and institutional changes emanating at the European
level (or some higher level, such as the effect of changing global economic
conditions). Time fixed effects capture these move-
ments. Since Europe for the past twenty years has been a laboratory for
economic integration and rule-bound policy-making, it will be very interesting
to see what pattern emerges, in relation to three specific events especially: the
Single European Act (1987), the Maastricht Treaty on European Union
(1993) and the introduction of the euro (1999).
In our model, several significant forms of unemployment are subject
to policy control and so are involuntary in Keynes’ (1936) meaning. These
include, particularly, the growth rate, the degree of pay inequality at the
regional level, and the contribution of European-level economic policy and
institutional change to European unemployment. Other factors, including
population structure and national institutional characteristics, would have to
be considered as sources of frictional or even of voluntary unemployment.
So the analysis should be of considerable interest in sorting out the empiri-
cal relevance of these old theoretical questions.
Our framework may be applied to different subsets of the popula-
tion, which can be expected to have different degrees of responsiveness to
the forces at work. Women move in and out of work more than men.
Young people face an inevitable transition from school to work. The choice
for these groups is what job to aim for? A worker who once accepts a low-
wage job may be typed as low-productivity, and cannot make the transition
to higher pay as easily as a worker who has never been employed at all. For
this reason, young people especially have an incentive to resist taking bad
employment. Youth unemployment in unequal regions should therefore be
expected to be an especially serious problem.
Migration is a reinforcing consideration. Certain countries have larger
emigrant populations than others. Within any given population, older male
workers tend to be more mobile than either women or the very young. If
acceptable jobs are not available in their immediate surroundings, they can
be expected to search elsewhere, disappearing from the regional unem-
Unemployment, inequality and the policy of Europe: 1984-2000 9
ployment statistics. For this reason, the unemployment of less mobile
subpopulations should show higher sensitivity to regional conditions, and
less mobile subpopulations should generally experience higher unemploy-
ment rates, than more mobile subpopulations.
3. Data and model
Use of the region rather than the nation as the unit of geographic analysis
has two distinct advantages. The first is that regions are more numerous:
159 in ‘Old Europe’ alone. The second is that regions are also more homo-
geneous: the standard deviation of population size for regions is merely a
tenth of what it is for countries. Table 1 gives this information.
We propose a model in which regional unemployment rates depend
on four regional factors: pay inequality (+), the youth proportion in the
population (+), economic growth rate (–) and relative wages (–). The first
two of these factors influence the supply of unemployed labor; the second
two affect the demand for labor (or supply of jobs). In addition, we expect
to find national differences in average unemployment rates and variations
in unemployment common to all regions in Europe. These may be meas-
ured by country fixed effects and time fixed effects, respectively.
TABLE 1
POPULATION DIFFERENTIALS FOR NATIONS
AND REGIONS IN EUROPE, 1984-2000
Variable No.
observations Mean Standard
deviation Minimum Maximum
Nations 1169 28,128 25,164 355.9 80,759.6
Regions 1853 2,306 2,556 22.5 17,663.2
Population in thousands.
The main empirical innovation in the present paper lies in nearly
comprehensive measures of pay inequality calculated across broad eco-
nomic sectors at the level of European regions – the 159 entities over 17
years (1984-2000).
We employ the between-groups component of Theil’s T statistic to
measure pay inequality. The methodology has been proposed in Conceição
and Galbraith (2000) and in Conceição, Galbraith and Bradford (2001),
building on Theil (1972). Theil’s T statistic can be expressed as follows:
BNL Quarterly Review
10
1yiyi
T = n
n
Σ
i = l µ⋅ log [µ](1)
where i
y denotes the income of an individual region indexed by i, n is the
number of individuals in the population and µ is the average income.
One of the most attractive features of this statistic is its decomposi-
tion property. As long as a distribution of income and a distribution of
individuals are grouped into mutually exclusive and completely exhaustive
groups, overall inequality can be decomposed into a between-groups com-
ponent and a within-groups component. The between-groups measure is
derived from group means for payroll and group population weights; the
within-groups measure is a weighted average of the Theil inequality index
for each group. Formal expressions for both components are included in
Appendix 2; this study takes advantage of the fact that, under some very
general conditions, the dynamics of overall inequality can be captured using
only the between-groups component.
This between-sectors calculation provides a new source of informa-
tion on the relative inequality of the pay structures in the regions of
Europe, and because the sector categories are standardized, the measures
are comparable across regional (and national) boundaries as well as through
time. Our data are from Eurostat’s REGIO data base (http://www.eu-
datashop.de). We use compensation of employees (e2rem95) and employ-
ment (e2empl95) for 159 regional entities among 16 major economic
sectors. Regions are classified by NUTS level 2 except for the regions of
Germany and United Kingdom, where data are only available at NUTS
level 1. A list of economic sectors and regions is included in Appendix 3.
The relative wage variable (RelWage) is the ratio of each region’s av-
erage payroll per worker relative to the average payroll per worker of
Europe as a whole. Average payroll is derived by dividing total compensa-
tion of employees by employment for each year. The remaining regional
variables – growth of GDP and proportion of the population under 24
years of age – are constructed conventionally from REGIO.
We now turn to a regression analysis, with the following reduced
form, two-way fixed-effects model:
UN = a +B1Theil +B2 RelWage + B3 GDPG +
B4 PopUn24 + Di Country + DjTime
where:
Unemployment, inequality and the policy of Europe: 1984-2000 11
UN = regional unemployment rate
Theil = pay inequality across sectors for each region
RelWage = average regional wages relative to the European average
GDPG = growth rate of GDP at the regional level
PopUn24 = proportion of the regional population under 24 years of age
Country = dummy to capture fixed country effects
Time = dummy to capture fixed time effects.
The model can be fitted for all of Europe using annual data from
1984 to 2000, with full information for a total of 1465 region-year observa-
tions. The coefficients on the regional variables are reported in Table 2.
Different models reflect estimates for the whole population and its compo-
nent parts: men, women, older and younger workers (ages greater or less
than 25 years). We report a linear version of the model, a log-log version
gave similar results and is not reported.
TABLE 2
COEFFICIENT ESTIMATES: LINEAR MODEL (1984-2000)
Total Male Female < 25 Yrs > 25 Yrs
Beta Pvalue Beta Pvalue Beta Pvalue Beta Pvalue Beta Pvalue
Theil 4.97 0.04 3.22 0.13 6.80 0.04 11.97 0.03 4.08 0.04
PopUn24 57.02 0.00 50.58 0.00 76.46 0.00 112.32 0.00 38.04 0.00
RelWage –7.08 0.00 –4.95 0.00 –9.91 0.00 –6.37 0.00 –7.43 0.00
G-GDP –4.48 0.02 –5.67 0.00 –2.35 0.39 –6.30 0.17 –4.69 0.00
R¯2 0.61 0.59 0.65 0.62 0.58
No. observa-
tions
1465
1465
1465
1465
1465
All the variables have the correct sign and all but three are significant
at conventional significance levels. Coefficients are systematically higher
for less-mobile populations, except that GDP growth rates matter less for
women – no surprise. R2 is in the range of 60% for all models.
Higher growth at the local level reduces unemployment. Larger
numbers of young people are associated with higher unemployment. The
data on unemployment and inequality at the level of European regions
support our hypothesis of a positive relationship between these two vari-
BNL Quarterly Review
12
ables, though at a moderate significance level. In areas with high levels of
pay inequality and high numbers of young people, the two effects would
appear to combine to yield significantly higher unemployment rates.
Inequality across Europe (measured by the RelWage variable) also ap-
pears to affect local unemployment rates. If the regression were taken
literally, it would imply that reduction in the inequality of incomes across
Europe would reduce unemployment in the poor countries. But at the same
time it would increase it in the rich countries. Therefore this result is am-
biguous in policy terms.
The regional variables taken together play a considerable role in the
explanation of variance, but each level of analysis – regional, national and
European – has a role to play. Table 3 provides measures of the variance
explained (for unemployment of all workers) when the model is specified
without fixed effects, with one-way fixed effects, and with two-way fixed
effects. Coefficient estimates on the regional variables are also shown; these
are notably stable except that the effect of GDP growth is to some extent
absorbed by the introduction of country and time effects.
TABLE 3
ANALYSIS OF VARIANCE EXPLAINED UNDER DIFFERENT SPECIFICATIONS*
Region only Region & country Region & time All variables
Beta Pvalue Beta Pvalue Beta Pvalue Beta Pvalue
Theil 4.03 0.18 4.81 0.04 5.39 0.09 4.97 0.04
PopUn24 50.20 0.00 48.64 0.00 54.23 0.00 57.02 0.00
RelWage –2.82 0.00 –6.81 0.00 –2.21 0.00 –7.08 0.00
G-GDP –11.83 0.00 –8.56 0.00 –9.49 0.00 –4.48 0.02
Regional X X X X
Country X X
Time X X
R¯2 0.16 0.57 0.21 0.61
* Dependent variable is Total Unemployment.
It turns out that country fixed effects are relatively unimportant for
large countries, with two exceptions. Taking France (with the closest to
average unemployment for the period) as the base case and plus or minus
3% as the threshold, only Spain has much higher unemployment ceteris
paribus than one would otherwise expect. In the UK, on the other hand,
unemployment is lower than otherwise expected. Germany, with a positive
Unemployment, inequality and the policy of Europe: 1984-2000 13
fixed effect just over 3%, is a borderline case; most of the German fixed
effect is surely due to the special circumstances following reunification.2
Apart from this, neither the large countries nor Scandinavia have
large differences in unemployment rates apart from those captured by the
regional variables. Whether the Spanish and UK cases can be traced to
particular causes is a matter for later research; we would want to investigate
closely the effect of the cash economy in Spain and that of credit institu-
tions in the UK. But neither value can be attributed to Spanish wage rigidity
or British flexibility, since the inequality of pay structures is already taken
directly into account at the regional level.
There are, however, large negative fixed effects for small countries
(Austria, Ireland, Portugal, Greece and, to a lesser extent, the Netherlands).
Figure 1 provides a map of the country fixed effects; Table A1 (in the
Appendix) presents the coefficient estimates. This effect may possibly be
explained in some cases by the existence of large emigrant populations. The
Portuguese in France are absent from the labor force measured in Portugal
and therefore do not figure in Portuguese unemployment.
Austria is a more difficult case to explain. But the Austrian result may
be due to strategic wage-setting, with Austrian workers close substitutes for
Germans in competing sectors, but cheaper. In an exploration reported in
Appendix 6 we find that Austrian wages are indeed systematically lower than
German on average in manufacturing, but the sector averages are actually
higher than German in non-traded sectors. Similarly, Irish wages are lower
than British; this could help account for the explosion of jobs that brought
Irish unemployment down so sharply in the late 1990s. Austrian and Irish
wages are set substantially by central bargaining, and it appears that in these
countries wage competitiveness may be concentrated where it is useful.3
FIGURE 1
EUROPEAN UNEMPLOYMENT: ALL WORKERS COUNTRY FIXED EFFECTS
––––––––––
2 There is also an interesting negative effect for youth unemployment in Germany,
which could be picking up the effects of the apprentice system.
3 We thank Richard Freeman and David Howell for jointly making the suggestion that
we compare Austrian to German wages.
BNL Quarterly Review
14
In Figure 2 we present the time effects associated with the two-way
panel. These estimates show a striking increase in the pan-European com-
ponent of the unemployment rate from 1993 to the end of the decade,
rising to a peak value of 4.6 points above the 1985 baseline in 1994, and
settling above 2 full percentage points for most of the rest of the decade.
This provides, in our view and based solely on the coincidence of timing, a
very succinct measure of the employment penalty associated with the
events of 1992, notably the Maastricht Treaty and its implementation. (The
European Exchange Rate Mechanism also collapsed in 1992. But Gordon
(1999) pins the responsibility for rising European unemployment at this
time on the fiscal tightening required by the Maastricht Treaty.) On a
brighter note, excess youth unemployment in Europe has been reduced
sharply since 1997 if these measures are correct.4 Overall, it seems possible
that the fixing of exchange rates and the introduction of the euro in 1999
had a good effect, as the pan-European component of unemployment
declined toward the end of the decade. Table A2 in the Appendix reports
the time effects and their significance levels.
FIGURE 2
EUROPEAN UNEMPLOYMENT – EUROPEAN TIME EFFECTS
––––––––––
4 Richard Freeman suggests a link to large increases in university enrollment, especially
in Spain. We are looking for evidence on this conjecture.
Fixed effect
–11 - –5
–4
–3-3
4-5
Unemployment, inequality and the policy of Europe: 1984-2000 15
4. Implications for unemployment policy in Europe
These results, so different from those implied by the standard view, should
be treated with caution. Much work remains to be done to establish the
general validity of the models advanced here, and to corroborate specific
explanations here suggested. Nevertheless, we feel that the hierarchical and
panel structure of our model represents a useful advance over work that is
tied to the national level of analysis. Something like our approach may be
the wave of the future as economists come to grips with regional, national
and continental economic integration.
We draw a number of potential implications of this model for the de-
sign of unemployment policy in Europe. On the demand side, to state the
least questionable inference, raising the growth rate of GDP reduces un-
employment. That regional income convergence would do so is not readily
determined from our information, since our variable measures relative
wages. However, our model does suggest that income convergence would
help the poorer regions, and that policies explicitly targeted to achieve
regional income convergence would also reduce the divergence in unemploy-
ment rates, if not necessarily their average level. Policies that promoted
BNL Quarterly Review
16
income equalization for individuals – such as, for instance, measures that
raised the payout of non-wage incomes such as pensions in the poorer
regions – could in principle be expected to have this effect.
Targeted measures that provide pre-labor market opportunities for
European youth would appear to help on the supply-side (and may already
be doing so). Such opportunities would enable young people to time their
entry into paid employment so as to escape being tarred as either relatively
unproductive, or as having started working life with a long stretch of un-
employment. It may perhaps be noted that the United States does this very
effectively, with high levels of university enrollment, military enlistment –
and unfortunately also incarceration – all targeted to keeping youth off the
streets. As a result, youth unemployment in the United States is not (except
for certain relatively small populations) nearly as serious a social problem as
it is in Europe.
Perhaps our most interesting implication is that measures to reduce the
inequality of European wages at the regional level – for example, industrial
development policies in poor regions – would help reduce chronic unem-
ployment on average among Europeans. This is quite the opposite of the
common view that Europe needs more pay inequality (‘flexibility’) rather
than less. There is no support in our data for the idea that European unem-
ployment is due to excessive solidarity in the European wage structure. It is
possible, however, that some small countries have gamed the system at the
expense of their larger neighbours; by exercising solidarity and discipline
they have made themselves into attractive competitors for jobs in the
traded-goods sectors.
Our analysis of country fixed effects lends little encouragement to the
search for magic bullets in the form of national labor market institutional
reforms. Perhaps the other large European countries should investigate the
UK case very closely. Perhaps they should investigate Spain to learn what
to avoid (except for the fact that, not being Spain, they have already
avoided it). Perhaps there is something modest to be learned from Dutch
active labor market policies; Holland (with low emigration) has somewhat
lower-than-expected unemployment. (On the other hand, Holland also has
high rates of disability and part-time work, social accommodations to a
shortage of work that other countries may prefer to shun.) Apart from that,
there is little evidence that institutional differences among France, Ger-
many, Italy and the Nordic countries make big differences to their unem-
ployment rates; most of the differences between these countries experi-
ences seem fully accounted for by the regional variables.
Unemployment, inequality and the policy of Europe: 1984-2000 17
Finally, our evidence points a reproving finger at the institutions and
policy-makers of the European Union. It appears from our evidence that
European policy strongly contributed to a continent-wide increase in
unemployment in the 1990s. In a word, the Maastricht Treaty opened a
half-decade that can be qualified as disastrous, and from which recovery is
still incomplete. Overcoming the high unemployment visited on Europe as
a whole by the misgovernment of macroeconomic policy at the continental
level under recent leadership emerges from this analysis as a high priority.
Though some progress appears to have been made in the late 1990s, a
return even to the by-no-means-optimal conditions of the mid-1980s
remains quite far from complete.
BNL Quarterly Review
18
APPENDIX 1
Country and time fixed effects
TABLE A1
NATIONAL DUMMIES – LINEAR MODEL (1984-2000)
Model 1 Model 2 Model 3 Model 4 Model 5
Total Pvalue Male Pvalue Female Pvalue <25 Yrs Pvalue >25 Yrs Pvalue
BE 1.54 0.02 –0.35 0.53 5.16 0.00 –2.44 0.10 2.30 0.00
DE 3.32 0.00 4.12 0.00 2.97 0.00 –7.59 0.00 3.93 0.00
GR –5.20 0.00 –5.12 0.00 –3.64 0.00 1.45 0.42 –6.82 0.00
ES 5.04 0.00 3.70 0.00 8.96 0.00 9.71 0.00 2.86 0.00
IE –9.70 0.00 –6.48 0.00 –14.57 0.00 –24.12 0.00 –7.47 0.00
IT 0.53 0.17 –0.24 0.48 3.46 0.00 9.28 0.00 –1.69 0.00
NL –3.69 0.00 –3.16 0.00 –4.03 0.00 –13.00 0.00 –2.79 0.00
AT –6.03 0.00 –4.90 0.00 –7.05 0.00 –17.09 0.00 –5.12 0.00
PT –10.79 0.00 –8.25 0.00 –13.86 0.00 –16.81 0.00 –10.43 0.00
FI 0.90 0.24 3.26 0.00 –1.97 0.06 3.30 0.06 0.42 0.51
SE –1.06 0.11 1.88 0.00 –4.41 0.00 –3.70 0.02 –0.95 0.08
UK –4.10 0.00 –0.28 0.60 –9.09 0.00 –12.64 0.00 –3.50 0.00
TABLE A2
TIME DUMMIES – LINEAR MODEL (1984-2000)
Model 1 Model 2 Model 3 Model 4 Model 5
Total Pvalue Male Pvalue Female Pvalue <25 Yrs Pvalue >25 Yrs Pvalue
1984 –0.36 0.70 –0.17 0.83 –0.70 0.58 0.06 0.98 –0.50 0.51
1986 1.11 0.18 1.60 0.03 0.36 0.75 2.35 0.22 0.75 0.28
1987 –0.10 0.91 0.08 0.91 –0.30 0.79 –0.14 0.94 –0.22 0.74
1988 1.76 0.03 1.38 0.06 2.38 0.04 1.70 0.37 1.72 0.01
1989 –0.17 0.83 –0.14 0.84 –0.27 0.80 –2.90 0.12 0.56 0.40
1990 –0.99 0.21 –0.83 0.23 –1.31 0.23 –4.59 0.01 0.04 0.96
1991 –1.11 0.17 –0.98 0.17 –1.45 0.19 –5.51 0.00 0.19 0.78
1992 –0.28 0.73 –0.09 0.90 –0.81 0.47 –3.44 0.07 0.84 0.22
1993 1.86 0.04 1.96 0.01 1.53 0.21 1.28 0.54 2.53 0.00
1994 4.57 0.00 4.31 0.00 4.70 0.00 5.72 0.01 5.09 0.00
1995 2.32 0.00 2.46 0.00 1.95 0.07 3.33 0.06 2.95 0.00
1996 2.74 0.00 2.88 0.00 2.45 0.02 4.39 0.01 3.30 0.00
1997 2.76 0.00 3.04 0.00 2.23 0.04 4.37 0.02 3.34 0.00
1998 2.06 0.01 2.03 0.00 1.97 0.07 2.63 0.14 2.74 0.00
1999 1.55 0.05 1.65 0.02 1.31 0.23 1.22 0.51 2.36 0.00
2000 0.83 0.33 1.25 0.10 0.21 0.86 0.05 0.98 1.64 0.02
Unemployment, inequality and the policy of Europe: 1984-2000 19
APPENDIX 2
Constructing the Theil statistic
The Theil statistic is composed of two elements: a between-group inequality
component and a within-group inequality component:
T ≡ TB + T
_
w
(1)
where:
T = total Theil
TB = between-groups Theil component
T
_
w = within-group Theil component.
The between-groups component can be represented by the following two
equations:
∑∑
∑
∑
=
=
=
=
=n
1i n
1i ii
n
1i ii
n
1i i
i
Bee
ww
ln
w
w
T (2)
=∑∑
=
Y
i
Y
i
n
1j
i
Bw
w
ln
w
w
ej
e
T (2′)
The within group component equals:
w
n
1i
i
wT
w
w
T•
=∑
= (3)
•
=
iij
iij
i
ij
wee
ww
ln
w
w
T (4)
If we index regions with the subscript i and sectors with the subscript j, then
wij = the total compensation received in region i and sector j
eij = total people employed in region i and sector j
−
wi = average income of region i
−
wY = average income of all regions.
⋅
j
¯
BNL Quarterly Review
20
APPENDIX 3
List of regions and sectors in the REGIO data set
TABLE A3
LIST OF REGIONS – NUTS LEVEL 1 FOR DE AND UK,
NUTS LEVEL 2 FOR REMAINING COUNTRIES
1 be1 Région Bruxelles-hoofdstad gewest
2 be21 Antwerpen
3 be22 Limburg (B)
4 be23 Oost-Vlaanderen
5 be24 Vlaams Brabant
6 be25 West-Vlaanderen
7 be31 Brabant Wallon
8 be32 Hainaut
9 be33 Liège
10 be34 Luxembourg (B)
11 be35 Namur
12 de1 Baden-Württemberg
13 de2 Bayern
14 de3 Berlin
15 de4 Brandenburg
16 de5 Bremen
17 de6 Hamburg
18 de7 Hessen
19 de8 Mecklenburg-Vorpommern
20 de9 Niedersachsen
21 dea Nordrhein-Westfalen
22 deb Rheinland-Pfalz
23 dec Saarland
24 ded Sachsen
25 dee Sachsen-Anhalt
26 def Schleswig-Holstein
27 deg Thüringen
28 def Schleswig-Holstein
29 deg Thüringen
30 gr11 Anatoliki Makedonia, Thraki
31 gr12 Kentriki Makedonia
32 gr13 Dytiki Makedonia
33 gr14 Thessalia
34 gr21 Ipeiros
35 gr22 Ionia Nisia
36 gr23 Dytiki Ellada
37 gr24 Sterea Ellada
38 gr25 Peloponnisos
39 gr3 Attiki
40 gr41 Voreio Aigaio
41 gr42 Notio Aigaio
42 gr43 Kriti
43 es11 Galicia
44 es12 Principado de Asturias
45 es13 Cantabria
46 es21 Pais Vasco
47 es22 Comunidad Foral de Navarra
48 es23 La Rioja
49 es24 Aragón
50 es3 Comunidad de Madrid
51 es41 Castilla y León
52 es42 Castilla-la Mancha
53 es43 Extremadura
54 es51 Cataluña
55 es52 Comunidad Valenciana
56 es53 Illes Balears
57 es61 Andalucia
58 es62 Murcia
59 es63 Ceuta y Melilla (ES)
60 es7 Canarias (ES)
61 fr1 Île de France
62 fr21 Champagne-Ardenne
63 fr22 Picardie
64 fr23 Haute-Normandie
65 fr24 Centre
66 fr25 Basse-Normandie
67 fr26 Bourgogne
68 fr3 Nord-Pas-de-Calais
69 fr41 Lorraine
70 fr42 Alsace
71 fr43 Franche-Comté
72 fr51 Pays de la Loire
73 fr52 Bretagne
74 fr53 Poitou-Charentes
75 fr61 Aquitaine
76 fr62 Midi-Pyrénées
77 fr63 Limousin
78 fr71 Rhône-Alpes
79 fr72 Auvergne
80 fr81 Languedoc-Roussillon
81 fr82 Provence-Alpes-Côte d’Azur
82 fr83 Corse
83 ie01 Border, Midlands and Western
84 ie02 Southern and Eastern
85 it11 Piemonte
86 it12 Valle d’Aosta
87 it13 Liguria
88 it2 Lombardia
89 it31 Trentino-Alto Adige
90 it32 Veneto
91 it33 Friuli-Venezia Giulia
92 it4 Emilia Romagna
93 it51 Toscana
94 it52 Umbria
95 it53 Marche
96 it6 Lazio
97 it71 Abruzzo
98 it72 Molise
99 it8 Campania
100 it91 Puglia
101 it92 Basilicata
102 it93 Calabria
103 ita Sicilia
104 itb Sardegna
Unemployment, inequality and the policy of Europe: 1984-2000 21
105 lu Luxembourg
106 nl11 Groningen
107 nl12 Friesland
108 nl13 Drenthe
109 nl21 Overijssel
110 nl22 Gelderland
111 nl23 Flevoland
112 nl31 Utrecht
113 nl32 Noord-Holland
114 nl33 Zuid-Holland
115 nl34 Zeeland
116 nl41 Noord-Brabant
117 nl42 Limburg (NL)
118 at11 Burgenland
119 at12 Niederösterreich
120 at13 Vienna
121 at21 Kärnten
122 at22 Steiermark
123 at31 Oberösterreich
124 at32 Salzburg
125 at33 Tirol
126 at34 Vorarlberg
127 pt11 Norte
128 pt12 Centro (PT)
129 pt13 Lisboa e Vale do Tejo
130 pt14 Alentejo
131 pt15 Algarve
132 pt2 Açores (PT)
133 pt3 Madeira (PT)
134 fi13 Itä-Suomi
135 fi14 Väli-Suomi
136 fi15 Pohjois-Suomi
137 fi16 Uusimaa (suuralue)
138 fi17 Etelä-Suomi
139 fi2 Åland
140 se01 Stockholm
141 se02 Östra Mellansverige
142 se04 Sydsverige
143 se06 Norra Mellansverige
144 se07 Mellersta Norrland
145 se08 Övre Norrland
146 se09 Småland med öarna
147 se0a Västsverige
148 ukc North East
149 ukd North West (including Merseyside)
150 uke Yorkshire and The Humber
151 ukf East Midlands
152 ukg West Midlands
153 ukh Eastern
154 uki London
155 ukj South East
156 ukk South West
157 ukl Wales
158 ukm Scotland
159 ukn Northern Ireland
TABLE A4
SECTORIZATION USED TO CALCULATE REGIONAL INEQUALITY
Sectors by NACE-CLIO
(1984-1994)
Sectors by NACE
(1995-2000)
Fuel and power products
Ferrous and non-ferrous ores and metals, other
than radioactive
Non-metallic minerals and mineral products
Chemical products
Metal products, machinery, equipment and
electrical goods
Transport equipment
Food, beverages, tobacco
Textiles and clothing, leather and footwear
Paper and printing products
Products of various industries
Building and construction
Recovery, repair, trade, lodging and catering
services
Transport and communication services
Services of credit and insurance institutions
Other market services
Non-market services
Agriculture, hunting and forestry
Fishing
Mining and quarrying
Manufacturing
Electricity, gas and water supply
Construction
Wholesale and retail trade; repair of motor
vehicles*
Hotels and restaurants
Transport, storage and communication
Financial intermediation
Real estate, renting and business activities
Public administration and defence; compul-
sory social security
Education
Health and social work
Other community, social, personal service
activities
Private households with employed persons
* Motorcycles and personal and household goods.
BNL Quarterly Review
22
APPENDIX 4
Sensitivity analyses
The REGIO data set permits us to extract annual data set from 1984 to 2000 for
the major countries of Europe. However, for a number of the small countries,
including Greece, Austria, Ireland and Portugal, full data are available only for the
second half of the 1990s. This raises two questions: whether those years are repre-
sentative of the whole period for these countries, and whether the panel analysis as
a whole would be different if they were excluded.
Examination of the unemployment rates for the four countries suggests that
the relatively low unemployment rates seen in Austria, Greece and Portugal in the
late 1990s are not wildly unrepresentative of their experience over the whole
period, even though the absolute levels of unemployment do vary through time.
The Irish case is very different, as Ireland passed from a high- to a low-
unemployment country in the mid-1990s. It would thus be inappropriate to regard
the low country fixed effect found for Ireland as representative of institutions
producing low unemployment throughout the period. It represents, rather, the
exceptional experience of the late 1990s, when Ireland experienced a powerful
economic boom.
To test the second question, we ran the full panel regression, with two-way
fixed effects, on a panel excluding Greece, Austria, Ireland and Portugal. The
results for the whole population are given in Table A5. Results for the male,
female, young and older subpopulations tell a similar story and are available from
the authors.
Unemployment, inequality and the policy of Europe: 1984-2000 23
TABLE A5
SENSITIVITY ANALYSIS – MODEL 1
(TOTAL UNEMPLOYMENT) – EXCLUDING AU, IE, GR, PT
Model 1
Total Pvalue
Theil 31.75 0.00
PopUn24 71.48 0.00
RelWage –6.15 0.00
G-GDP –6.92 0.00
BE 1.29 0.05
DE 4.54 0.00
ES 4.21 0.00
IT 0.32 0.43
NL –3.47 0.00
FI 1.38 0.07
SE –0.52 0.43
UK –4.69 0.00
84 –0.36 0.70
86 1.11 0.18
87 –0.10 0.91
88 1.76 0.03
89 –0.17 0.83
90 –0.99 0.21
91 –1.11 0.17
92 –0.28 0.73
93 1.86 0.04
94 4.57 0.00
95 2.32 0.00
96 2.74 0.00
97 2.76 0.00
98 2.06 0.01
99 1.55 0.05
00 0.83 0.33
R
– 2 0.63
No. observations 1240
The model is substantially unaffected by the exclusion of the four small coun-
tries. All coefficients have the same sign and all remain significant. One difference
is that the relationship between inequality and unemployment is stronger, and the
significance of the coefficient estimate on the inequality variable rises eight-fold,
when the four small countries are not included. We take this as confirmation that
the inequality-unemployment relation is not an artifact of the inclusion of the small
countries in the late 1990s.
BNL Quarterly Review
24
APPENDIX 5
Wage and employment effects on inequality
The between-groups component of Theil’s T statistic is a compound measure
influenced by both the relative wage rates between groups and the relative size of
each group. A region with high inequality may have a large differential between the
best and worst paid, or a marked bimodalism in the structure of employment, or
some combination of both factors. It is worth noting that the line of causality
traditionally argued to hold in economics, which runs from unemployment rates to
the pay structure, does not imply anything in particular about the structure of
employment. If there exists a large excess of unskilled workers, that should reduce
the relative pay of unskilled workers, increasing inequality, but it would not neces-
sarily change the technology employed in particular processes of production.
To provide an illustration of the roles of these two factors we examine the struc-
ture of pay and employment in four European regions, two with high and two with
low unemployment in the year 2000. The following regions are included in the
analysis: Andalucia and Extremadura with high unemployment rates, and Navarra
and Stockholm with low unemployment rates:
– Extremadura (24.4%)
– Andalucia (25%)
– Navarra (4.8%)
– Stockholm (3.7%).
TABLE A6
SUMMARY STATISTICS FOR AVERAGE WAGES
ACROSS 16 SECTORS FROM 1995-2000
Mean Min. Max.
No.
observations
Extremadura 21.49 5.4 65.5 72
Andalucia 22.65 5.1 79.7 82
Navarra 25.93 7.5 52.1 72
Stockholm 35.59 16.7 64 88
Ranges for low-unemployment regions are much lower than for high-
unemployment regions. We also find that low unemployment regions have sub-
stantially larger shares of their employment near the mean, and less associated with
the extremes of the distribution.
APPENDIX 6
Unemployment, inequality and the policy of Europe: 1984-2000 25
Evaluating the strategic-wage conjecture
The conjecture that certain small countries with strong collective wage bargaining
might generate domestic full employment at the expense of a larger neighbour can
be evaluated directly for the case of Austria and Germany. The evidence is sugges-
tive. As Table A7 shows, average wages in Austria are systematically higher than in
Germany except in two sectors: manufacturing and real estate. Manufacturing is, of
course, by far the largest of these sectors. Is this the secret of Austrian unemploy-
ment rates consistently half those of Germany?
TABLE A7
RATIO OF AUSTRIAN TO GERMAN AVERAGE WAGES, BY MAJOR SECTORS
1995 1996 1997 1998 1999 2000
Mining and quarrying 1.04 1.01 1.01 1.06 1.09 0.98
Manufacturing 0.88 0.88 0.88 0.89 0.92 0.86
Electricity, gas and water supply 1.22 1.19 1.21 1.26 1.22 1.14
Construction 1.04 1.03 1.06 1.11 1.27 1.20
Transport, storage and communication 1.03 1.00 1.03 1.07 1.18 1.14
Financial intermediation 1.06 1.07 1.08 1.09 1.23 1.18
Real estate, renting and business activities 0.99 0.96 0.94 0.90 1.09 0.95
Public administration and defence; compulsory social
security 1.16 1.15 1.13 1.10 1.12 1.12
Table A8 gives a similar analysis of relative wages in Ireland and the UK in the
late 1990s; if the data are accurate a similar story may apply. Indeed it is striking
how much higher average pay in such sectors as finance, health and education
appears to be in Ireland than in England. But manufacturing pay is lower, and this
could well have given Ireland the edge in the location of new industry during the
technology boom.
BNL Quarterly Review
26
TABLE A8
RATIO OF IRISH TO BRITISH AVERAGE WAGES, BY MAJOR SECTORS
1995 1996 1997 1998
Mining and quarrying 0.71 1.05 0.86 0.87
Manufacturing 0.81 0.84 0.75 0.71
Electricity, gas and water supply 0.74 0.65 0.70 0.63
Construction 1.32 1.27 1.17 1.11
Wholesale and retail trade* 1.35 1.39 1.32 1.29
Hotels and restaurants 1.15 1.05 0.97 0.90
Transport, storage and communication 0.79 0.87 0.76 0.70
Financial intermediation 1.51 1.49 1.20 1.11
Real estate, renting and business activities 1.19 1.13 1.07 1.02
Public administration and defence** 1.08 1.17 1.11 1.18
Education 1.27 1.30 1.17 1.10
Health and social work 1.52 1.48 1.39 1.22
Other community, social, personal service activities 0.97 0.90 0.66 0.57
** Including repair of motor vehicles, motorcycles and personal and household goods.
** Including compulsory social security.
APPENDIX 7
Coverage by country and year
TABLE A9
DATA COVERAGE BY COUNTRY AND YEAR
(number of regions in parentheses)
Year
No.
observa-
tions
1984 35 be (8) it (20) uk (7)
1985 35 be (8) it(20) uk (7)
1986 56 be (8) es (17) it (20) pt (4) uk (7)
1987 69 be (8) es (17) fr (20) it (20) pt (4)
1988 63 be (8) es (18) it (20) nl (12) pt (5)
1989 84 be (8) es (18) fr (21) it (20) nl (12)
1990 86 be (8) es (18) fr (21) it (20) nl (12)
1991 78 es (18) fr (21) it (20) nl (12) pt (7)
1992 78 es (18) fr (21) it (20) nl (12) pt (7)
1993 57 es (18) it (20) nl (12) pt (7)
1994 45 es (18) it (20) pt (7)
1995 133 de (16) gr (13) es (18) fr (21) it (20) nl(12) at (9) pt (7) fi (4) se (6) uk (7)
1996 139 de (16) gr (13) es (18) fr (21) it (20) nl(12) at (9) pt(7) fi (6) se (6) uk (11)
1997 136 de (16) gr (12) es (18) fr (21) it (20) nl(12) at (9) pt (7) fi (4) se (6) uk (11)
1998 144 de (16) gr (12) es (18) fr (21) ie (2) it(20) nl (12) at (9) pt (7) fi (6) se (8) uk (12)
1999 131 de (16) gr (12) es (18) fr (21) ie (2) it(20) nl (12) at (9) pt (7) fi (5) se (8)
2000 96 de (16) gr (13) fr (21) ie (2) it (20) at(9) fi (6) se (8)
1465
Unemployment, inequality and the policy of Europe: 1984-2000 27
BNL Quarterly Review
28
REFERENCES
BAKER, D., A. GLYN, D. HOWELL and J. SCHMITT (2003), “Labor market institutions and
unemployment: a critical assessment of the cross-country evidence”, Harvard Univer-
sity, Minda de Gunzburg Center for European Studies, Working Paper, no. 98.
BLANCHARD, O. and J. WOLFERS (1999), “The role of shocks and institutions in the rise of
European unemployment: the aggregate evidence”, NBER Working Paper, no. 7282.
CONCEIÇÃO, P., P. FERREIRA and J.K. GALBRAITH (1999), “Inequality and unemployment in
Europe: the American cure”, New Left Review, no. 237, September-October, pp. 28-51.
CONCEIÇÃO, P. and J.K. GALBRAITH (2000), “Constructing long and dense time series of
inequality using the Theil statistic”, Eastern Economic Journal, vol. 26, no. 1, pp. 61-74.
CONCEIÇÃO, P., J.K. GALBRAITH and P. BRADFORD (2001), “The Theil index in sequences
of nested and hierarchical grouping structures: implications for the measurement of
inequality through time, with data aggregated at different levels of industrial classifica-
tion”, Eastern Economic Journal, vol. 27, no. 4, pp. 491-514.
GORDON, R.J. (1999), “The aftermath of the 1992 ERM breakup: was there a macroeco-
nomic free lunch?”, NBER Working Paper, no. 6964.
HARRIS, J.R. and M.P. TODARO (1970), “Migration, unemployment and development: a two-
sector analysis”, The American Economic Review, vol. 60, issue 1, pp. 126-42.
KEYNES, J.M. (1936), The General Theory of Employment, Interest and Money, Macmillan, London.
KUZNETS, S. (1955), “Economic growth and income inequality”, American Economic Review,
vol. 65, no. 1, pp. 1-28.
NICKELL, S. (1997), “Unemployment and labor market rigidities: Europe versus North
America”, Journal of Economic Perspectives, vol. 11, no. 3, pp. 55-74.
PALLEY, T.I. (2004), “The causes of high unemployment: labor market sclerosis versus
macroeconomic policy”, in J. Stanford and L. Vosko eds, Challenging the Market: The
Struggle to Regulate Work and Income, McGill-Queens University Press,
Toronto, forthcoming.
THEIL, H. (1972), Statistical Decomposition Analysis: With Applications in the Social and Administra-
tive Sciences, North Holland Publishing Company, Amsterdam-London.
