Cambridge Journal of Economics 2006, 30, 181–199
doi:10.1093/cje/bei054
Advance Access publication 6 June, 2005
The sensitivity of poverty rates
to macro-level changes in the
European Union
Herwig Immervoll, Horacio Levy, Christine Lietz,
Daniela Mantovani and Holly Sutherland*
The authors use the European Union-wide tax–benefit model, EUROMOD, to
establish baseline rates of relative poverty in 1998 for each of the Member States
and then explore their sensitivity to (a) an increase in unemployment, (b) real
income growth and (c) an increase in earnings inequality. They find that poverty
rates are sensitive to such ‘macro-level’ changes but that the size—and in some
cases the direction—of the effect varies across countries. If such indicators are to be
used in judging the effectiveness of social policies, it is important that differences in
responsiveness are fully understood.
Key words: European Union, Microsimulation, Social indicators
JEL classifications: C81, D31, I32
1. Introduction
Late in 2001, the European Union (EU) adopted a set of commonly agreed indicators to
assess social inclusion. The main impetus for this achievement arose, first, through the
agreement at the Lisbon European Council to promote social inclusion as a key
component of the strategy of the EU, and then with the adoption of the open method of
coordination at the Nice Summit. The process involves Member States submitting
National Action Plans for Inclusion (NAPIncl), which spell out social policy initiatives
designed to reduce social exclusion and to promote inclusion. The extent to which these
objectives are met is assessed both by Member States in their NAPIncl reports and by the
Commission (together with Member States) in their Joint Report on Social Inclusion. The
‘toolbox’ for this assessment consists of indicators that are relevant in specific national
contexts, alongside the common indicators which act as ‘measuring instruments allowing
Member States to use a common language for the assessment of the various phenomena at
stake’ (Atkinson et al., 2002B, p. 8). These include income-based measures of inequality
Manuscript received 6 October 2003; final version received 27 October 2004.
Address for correspondence: Holly Sutherland, Institute for Social and Economic Research, University of
Essex, Wivenhoe Park, Colchester CO4 3SQ, UK; email: [email protected]
* University of Cambridge and European Centre for Social Welfare Policy and Research, Vienna;
Universitat Autonoma de Barcelona; University of Cambridge; University of Cambridge, and University
of Essex and DIW Berlin, respectively.
Ó The Author 2005. Published by Oxford University Press on behalf of the Cambridge Political Economy Society.
All rights reserved.
182
H. Immervoll et al.
and relative and absolute poverty, as well as non-monetary indicators (Social Protection
Committee, 2001).
It is to be hoped that Member States can devise policies that will reduce poverty and social
exclusion and that these reductions will be reflected in improvements in the indicators.
However, any positive effects of targeted social policy initiatives may be mitigated by other,
independent changes in the economy or society. Macro-level changes such as increasing
unemployment may inhibit the movement of a social indicator in the desired direction. At
the same time, governments may rely on macro-level economic policies to achieve social
objectives: growth may contribute to raising the incomes of the poor. In general, it is
important to recognise and understand cross-country differences in sensitivity of the chosen
indicators to macro-level changes. In monitoring progress across countries, account needs
to be taken of the effect of cyclical changes in activity on indicators. The timing and size of
cyclical effects may differ across countries, but so might the sensitivity of the indicator to the
macro-level change. Not only do national population characteristics differ, but also tax and
benefit systems vary in the extent to which they automatically absorb the impact of changes
and protect populations from any adverse effects.
In the initial period of the open method of coordination, the performance of the agreed
indicators is being tracked using data for successive years from existing EU sources of data
(mainly the Eurostat Labour Force Survey and European Community Household Panel).
Thus, an observed change in an indicator will reflect not only the impact of policy reforms
intended to reduce exclusion. It will also reflect (i) the impact of other policy reforms, with
other goals, and (ii) the impact of other influences such as changes in the level of economic
activity, changes in demographic composition or changes in the distribution of sources of
primary income. While we should like to assess the effect of policies intended to promote
inclusion, it is difficult or impossible to decompose the observed change in the value of
the indicator into the parts that are due to each influence, not least because they are not
independent of each other. We can, however, use static microsimulation methods to hold
most influences constant and to focus on the effect of one change at a time.1 Typically,
static microsimulation models are used to explore the direct, first-order effects of policy
changes, while holding higher order and exogenous effects constant (Sutherland, 2002).
In this paper, we hold the tax–benefit policy systems constant (as in 1998) and simulate
the effect of a series of changes on the underlying populations and income distributions.
We consider the impact of increasing unemployment, real income growth and increasing
earnings inequality.2
There is an existing literature on the effect of such macro changes on the income
distribution that is based on time-series analyses of relevant variables. Parker (1998–99)
provides a review of such studies. The impact of UK economic conditions has been
explored more recently by Jäntti and Jenkins (2001), who summarise the findings as
‘unemployment had a regressive impact [on inequality] and no statistically significant
association with inflation [could be found]’ (p. 2). As with the monitoring of social
indicators, the use of time series data makes the identification of the role of specific factors
difficult to achieve. In the present study, microsimulation methods allow us to focus on one
1
See Harding (1996) and Mitton et al. (2000) for useful overviews of ‘static’ and ‘dynamic’ microsimulation methods.
2
Nolan (1987) identifies the main ‘channels of influence’ of cyclical macroeconomic changes on the
distribution of income as being unemployment and changes in the shares of income by type. Here, partly
because of poor-quality data on unearned income in the micro-datasets that we make use of, we focus on
changing shares of earned income, as well as unemployment.
Macro-level changes in the European Union
183
change at a time. The main drawbacks of this approach are that we must specify the precise
form of the macro change and consider how to introduce it consistently across countries,
and that we do not capture second-order or higher-order effects. The advantages are that
we have no identification problem: the results are transparent, and that the same
‘experiment’ can be implemented in different countries.
We explore the ways that the 15 national tax–benefit systems respond to the macro-level
changes and the resulting consequences for one of the Social Protection Committee’s
primary indicators: the ‘low income rate after transfers’. For convenience, we refer to this
as the poverty rate. This is based on the number of people living in households with
disposable income (i.e., income after taxes and benefits) less than 60% of the median.1
Since our main interest is in differences in responsiveness across countries, we use
a microsimulation model that is specifically designed for comparisons across EU Member
States: EUROMOD. Section 2 describes this model and presents the poverty rates
calculated using simulated incomes for 1998 from EUROMOD. We refer to these incomes
as the ‘baseline’ against which incomes calculated following simulated macro-level changes
may be compared. Section 3 explains these changes and discusses the impact on poverty
rates that we might expect, a priori. Section 4 presents the results and Section 5 concludes.
2. EUROMOD and social indicators
EUROMOD is a tax–benefit model for the EU. (See Immervoll et al. (1999) for a general
description.) Tax–benefit models calculate disposable income for each household in
a representative set of micro-data. The datasets used as the basis for this paper are listed in
Appendix 1. They were chosen on the grounds that they provide the best quality input for
a tax–benefit model and are at the same time available and accessible to an international
scientific project. Although they include data collected at various points in time in the
period 1993–99, they have all been adjusted to mid-1998 prices and incomes and, where
necessary, gross incomes have been imputed from net (Immervoll and O’Donoghue,
2001). The calculation of household disposable income is made up of elements of gross
income taken (or imputed) from the survey data combined with elements of income—taxes
and benefits—that are simulated by the model. The calculations are performed once for
the 1998 tax–benefit system and population (the ‘baseline’), and again for each alternative
scenario. The first round effect of the simulated change is the arithmetic difference
between the ‘before’ and ‘after’ calculations.
In this exercise, we estimate poverty using two different types of poverty line. The first is
calculated using 60% of the median of baseline incomes. We refer to this as the ‘fixed’
poverty line. The second or ‘relative’ poverty line uses the median that is recalculated after
the macro change has taken effect. This is referred to as the ‘within-scenario’ median. We
are thus able to distinguish between changes in poverty rates that are due to incomes rising
above or falling below a fixed income level from those due to shifts in the poverty line. It is
often possible to predict the direction of each change individually, but measurement of the
relative size of each effect, and the net consequences for the indicator and the composition
of the poor, require detailed micro-level calculations.
Poverty rates as calculated by EUROMOD using 1998 taxes, benefits, prices and
incomes for the 15 Member States are shown in Figure 1, using 60% of the 1998 national
1
Other income-based indicators, as adopted by the Laeken European Council and recommended by the
Indicators Sub-Group of the EU Social Protection Committee (2001), and Atkinson et al. (2002A) are
considered in a longer version of this paper (Feres et al., 2002).
H. Immervoll et al.
Percentage of population living in households with
income below 60% of the national median
184
25.0%
20.0%
15.0%
10.0%
5.0%
0.0%
PT
IR
IT
Self-employed
GR
UK
Employee
SP
BE
Unemployed
FR
LU
Children
AT
DK
NL
FI
GE
SW
Pensioners and others
Fig. 1. Percentage of population living in households with income below 60% of the national median,
decomposed by economic status: EUROMOD baseline 1998.
Source: EUROMOD (see Appendix 1 for country codes).
‘baseline’ medians as the poverty lines.1 Countries are ranked according to the poverty
rate: Portugal has the highest and Sweden the lowest rate. Generally, the rates and their
rank position correspond well with estimates obtained from other sources.2 The composition of the poor is shown according to their economic status. The variation across
countries in the prevalence of unemployed, employed and self-employed among the poor,
shown in Figure 1, indicates that we might expect changes in unemployment and in
earnings level and distribution to have a non-uniform cross-country impact on poverty
indicators. At the same time, such information does not tell the whole story. Children make
up a large proportion of the poor in many countries and the impact of macro-level changes
on them will depend on the economic status of the adults with whom they live. More
generally, it is the composition of household income that is of critical importance.
3. Real issues and simulated scenarios
Two practical concerns lie behind the choice of macro-level changes considered in this
paper. The first is that recession, and particularly an increase in demand-driven
worklessness, will undermine the efforts of social policy-makers to improve the chances
of those at risk of poverty. The relevant questions we address are: How large is the effect?
and are poverty rates in some Member States more sensitive than in others?
The second concern is more of an open question: whether the factors driving market
income growth are forces that tend to improve the performance of social indicators
1
Poverty rate calculations here and throughout use the methods recommended by Eurostat and made use
of by Dennis and Guio (2003). Household disposable incomes are equivalised using the modified OECD
equivalence scale (1 for a single person; þ0.5 for each additional person aged 14þ and þ0.3 for each person
aged under 14). The median is calculated by ranking individuals by their equivalised household disposable
income and the poverty rate is the percentage of individuals living in households with income below 60% of
the median.
2
Microsimulation model estimates are subject to many sources of error and their quality may also vary
across country. For a description of the assumptions behind the calculations, a discussion of issues affecting
the quality and comparability of results and comparisons of EUROMOD 1998 ‘baseline’ poverty rate
estimates with other sources see Sutherland (2001) and Mantovani and Sutherland (2003).
Macro-level changes in the European Union
185
(i.e., by reducing measured poverty and social exclusion). These are big issues that we do
not aim to address directly. Such an analysis would require a dynamic approach and
a theoretical framework and methodology which relates macro change to micro outcome.
Instead, we focus on the mechanics of the relationship between income growth and the
behaviour of the ‘headline’ relative poverty indicator, based on measures of median income.
Our static microsimulation approach focuses on some of the mechanisms over which
governments have control—tax and benefit rules.
The aim is not to try to create scenarios that are realistic for each country but, instead, to
simulate simple changes that can be operationalised in a common way across countries,
making use of the information contained in the national databases to provide national
character. The purpose is to compare effects across European countries with different tax–
benefit systems and population characteristics. We examine three changes:
(a) an increase in unemployment
(b) growth in earned incomes
(c) an increase in earnings inequality.
In each case, we assume that governments make no compensating reforms to tax and
benefit rules. The 1998 rules are simply applied to the new sets of incomes and
circumstances. By construction, the scenarios are not budget-neutral. The three macrolevel changes are discussed in turn in terms of the way they are simulated and the effect we
might expect a priori on incomes in relation to the poverty line.
(a) An increase in unemployment
An increase in the unemployment rate of 5 percentage points is simulated. We assume that
the characteristics of the new unemployed are the same as those of the existing
unemployed. We reweight the existing populations to increase the importance of households containing an unemployed person, reducing the importance of households that are
similar in other respects. Appendix 2 explains the details of how this is done.
We might expect households with unemployed people to have lower incomes than other
demographically-equivalent households. This may not be the case if other groups such as
the inactive have lower incomes, or if unemployed people share households with people
with medium or high earnings. But generally, we should expect the increase in the
prevalence of unemployment to increase the poverty rate, if the poverty line is unchanged.
We should, however, expect the impact on median income and hence the relative poverty
line to be in a downward direction. It is therefore possible that an expansion of
unemployment could reduce the relative poverty rate: the numbers falling into poverty
because of becoming unemployed might be outweighed by the numbers appearing above
the falling line.
Table 1 shows that the rise in modelled unemployment results in a reduction in gross
earnings in all countries, offset to some extent by reduced taxes and contributions and
increased benefits. Given the way in which increasing unemployment is modelled, the size
of the aggregate reduction in market income depends on the average earnings of
households containing unemployed people and the earnings in households that are
demographically similar. The table shows that the change in market income as a proportion
of the pre-change (baseline) disposable income varies across countries. It falls by most in
Ireland (4.7%), Denmark (4.0%), Germany and Finland (3.7%), and by least in Greece
(1.1%), Luxembourg (1.9%) and Austria (2.1%). The increase in unemployment results
186
%
(a) An increase in unemployment
Change in market income
þChange in benefits
Change in social contributions
Change in taxes
¼Change in disposable income
(b) Real earnings growth
Change in market income
þChange in benefits
Change in social contributions
Change in taxes
¼Change in disposable income
(c) An increase in earnings inequality
Change in market income
þChange in benefits
Change in social contributions
Change in taxes
¼Change in disposable income
a
AT
BE
DK
FI
FR
GE
GR
IR
IT
LU
NL
PT
SP
SW
UK
2.1
0.4
0.3
0.6
0.9
3.0
0.9
0.3
0.8
1.0
4.0
2.0
0.3
0.8
0.9
3.7
1.0
0.2
0.9
1.6
2.9
0.7
0.5
0.2
1.5
3.7
1.0
0.5
0.9
1.3
1.1
0.0
0.1
0.2
0.8
4.7
1.2
0.2
0.9
2.5
3.1
0.0
0.3
0.6
2.2
1.9
0.6
0.1
0.4
0.8
3.0
1.0
0.4
0.3
1.3
3.5
0.7
0.3
0.7
1.7
2.7
0.4
0.2
0.6
1.6
2.5
1.2
0.0
0.4
0.8
3.6
0.8
0.2
0.6
2.0
9.5
0.0
1.3
3.0
5.1
9.0
0.1
1.0
3.5
4.4
12.5
0.2
1.1
5.9
5.3
8.2
0.1
0.7
3.8
3.7
8.3
0.2
1.5
1.0
5.6
10.2
0.2
1.3
3.6
5.0
9.2
0.0
0.8
2.1
6.3
9.5
0.1
0.4
3.2
5.8
8.2
0.1
0.7
2.6
4.8
8.7
0.0
0.7
2.8
5.2
10.0
0.1
1.4
3.0
5.4
9.6
0.1
0.9
2.2
6.4
8.8
0.0
0.2
2.5
6.0
8.3
0.2
0.4
3.6
4.2
8.8
0.1
0.5
2.4
5.9
0.0
0.2
0.9
1.9
0.7
0.0
0.4
0.1
1.1
0.6
0.0
0.4
0.0
1.3
0.8
0.0
0.3
0.0
1.2
0.9
0.0
0.5
0.4
1.8
0.9
0.0
0.6
1.0
1.6
0.1
0.0
0.1
0.4
2.6
2.2
0.0
0.6
0.3
1.5
0.6
0.0
0.1
0.1
1.2
1.0
0.0
0.3
0.8
2.2
1.1
0.1
0.4
2.1
3.5
1.1
0.0
0.4
0.0
1.9
1.5
0.0
0.1
0.3
2.5
2.2
0.0
0.6
0.4
1.1
0.1
0.0
0.4
0.5
1.3
0.3
Source: EUROMOD (see Appendix 1 for country codes).
Income is measured per household and not equivalised. Change is measured as a percentage of pre-change ‘baseline’ household disposable income.
H. Immervoll et al.
Table 1. Simulating alternative scenarios: percentage change in household disposable income and its componentsa
Macro-level changes in the European Union
187
in an increase in benefits in all countries except Italy, where there are no unemployment
benefits, and Greece, where few people qualify and benefit amounts are small. At the same
time, the fall in gross income is in all countries automatically accompanied by a decrease in
social contributions and income tax such that the tax–benefit systems absorb part of the
loss of income associated with unemployment. In Belgium, Denmark, Germany and
Sweden taxes, and benefits absorb more than 2/3 (3/4 in Denmark) of the fall in average
market incomes, while in Greece and Italy this ratio is less than 1/3. As a result, average
household disposable income falls in all countries, but significantly less than market
incomes. It falls by most in Ireland, Italy and the UK (by 2.5%, 2.2% and 2.0%,
respectively) and by least in Greece, Luxembourg, Sweden (0.8%), Austria and Denmark
(0.9%).
Table 2 shows that the modelled increase in unemployment reduces equivalised median
incomes, and hence the national relative poverty lines, in all 15 countries. The decrease in
the poverty line follows patterns similar to the fall in mean disposable income (see Table
1), falling by most in Ireland, UK and Italy, and by least in Luxembourg, Austria and
Greece.
(b) Real earnings growth
Gross earned incomes are inflated by an illustrative common factor (10%) to represent real
income growth over some period of time, while keeping the parameters of the tax and
benefit system—such as tax thresholds and benefit rates—constant.1 This corresponds to
the typical situation where benefit payments and the cash value of tax concessions do not
keep pace with market income growth. In many Member States, the main components are
annually indexed for inflation, but this is by no means universal practice (Immervoll, 2004;
Messere, 1998). For example, there is no statutory indexation in Ireland. It is rare for
increases to match changes in earnings or incomes more generally. At the same time,
income taxes are buoyant, meaning that liabilities naturally grow with income. If tax
thresholds are indexed only for inflation (or not at all), tax burdens rise. This phenomenon
is known as fiscal drag. A similar mechanism exists in reverse for income-tested benefits: as
incomes rise, entitlements fall unless benefit levels are linked to income growth. More
generally, without indexation, benefit incomes fall relative to market incomes. Benefits are
typically the opposite of buoyant: they must be increased to make up for inflation and by
more if they are to keep pace with real income growth.
In our simulations, the 1998 tax–benefit rules are applied to gross incomes that have
been inflated by a 10% increase in current employment and self-employment incomes.
This will have the effect of increasing incomes for those in work, such that median
household income rises. Whether the corresponding rise in the relative poverty line
increases the net numbers counted as poor depends on the extent to which poor
households contain people in paid work. Table 1 shows that the percentage increase in
household disposable income following 10% real earnings growth varies from 3.7% in
Finland to 6.4% in Portugal. This is the net effect of changes in market income (Table 1
shows that this, as a proportion of prechange ‘baseline’ disposable income, varies from
8.2% in Finland and Italy to 12.5% in Denmark); changes in benefits that are earnings
tested (which are small and negative) and changes in taxes and contributions (which
are positive). Where average tax rates are relatively low and/or where the degree of
1
Our calculations ignore the fact that pensions in some contributory systems were, in 1998, automatically
linked to current levels of earnings. Pension income remains constant by assumption.
188
EURO
per montha
Baseline 1998
poverty line
Poverty line after:
(a) An increase in
unemployment
(b) Real earnings
growth
(c) An increase in
earnings inequality
Euro exchange rate
31 Dec 1998
% change
(a) An increase in
unemployment
(b) Real earnings
growth
(c) An increase in
earnings inequality
a
AT
BE
DK
FI
FR
GE
GR
IR
IT
LU
NL
PT
SP
SW
UK
749
613
859
674
716
711
310
535
523
1,072
684
263
361
642
647
745
607
849
663
701
702
308
517
507
1,066
671
261
354
636
626
789
638
908
702
758
746
330
570
549
1,118
724
282
380
675
683
711
583
817
644
662
684
286
489
495
1,014
643
226
331
614
594
13.76
40.34
7.459
5.946
6.56
1.956
340.75
0.7876
1936.3
40.34
2.204
200.48
166.39
9.512
0.7032
0.6
1.0
1.1
1.6
2.1
1.2
0.7
3.4
3.0
0.5
1.9
1.0
2.0
1.0
3.3
5.3
4.1
5.7
4.2
5.9
5.0
6.4
6.5
4.9
4.3
5.9
7.4
5.4
5.1
5.6
5.1
4.8
4.8
4.5
7.4
3.9
7.5
8.6
5.4
5.4
5.9
14.1
8.4
4.3
8.1
Source: EUROMOD (see Appendix 1 for country codes).
Equivalised using the modified OECD equivalence scale.
H. Immervoll et al.
Table 2. Simulating alternative scenarios: changes in the poverty line (60% equivalised median)
Macro-level changes in the European Union
189
progressivity is low, fiscal drag has a correspondingly small effect (as in France, Greece and
Portugal).1 Increases in median equivalised income, and hence the relative poverty lines,
are shown in Table 2.
(c) An increase in earnings inequality
In this third experiment, we increase earnings inequality while keeping mean earnings
constant. Gross earnings are adjusted according to the simple formula: Ynew¼KY n, where
n¼1.3, and K is a scaling factor, determined such that the mean of Yand Ynew are the same
within each country. The value of 1.3 was chosen to secure a large but plausible illustrative
increase in earnings inequality: the Gini coefficient for individual gross employment
income rises by between 6 percentage points (in Sweden) and 11 percentage points (in
Portugal).2 Thus, low earners face a reduction in market income, and high earners an
increase. The break-even point for employment earnings is well above the mean (varying
from 26.8% above the mean in Italy to 57.1% in Portugal), indicating that most
individuals’ earnings will fall in this scenario.
Although mean gross earnings of individuals are held constant, Table 1 shows that the
change in their distribution results in a reduction in average household disposable incomes.
The net reduction varies from 0.1% in Germany and Sweden to 2.2% in Greece and Spain.
Benefits increase a little (most in Sweden and Ireland, least in Greece and Spain). The
progressive income tax systems play the biggest role, with taxes rising in all countries: more
tax is collected on increased high incomes than is foregone on reduced low incomes. The
strongest effects are found in the Netherlands, Greece and Spain; the weakest in Belgium
and Sweden. In most countries social contributions act in the opposite direction—total
contributions fall. Ceilings on contributions mean that the extra contributions paid by high
earners are limited and in aggregate are more than matched by reductions in contributions
among the lower paid. The exceptions—Denmark, Finland and Portugal—are systems
that levy contributions mainly on a proportional basis. In such cases, the distribution of
earnings has no effect on the total contributions that are collected. Table 2 shows that in all
countries the relative poverty line falls. In proportional terms, the drop is largest in
Portugal (14.1%) and smallest in Germany (3.9%).
4. Results
Table 3 shows the poverty rates for the 1998 baseline and after each of the illustrative
macro-level changes. The rates are shown using both the fixed poverty line (using median
incomes from the baseline) and the poverty line calculated from median incomes after the
macro-level change is modelled. Table 2 shows the change in the relative poverty line
following each macro-level change. We discuss the effect on poverty of each of the macrolevel change individually below, summarising the main effects graphically.
1
In the case of France, social contributions are more important than income tax in reducing the effect of
earnings growth on disposable income.
2
In practice, there are several variables which together make up gross earnings. In all countries, there are
at least two variables (corresponding to earnings from employment and self-employment) but in some
there are more (e.g., the value of 13th and 14th month salaries). For a given value of n, convergence to
a single balancing value of K would be complex to achieve. We approximate by allowing K to be different
for each earnings component. We assume that no other changes take place that might, in practice,
accompany a change in earnings distribution or an individual change in earnings (such as changes in hours
of work).
190
% with income below:
Baseline
60% median
(a) An increase in unemployment
60% baseline median
60% within-scenario
median
(b) Real earnings growth
60% baseline median
60% within-scenario
median
(c) An increase in earnings inequality
60% baseline median
60% within-scenario
median
AT
BE
DK
FI
FR
GE
GR
IR
IT
LU
NL
PT
SP
SW
UK
10.7
14.8
10.6
9.2
13.0
8.3
20.5
21.2
20.8
11.6
10.0
22.0
18.1
6.0
19.9
11.1
10.7
15.3
14.8
10.5
9.9
10.0
9.1
14.0
12.8
8.7
8.1
20.8
20.6
23.0
19.9
22.8
21.5
11.7
11.6
10.8
9.6
22.2
21.5
19.0
18.0
6.0
5.8
21.5
19.7
9.6
11.4
13.8
15.3
9.9
12.7
8.8
10.7
11.5
14.2
7.7
10.0
18.4
20.8
20.4
24.3
19.3
21.1
9.7
11.5
8.9
12.2
19.3
23.3
16.4
18.2
5.7
6.6
19.2
22.3
14.2
11.6
16.6
13.7
12.3
9.7
11.6
9.1
18.1
13.2
10.6
8.4
26.8
23.6
25.0
17.7
24.2
22.0
18.7
15.8
12.9
9.0
30.1
22.6
24.4
20.7
6.7
6.1
22.3
16.6
Notes: Poverty rates are calculated from on household disposable income, equivalised using the modified OECD equivalence scale and weighted by household size.
The ‘baseline’ is the EUROMOD 1998 distribution of equivalised household disposable incomes before any macro-level changes. The ‘within scenario median’ is the
median of equivalised household disposable income after the macro-level change (unemployment, real earnings growth, or increased earnings inequality) has been
simulated.
Source: EUROMOD (see Appendix 1 for country codes).
H. Immervoll et al.
Table 3. Poverty rates under the 1998 EUROMOD baseline and simulated macro-level change scenarios
Macro-level changes in the European Union
191
(a) An increase in unemployment
Using a fixed poverty line, increasing the unemployment rate by 5 percentage points causes
poverty rates to rise in all countries but with a very small effect in some cases.1 Percentage
point increases in the poverty rate range from negligible in Sweden and 0.1 in Luxembourg
to 2.0 in Italy, 1.7 in Ireland and 1.6 in UK. If the poverty line is recalculated using median
income after the increase in unemployment, a very mixed picture emerges: the change in
the relative poverty rate varies from an increase of 0.7 percentage points to a decrease of 1.3
points and, in the majority of cases, it falls slightly or remains unchanged. This is illustrated
in Figure 2.
We have seen that the fall in the relative poverty line varies from 0.5% (Luxembourg) to
3.4% (Ireland). One might expect countries such as Ireland to be among those with the
largest reductions in poverty rate and countries such as Luxembourg to be among those
with the largest increases (or smallest reductions). Figure 3, which plots the percentage
point reduction in the relative poverty rate by the percentage change in level of the poverty
line, shows that there is no such clear relationship. While Ireland does indeed experience
the largest percentage point drop in relative poverty, in Italy where the downward shift in
poverty line is almost as large, the relative poverty rate increases more than in any other
country. Where the drop in the poverty line is significant, the effect on the poverty rate
depends on the density and composition of the distribution around the falling line. In Italy
the numbers who fall into poverty because of unemployment outweigh the numbers
brought out of poverty by the falling line. In Ireland, where the proportions brought into
poverty by unemployment are almost as large, these are more than outweighed by the
numbers rising above the falling line.
Even when using the fixed poverty line, in some countries (Luxembourg, Portugal,
Denmark and Sweden) increasing unemployment does not result in a significant increase
in poverty. In these countries, the unemployed are protected from poverty either by the
incomes of other household members or by the operation of the tax and benefit systems (or
some combination of the two). According to Figure 1, less than 5% of the poor are
unemployed in those countries. On the other hand, in Finland, the Netherlands and Italy,
this proportion is above 15%.
(b) Real earnings growth
Not surprisingly, earnings growth without changes in tax and benefit rules reduces the
numbers with incomes below a fixed poverty line. However, as shown in Figure 4, the
extent of the reduction varies from just 0.3 percentage points in Sweden and 0.4 in
Finland to 2.7 in Portugal and 2.0 in Luxembourg and Greece. In countries where there
are few working poor (see Figure 1), the effect on poverty measured with a fixed line is
likely to be small. Using the relative poverty line, recalculated after earnings growth, the
impact is such as to increase the poverty rate in all countries except Luxembourg. The
rise in median incomes offsets the effect of increasing the earned income of some of the
poor. The net effect is negligible in Spain and very small in Greece and Italy. The largest
percentage point increases in the relative poverty rate are found in Ireland (3.1), the UK
(2.5), the Netherlands (2.2) and Denmark (2.1).
Figure 5 relates the percentage point increase in the poverty rate to the proportional
increase in the poverty line. It shows that the poverty rate is less sensitive to changes in the
1
In fact in Denmark the poverty rate falls slightly: the Danish unemployed have a lower risk of poverty than
the general population.
192
H. Immervoll et al.
Percentage point change in the poverty rate
2.5
2.0
2.0
1.7
1.6
1.5
1.0
0.5
0.8
0.5
0.3
0.2
0.2
1.0
0.9
0.7
0.8
0.5
0.1
0.0
0.4
0.1
0.0
0.0
-0.2
-0.5
-0.2
-0.2
-0.1
-0.1
-0.2
-0.2
-0.6
-1.0
-1.5
-0.1
-0.4
-0.5
-1.3
-2.0
-2.5
PT
IR
IT
GR
Unemployment: fixed line
UK
SP
BE
FR
LU
AT
DK
NL
FI
GE
SW
Unemployment: within scenario line
Percentage point change in the poverty rate
Fig. 2. Percentage point change in the poverty rate due to increasing unemployment, using fixed and
relative (within-scenario) poverty lines.
Source: EUROMOD (see Appendix 1 for country codes).
1.0
IT
0.5
BE
0.0
SP
FR
UK
-0.5
FI
SW
GE
NL
DK
GR
AT
LU
PT
-1.0
IR
-1.5
-4.00
-3.50
-3.00
-2.50
-2.00
-1.50
-1.00
-0.50
0.00
Percentage change in the relative poverty line
Fig. 3. Change in the relative poverty line and poverty rate with an increase in unemployment of
5 percentage points.
Source: EUROMOD (see Appendix 1 for country codes).
poverty line in Luxembourg, Spain, Greece and Portugal. In part, this can be explained by
the fact that relatively large proportions of the poor in these countries are in work (see
Figure 1). On the other hand, in countries with low rates of working poor (Finland and
Ireland) the effect of the upward shift in the poverty line is dominant, and poverty rates rise
substantially.
(c) An increase in earnings inequality
Increasing earnings inequality while maintaining the same mean value has the effect
of reducing incomes in poor households, to the extent that the low paid live in low income
households. Thus poverty rates rise if the poverty line is held constant. Figure 6 shows that
poverty rates increase by most in the countries with greater levels of working-poor
(Portugal, Luxembourg, Greece and Spain), suggesting that these countries also have
Macro-level changes in the European Union
193
Percentage point change in the poverty rate
4.0
3.1
3.0
2.5
2.2
2.1
2.0
1.6
1.3
1.7
1.2
0.8
1.0
0.3
0.6
0.5
0.4
0.0
0.0
-0.1
-1.0
-0.6
-0.3
-1.1
-1.5
-1.8
-2.0
-3.0
-1.0
-1.0
-1.5
-2.0
-0.4
-0.7
-0.7
-0.8
-2.0
-2.7
-4.0
PT
IR
IT
GR
UK
Earnings growth: fixed line
SP
BE
FR
LU
AT
DK
NL
FI
GE
SW
Earnings growth: within scenario line
Percentage point change in the poverty rate
Fig. 4. Percentage point change in the poverty rate due to real earnings growth, using fixed and relative
(within-scenario) poverty lines.
Source: EUROMOD (see Appendix 1 for country codes).
3.5
IR
3.0
UK
2.5
2.0
1.5
0.0
-0.5
4.00
PT
FR
1.0
0.5
DK NL
GE
FI
BE
IT
LU
4.50
5.00
SW
AT
GR
SP
5.50
6.00
6.50
7.00
7.50
Percentage change in the relative poverty line
Fig. 5. Change in the relative poverty line and poverty rate with a real increase in earnings.
Source: EUROMOD (see Appendix 1 for country codes).
concentrations of working households just above the poverty line. Poverty rates rise by least
in Sweden (0.8 percentage points), Denmark (1.8) and Belgium (1.9).
If the poverty line is based on the within-scenario median, the net effect is to increase the
poverty rate in some countries and to reduce it in others. The percentage point increase is
particularly strong in Luxembourg (4.2), Greece (3.1) and Spain (2.5), and the decrease is
strong in Ireland (3.5) and the UK (3.3).
Figure 7 shows the relationship between the shift in the poverty line and the change in
the relative poverty rate. There appear to be some distinct groups of countries. Greece
and Spain show relatively large downward shifts in their poverty lines and increases in
their poverty rates. In contrast, UK and Ireland also show a relatively large shift in the
line but the rates fall. One contributory factor is the protection offered to the low paid by
in-work benefits in these two countries. Another is that both have a large number of
194
H. Immervoll et al.
Percentage point change in the poverty rate
10.0
8.1
8.0
7.1
6.4
6.3
6.0
5.1
4.2
3.7
4.0
3.5
3.5
3.1
2.4
2.0
2.5
2.9
1.9
2.4
2.3
1.8
1.2
1.0
0.6
0.2
0.8
0.1
0.1
-0.1
0.0
-0.9
-1.1
-2.0
-4.0
-1.0
-3.3
-3.5
-6.0
PT
IR
IT
GR
UK
Earnings inequality: fixed line
SP
BE
FR
LU
AT
DK
NL
FI
GE
SW
Earnings inequality: within scenario line
Percentage point change in
the poverty rate
Fig. 6. Percentage point change in the poverty rate due to an increasing earnings inequality, using fixed
and relative (within-scenario) poverty lines.
Source: EUROMOD (see Appendix 1 for country codes).
5.0
LU
4.0
3.0
SP
2.0
1.0
0.0
GR
IT
PT
FR
AT
NL
-1.0
SW
GE
FI
DK
BE
-2.0
-3.0
-4.0
-15
IR
-13
-11
-9
UK
-7
-5
-3
Percentage change in the relative poverty line
Fig. 7. Change in the relative poverty line and poverty rate with an increase in earnings inequality.
Source: EUROMOD (see Appendix 1 for country codes).
elderly people—who are unlikely to be directly affected by reductions in earnings—with
incomes just below the 1998 (baseline) poverty line.1
Figure 6 shows that poverty rate changes in Portugal and Luxembourg are very similar
using a fixed poverty line, but very different when using a relative poverty line. An
explanation for this emerges from Figure 7. While in Luxembourg the fall in the poverty
line is relatively small (5.4%), in Portugal the fall is so large (14.1%) that the rise in poverty
among workers is compensated by a reduction of the poverty rate among the elderly and
others without earnings.
1
The poverty rate among people aged 65þ falls by 17 percentage points in Ireland, 8 in the UK, 2 in Spain
and 0 in Greece. See Mantovani and Sutherland (2003) for further details and Feres et al. (2002) for a more
comprehensive discussion of the effects by age group.
Macro-level changes in the European Union
195
5. Conclusions
Our results confirm that the poverty indicator can be sensitive to the types of macro changes
that we have considered. The extent of sensitivity does differ markedly across countries. It is
clear that detailed micro-level simulations are required to establish the separate effects of
a shift in the poverty line and changes in the income of those at risk of poverty.
This is clearest in the case of rising unemployment, a scenario that is intuitively
associated with a rise in the proportion of households with low income. The relative
poverty rate falls slightly or remains unchanged in 11 out of 15 countries following
a simulated increase in unemployment of 5 percentage points.
Real earnings growth without changes to tax and benefit policies increases the relative
poverty rate in all countries, but the effect is negligible in some countries and substantial
in others.
Increasing earnings inequality has a variable impact across countries. Although most
employed people’s earnings fall under this scenario and the poverty rate using the fixed line
rises in all countries, the relative poverty rate rises in some countries and falls in others.
What can governments do to minimise the negative impact of macro changes on the
headline social indicator (and, more broadly, on those at risk of poverty and social
exclusion)? Our results suggest that:
Real earnings growth places an upward pressure on relative poverty indicators, particularly in Denmark, Ireland, the Netherlands and the UK. Tax allowances and
thresholds and benefit levels should keep pace with the growth in median incomes if
relative poverty is to be controlled. Alternatively, other measures to protect those at risk
of poverty need to be introduced and reviewed on a continuing basis, to compensate
for fiscal drag.1
Unemployment, as simulated, has a variable effect overall on relative poverty rates, and the
net national effects tend to be small. Clearly, governments should be concerned about
the social and economic effects of unemployment, aside from any impact on relative
poverty. Generally, some combination of the incomes of other household members and
the operation of tax–benefit systems offers some protection against poverty to the
unemployed. The combination varies across countries.
The earnings inequality scenario is an example of increasing the income of the rich at the
expense of the poor. In some countries, this results in a decrease in relative poverty.
Clearly, governments should not encourage the reduction of wages as a mechanism to
reduce relative poverty. At the same time, it is worth noting the role of in-work benefits
in UK and Ireland in partly protecting low income earners from the effect of falling
wages.
These last two points illustrate the dangers of relying on the relative poverty rate as a single
indicator and highlight the importance of maintaining a portfolio which includes
(a) indicators that relate directly to individual labour market experience (such as
unemployment or low wages) as well as household incomes
(b) indicators of absolute changes in real income level
(c) indicators calculated for sub-groups.
1
The fact that this scenario raises revenue makes this a feasible strategy that has been adopted in practice in
the UK (Hills and Sutherland, 2004).
196
H. Immervoll et al.
It is also important to acknowledge that the simulated scenarios should not be considered
as predictions of what would happen in any particular country in the event of an increase
in unemployment or change in the level or distribution of earnings. Real changes are
more complicated, and it is unlikely that they would take place in isolation (earnings
growth may well occur in combination with increasing inequality), that they would apply
so uniformly within a country (new unemployment may be concentrated in particular
regions or sectors), or that they would in fact occur in the same way across countries.
However, the precise pattern in which such changes occur is impossible to predict. The
point of this ‘forward-looking’ exercise has been to explore the mechanics of the
relationships between plausible macro-level changes and the ‘headline’ EU social
indicator, and to draw out the implications for cross-country monitoring of the evolution
of the indicator over time. Having established that macro-level changes can have
important consequences for evaluating progress towards social inclusion, similar methods
can, at a later stage, be used to assess what part of observed changes has in fact been due
to tax and benefit policy reform.
We believe that, if poverty rates based on median household incomes are to be used as
generally accepted measures of the outcomes of policy to promote social inclusion, it is
important that their sensitivities to other influences are fully understood. Indeed, it has yet
to be established what improvement in these indicators is realistic to expect, and to what
extent the same improvement is possible across countries (Atkinson et al., 2003, p. 25).
This paper has demonstrated that the ‘headline’ poverty indicator used by the EU—the
proportion living in households with incomes below 60% of the equivalised median—is
potentially vulnerable to both unemployment and growth, in certain circumstances.
Bibliography
Atkinson, A. B., Cantillon, B., Marlier, E. and Nolan, B. 2002A. Social Indicators: The EU and
Social Inclusion, Oxford, Oxford University Press
Atkinson, A. B., Cantillon, B., Marlier, E. and Nolan, B. 2002B. Indicators for social inclusion,
Politica Economica, 18, no. 1, 7–47
Atkinson, A. B., Marlier, E. and Nolan, B. 2003. ‘Indicators and Targets for Social Inclusion
in the European Union’, Economic and Social Research Institute Working Paper 151,
Dublin
Dennis, I. and Guio, A-C. 2003. Poverty and social exclusion in the EU after Laeken- part 1,
Statistics in Focus, Theme 3 8/2003, Eurostat
Feres, P., Immervoll, H., Levy, H., Lietz, C., Mantovani, D. and Sutherland, H. 2002.
‘Indicators for Social Inclusion in the European Union: how responsive are they to macro-level
changes?’, EUROMOD Working Paper EM3/02, Department of Applied Economics,
University of Cambridge
Gomulka, J. 1992. Grossing-up revisited, in Hancock, R. and Sutherland, H. (eds), Microsimulation Models for Public Policy Analysis: New Frontiers, STICERD Occasional Paper 17,
LSE
Harding, A. (ed.) 1996. Microsimulation and Public Policy, Amsterdam, North Holland
Hills, J. and Sutherland, H. 2004. ‘Ending Child Poverty in a Generation? Policies and Prospects
in the UK’, paper presented at the Conference on Cross-National Comparisons of Expenditures on Children, Princeton University, January
Immervoll, H. 2004. ‘Falling up the Stairs. An Exploration of the Effects of ‘‘Bracket Creep’’ on
Household Incomes’, EUROMOD Working Paper EM3/04, Department of Applied Economics, University of Cambridge
Immervoll, H. and O’Donoghue, C. 2001. ‘Imputation of Gross Amounts from Net Incomes in
Household Surveys: An Application using EUROMOD’, EUROMOD Working Paper EM1/
01, Department of Applied Economics, University of Cambridge
Macro-level changes in the European Union
197
Immervoll, H., O’Donoghue, C. and Sutherland, H. 1999. ‘An Introduction to
EUROMOD’, EUROMOD Working Paper EM0/99, Department of Applied Economics,
University of Cambridge
Jäntti, M. and Jenkins, S. P. 2001. ‘Examining the Impact of Macro-Economic Conditions on
Income Inequality’, ISA Discussion Paper No. 364, ISA, Bonn
Mantovani, D. and Sutherland, H. 2003. ‘Social Indicators and other Income Statistics using the
EUROMOD Baseline: A Comparison with Eurostat and National Statistics’ EUROMOD
Working Paper EM1/03, Department of Applied Economics, University of Cambridge
Messere, K. 1998. The Tax System in Industrialized Countries, Oxford, Oxford University Press
Mitton, L., Sutherland, H. and Weeks, M. (eds) 2000. Microsimulation Modelling for Policy
Analysis, Cambridge, Cambridge University Press
Nolan, B. 1987. Income Distribution and the Macroeconomy, Cambridge, Cambridge University
Press
Parker, S. C. 1998–99. Income Inequality and the Business Cycle: a survey of the evidence and
some new results, Journal of Post Keynesian Economics, vol. 21, no. 2, 201–25
Social Protection Committee. 2001. ‘Report on Indicators in the Field of Poverty and Social
Exclusion’
Sutherland, H. (ed.) 2001. ‘EUROMOD: an integrated European Benefit-tax model, Final
Report’, EUROMOD Working Paper EM9/01, Department of Applied Economics, University of Cambridge
Sutherland, H. 2002. ’Indicators for Social Inclusion in the European Union: the impact of
policy changes and the use of microsimulation models, Politica Economica, vol. 18, no. 1,
117–20
Note: This paper was written as part of the MICRESA (Micro Analysis of the European Social
Agenda) project, financed by the Improving Human Potential programme of the European
Commission (SERD-2001-00099). Horacio Levy is grateful for support from the Departament
d’Universitats, Recerca i Societat de la Informacio de la Generalitat de Catalunya. We are
indebted to our former colleague Cathal O’Donoghue for his invaluable contribution to the
construction of the EUROMOD model under project CT97-3060 and to all other past and
current members of the EUROMOD consortium. We are particularly grateful for assistance
from Patricio Feres and comments from Tony Atkinson, Roxana Gutierrez, Deborah Mabbett,
Panos Tsakloglou and Michael Wolfson, the participants of the MICRESA meeting in Athens
in May 2002 and three referees. The views expressed in this paper, as well as any errors, are
the responsibilities of the authors. In particular, this applies to the interpretation of model
results and any errors in its use. EUROMOD is continually being improved and updated and
the results presented here represent work in progress. EUROMOD relies on micro-data from
12 different sources for 15 countries. These are the European Community Household Panel
(ECHP) User Data Base made available by Eurostat; the Austrian version of the ECHP made
available by the Interdisciplinary Centre for Comparative Research in the Social Sciences; the
Panel Survey on Belgian Households (PSBH) made available by the University of Liège and
the University of Antwerp; the Income Distribution Survey made available by Statistics Finland;
the Enquête sur les Budgets Familiaux (EBF) made available by INSEE; the public use version
of the German Socio Economic Panel Study (GSOEP) made available by the German Institute
for Economic Research (DIW), Berlin; the Living in Ireland Survey made available by the
Economic and Social Research Institute; the Survey of Household Income and Wealth
(SHIW95) made available by the Bank of Italy; the Socio-Economic Panel for Luxembourg
(PSELL-2) made available by CEPS/INSTEAD; the Socio-Economic Panel Survey (SEP) made
available by Statistics Netherlands through the mediation of the Netherlands Organisation for
Scientific Research—Scientific Statistical Agency; the Income Distribution Survey made available by Statistics Sweden; and the Family Expenditure Survey (FES), made available by the
UK Office for National Statistics (ONS) through the Data Archive. Material from the FES is
Crown Copyright and is used by permission. Neither the ONS nor the Data Archive bears
any responsibility for the analysis or interpretation of the data reported here. An equivalent
disclaimer applies for all other data sources and their respective providers cited in this
acknowledgement.
198
H. Immervoll et al.
Appendix 1: EUROMOD base datasets
Base Dataset
for EUROMOD
Country
AT
Austria
BE
Belgium
DK
Denmark
FI
Finland
FR
France
GE
Germany
GR
IR
IT
LU
NL
PT
SP
SW
UK
Austrian version
of the European
Community Household
Panel (W5)
Panel Survey
on Belgian
Households (W6)
European
Community
Household Panel (W2)
Income distribution
survey
Budget de Famille
German SocioEconomic Panel (W15)
Greece
European Community
Household Panel
(W3)
Ireland
Living in Ireland
Survey (W1)
Italy
Survey of Households
Income and Wealth
Luxembourg PSELL-2 (W5)
Netherlands Sociaal-economisch
panelonderzoek (W3)
Portugal
European Community
Household Panel (W3)
Spain
European Community
Household Panel (W3)
Sweden
Income distribution
survey
UK
Family Expenditure
Survey
Type
Date of
collection
Reference
time period
for incomes
ECHP
1999
annual 1998
National Panel
1997
annual 1996
ECHP
1995
annual 1994
Registerþsurvey
1998
annual 1998
Household
Budget Survey
National Panel
1994/5 annual 1993/4
1998
annual 1997
ECHP
1996
annual 1995
National Panel
1994
month in 1994
Income survey
1996
annual 1995
National Panel
National Panel
1999
1996
annual 1998
annual 1995
ECHP
1996
annual 1995
ECHP
1996
annual 1995
Registerþsurvey
1997
annual 1997
Household Budget
Survey
1995/6 month in 1995/6
Appendix 2: Increasing unemployment
The aim is to inflate the weights of households containing unemployed people while keeping the
aggregate counts of other key characteristics constant.1 For our purposes, the unemployed are
defined as people aged 19–59 declaring themselves to be out of work and looking for a job, plus
any others in receipt of unemployment benefits during the period covered by the underlying data.
The within-database national ‘unemployment rate’ is calculated as the ratio of these unemployed
to those in the labour force, defined as the unemployed plus people aged 19–59 in receipt of
earnings or self-employment income.
1
We are very grateful to Joanna Gomulka for facilitating access to her grossing-up software (Gomulka,
1992).
Macro-level changes in the European Union
199
It is worth noting that differences in underlying data cause our estimates of the unemployed not to be comparable across all countries. The main source of difference arises from
the extent to which the recipients of benefit are the same as the people declaring themselves to
be unemployed. Where income data refer to the same period as the status variables (Ireland and
UK), the two groups will overlap more than where income variables refer to the previous year. In
this latter case—applying in most countries—incomes may in fact refer to a period of nonunemployment. Thus, our term ‘unemployment’ might be more accurately described as ‘at risk
of unemployment’.
The increased total number of unemployed people is calculated by adding 5 percentage points
to the ‘unemployment rate’ within each country, calculated as described above.
Household weights already exist, supplied with the national datasets. The baseline weights
have been calculated to adjust for sample design and/or differential non-response (see Sutherland, 2001, for details). Weights are then recalculated using the existing weights as a starting
point, but (a) using the increased number of unemployed as the control for unemployed, and (b)
also controlling for demographic and household composition variables, and region, using the
existing grossed-up totals for these categories as control totals. The specific variables used as
controls are:
Individual variables
Number aged 0–18 (¼ children)
Males aged 19–24
Females aged 19–24
Males aged 25–49
Females aged 25–49
Males aged 50–59
Females aged 50–59
Males aged 60þ
Females aged 60þ
Household variables
Households with 1 adult aged 19–59 only
Households with 2 adults aged 19–59 only
Households with 1 adultþ1 or more children only
Households with 2 adultsþ1 or more children only
Other households with children
Other households without children
Region
This method implies that the households without any unemployed people that are similar to
households with unemployed people (according to the above variables) will have their weights
reduced. In other words, these are the households who are ‘made unemployed’ in our exercise.