ArticlePDF Available

Back to Bismarck: Shifting Preferences for Intragenerational Redistribution in OECD Pension Systems

July 2014

Authors:

Tim Krieger

University of Freiburg

Stefan Traub

Helmut Schmidt University

Changes of Bismarckian factor and generosity index

…

classifies the countries according to their changes in the Bismarck-

…

: Classification of countries

…

also reports the annual change in life expectancy in terms of

…

: Bivariate correlations

…

Figures - uploaded by Tim Krieger

Content may be subject to copyright.

Content uploaded by Tim Krieger

Content may be subject to copyright.

Back to Bismarck?

Shifting Preferences for Intragenerational Redistribution

in OECD Pension Systems

Tim Kriegera,∗, Stefan Traubb

aDepartment of Economics, University of Paderborn, Germany

bDepartment of Economics, University of Bremen, Germany

May 2010

Abstract Using a sample of 20 OECD countries it is shown that the ma-

jority of countries decreased the level of intragenerational redistribution in

the ﬁrst pillar of their pension systems, though the evidence is weak in sta-

tistical terms. We ﬁnd strong correlations between changes of the so-called

Bismarckian factor and changes of the generosity of the pension system, the

shape of the income distribution in terms of its ﬁrst three central moments

and life expectancy. An economic laboratory experiment conﬁrms that these

variables could have been causal for the observed change.

JEL classiﬁcation: H55, D71, J18, D63, C92

Keywords: earnings-related and ﬂat-rate beneﬁts, Beveridge vs. Bismarck,

pension reform, relative deprivation, OECD countries, experiments

∗Corresponding author. Department of Economics, University of Paderborn, War-

burger Str. 100, 33098 Paderborn, Germany, phone +49-5251-602117, fax +49-5251-

605005, email tim.krieger@notes.upb.de.

1 Introduction

Recent studies suggested that the degree of intragenerational redistribution

in the ﬁrst pillar of many OECD countries’ pension systems has been de-

creasing over the last two decades (see, e.g., Fenge et al., 2003; Lindbeck and

Persson, 2003; Queisser, 2000; and Werding, 2003). Our study empirically

tests this hypothesis using microdata taken from the Luxembourg Income

Study (LIS). In order to estimate the level of intragenerational redistribu-

tion in the public pension system, we employ the Bismarckian factor (see,

e.g., Hassler and Lindbeck, 1997; and Cremer and Pestieau, 1998). Concep-

tionally, the Bismarckian factor divides the pension beneﬁt into a ﬂat com-

ponent (such as a basic or minimum pension) and into an earnings-related

component: the higher the Bismarckian factor, the more important is the

earnings-related part and, thus, the smaller is the degree of intragenerational

redistribution (under the proviso that contributions are collected as payroll

taxes). A pension system that emphasizes the earnings-related component

(such as in France or Germany) is called a Bismarckian pension system. If

the pension system accentuates the ﬂat beneﬁt part (such as in the UK), it

is called Beveridgean.

Our study considers pension systems in 20 OECD countries. Depending

on data availability, we used data for the time period from 1985 to 2000.

In line with the studies mentioned above, we empirically observe for most

countries an increase of the Bismarckian factor that was accompanied by

an increase of the “generosity” of the pension system (the share of pension

beneﬁts in total household income).1In 14 countries, the Bismarckian fac-

1The negative correlation between the level of intragenerational redistribution and the

tor increased; in 11 cases both the level of intragenerational redistribution

decreased and the generosity increased. It should be noted, however, that

the increases of the Bismarckian factor as well as the generosity index are

statistically insigniﬁcant at conventional terms. Hence, though we approve

our initial question that OECD pension systems actually tend to move“back

to Bismarck”, the empirical evidence is relatively weak.

The core of our analysis, however, is formed by the identiﬁcation of eco-

nomic factors that through political processes could exert inﬂuence on the

level of intragenerational redistribution in a country’s pension system. In our

paper, we mainly focus on two aspects that are likely to be important. First,

during the observation period, the income distributions of the 20 OECD

countries considered have changed dramatically. Not only real per capita

GDP has risen, but also the variance and the skewness of the income distri-

bution have increased. It does not seem to be too farfetched to assume that

these developments have somehow inﬂuenced the perceived level of inequal-

ity in the society and shaped redistribution policy (though the causality may

be mutual). Conde-Ruiz and Profeta (2007) recently emphasized the deci-

sive role of inequality for the level of intragenerational redistribution in the

pension system. With substantial inequality, a winning coalition of the rich

and the poor could implement a Beveridgean pension system, while a low

degree of inequality would allow the middle-class to introduce a Bismarckian

system.2Second, the average life expectancy of a male person at the age of

size of the pension system also was investigated by, for example, Cremer and Pestieau,

1998; Casamatta et al., 2000a, 2000b; K¨

othenb¨

urger et al., 2008; and Rossignol and Tau-

gourdeau, 2006).

2This outcome resembles to some degree the “paradox of redistribution” by Korpi and

65 has increased by about one and a half months per year. However, the

gain in life years was probably not uniformly distributed. Evidence for a

positive eﬀect of wealth on life expectancy has been provided, for example,

by Attanasio and Emerson (2003) or Deaton and Paxson (2001). So, the rel-

ative importance of retirement arrangements has increased, and the increase

has been more intense for the rich. A positive eﬀect of income on life ex-

pectancy is known to make a pension system regressive (see Coronado et al.,

2000; Gil and Lopez-Casanovas, 1998; and Reil-Held, 2000). These ﬁndings

also are consistent with Borck (2007) and Gorski et al. (2007) who showed

formally that a positive correlation between income and longevity dampens

redistribution from rich to poor in pension systems.

We choose a twofold approach to investigate these questions. In the ﬁrst

part of the analysis, we present empirical facts based on LIS data as to the

change of OECD pension systems, income distributions, and life expectancy.

Results are reported in terms of descriptive statistics and a correlation analy-

sis. Due to data limitations and unclear causal relationships of the empirical

analysis, the second part of the paper presents an economic laboratory ex-

periment where subjects had to chose the Bismarckian factor representing a

hypothetical social security system. The experiment involved 18 treatments

mimicking exogenous changes in the generosity of the pension system, in

the shape of the income distribution and both symmetric and asymmetric

increases of life expectancy.

In our experiment, we basically followed the so-called individual choice

Palme (1998), which has been supported empirically in the context of pension systems by

Lef`ebvre (2007).

approach to social welfare which was developed by Friedman (1953) and

Harsanyi (1953, 1955).3Our subjects were induced by the experimental de-

sign to act as involved social planners. They evaluated income distributions

from under a veil of ignorance, that is, they became a member of the re-

spective society after having done their choices, but they did not know their

own future income positions in advance. Note that the experiment involved

considerable monetary payoﬀs of up to e1,050. The individual choice ap-

proach was also employed by Bernasconi (2002), Bosmans and Schokkaert

(2004), and Traub et al. (2005, 2008) in order to test various hypotheses

concerning perceptions of justice and the consistency of choice behavior in

income distribution contexts. In terms of the individual choice approach,

income distributions resemble lotteries. The principle of insuﬃcient reason

implies that the social planner imagines that the chance of being any person

in the society is equally likely. Equal consideration is given to every member

of the society, thereby guaranteeing impartiality of the social planner. Ac-

cordingly, the social planner is assumed to choose that income distribution

that maximizes the expected utility.

Apart from theoretical concerns (see, for example, Mongin, 2001), the

literature suggests that the relationship between a social planner’s distribu-

tional preferences and income distributions is more complex than stated by

3The large existing body of experimental literature concerned with redistribution prob-

lems comprises, for example, Amiel and Cowell, 1992, 2000; Ballano and Ruiz-Castillo,

1993; Frohlich and Oppenheimer, 1994; Harrison and Seidl, 1994a, 1994b; Bernasconi,

2002; Bosmans and Schokkaert, 2004; Traub et al., 2005, 2008. However, to our best

knowledge this is the ﬁrst study to contrast empirical ﬁndings concerning the change of

institutions such as the pension system with experimental micro data.

the impartial observer theorem. Recent research has shown that inequal-

ity and risk preferences are not the same (on this issue, see Cowell and

Schokkaert, 2001). Bernasconi (2002) questioned both the utilitarian ap-

proach to social welfare and the non-utilitarian approach like Rawls’ (1971)

maximin criterion to be a meaningful description of distributional prefer-

ences. Instead, he found some evidence of randomization preferences, that

is, a procedural fairness motive (Diamond, 1967) which violates the betwee-

ness axiom of expected utility theory. Likewise, in an experimental “beauty

contest” of social welfare functions Traub et al. (2005) demonstrated that a

quadratic social welfare function (Epstein and Segal, 1992) which expresses

randomization preferences performed remarkably well. As a consequence of

such procedural fairness motives entering the self-interested social planner’s

preference, she is likely to avoid extreme outcomes in terms of insuﬃciently

low or excessively high incomes (see Traub et al., 2008).

A hypothesis that goes remarkably well with these observations is Bould-

ing’s (1962, p. 83) proposition that “society lays a modest table at which

all can sup and a high table at which the deserving can feast”. Boulding’s

hypothesis translates into a lexicographic social welfare function. First, the

social planner takes care that everyone in the society (including himself)

obtains enough to make ends meet. The income threshold associated with

it may be called living wage, subsistence level, or poverty line. Then, for

incomes above that threshold, the social planner’s preferences are best ex-

pressed by maximizing the average or expected utility of the society (as

assumed by Friedman-Harsanyi). In the ignorance scenario (no probabil-

ity information was given) of the Traub et al. (2005) “beauty contest” of

social welfare functions, the Boulding social welfare function was the best

performing standard of behavior among all subjects; it was among the four

top performers in the risk scenario.

Further complications seem to arise from the fact that the evaluation of

incomes is context dependent. Whether an income is perceived as barely

suﬃcient or excellent does not only depend on its absolute value but also on

the background context in terms of the overall shape of the income distri-

bution. The concept of relative deprivation, tracing back to Stouﬀer et al.

(1949) and advanced by Runciman (1966) and others, provides the missing

link between income evaluation and societal context.4Generically, relative

deprivation describes a subjectively perceived lack of something in relation

to a societal reference value (for example, a hot breakfast). The link between

relative deprivation and a society’s degree of inequality aversion was high-

lighted by the philosopher Temkin (1986, 1993) who argued that inequality

aversion was driven by the complaints of the poor about their situation in

relation to that of the rich. Devooght (2003) found experimental support

for Temkin’s model. Seidl et al. (2006) used Parducci’s (1965, 1968, 1974,

1982) range-frequency theory in order to experimentally investigate the con-

text dependence of the assessment of income distributions. They showed that

positively skewed income distributions (which happen to be the normal case

in OECD countries) generate more relative deprivation at the societal level

than negatively skewed income distributions. Our empirical observation that

the skewness of the OECD countries’ income distributions annually increased

4The literature on the economics of happiness (see, e.g., Oswald, 1997; and Easterlin,

2001) heavily relies on the concept of relative deprivation.

by more than 3% since 1985 leaves us anticipating an upswell of complaint.

The rest of the paper is structured as follows. In Section 2, we formally

discuss the Bismarckian factor and present the empirical analysis based on

LIS microdata for 20 OECD countries. The experiment is presented in Sec-

tion 3. Section 4 provides an extensive discussion of our results. Section 5

concludes.

2 Empirical Facts

2.1 Deﬁning the Bismarckian factor

In order to derive a workable and intuitive representation of the level of intra-

generational redistribution in a pension system, we deﬁne the Bismarckian

factor along the lines of the “index of non-contributiveness” (INC) (Lef`ebvre

and Pestieau, 2006; Lef`ebvre 2007). INC, denoted by β, is deﬁned as the

ratio of the income share of public pensions in the bottom quintile, B, to the

same share in the top quintile, T:

β≡PB/YB

PT/YT

=PB

·YT

,(1)

where Yiand Pi,i∈ {B, 2,3,4, T }, are the mean income and the mean

pension beneﬁt, respectively, of the ith quintile of the income distribution.

A purely Beveridgean pension system which pays equal beneﬁts to every

citizen implies PB=PT, such that βbev =YT/YB(≥1). If beneﬁts are

solely earnings-related, the respective purely Bismarckian pension system

yields PB/YB=PT/YTand, therefore, βbis = 1. Note that due to β∈

[1, YT/YB] INC is not normalized which is a bit unfavorable for cross-country

comparisons.

We use the deﬁnition of the pension beneﬁt of a representative member of

quintile iin order to derive the Bismarckian factor. Piis deﬁned as a convex

combination of a ﬂat payment (proportional to the mean income) and an

earnings-related component (see Casamatta et al., 2000a):

Pi=τ[αYi+ (1 −α)µ],(2)

where α∈[0,1] is the Bismarckian factor and µ≡PiYi/5 is the mean of

the society’s income distribution. A measurement of the generosity of the

pension system τ∈[0,1] reﬂecting the replacement ratio5is given by

τ≡PiPi

PiYi

.(3)

Henceforth, we will refer to τas the “generosity index”.

We plug equation (2) into the ratio of the pension beneﬁts of the bottom

and the top quintile PB/PT(the left fraction in the deﬁnition of INC). Since

τdrops out, solving this expression for αgives the Bismarckian factor

α≡(PT−PB)·µ

(PT−PB)·µ−PTYB+PBYT

.(4)

A purely Beveridgean pension system (PB=PT) yields αbev = 0 and a purely

Bismarckian pension system (PB/YB=PT/YT) gives αbis = 1. Hence, as

desired the Bismarckian factor is normalized6on the closed interval [0,1] and

5Note that this parameter could be interpreted as the pension system’s replacement

ratio, thereby also capturing inter generationally redistributive elements in the pension

formula, which are, however, not explicitly build into our approach.

6Alternatively, we can write α≡[(PT−PB)·µ]/[(PT−PB)·µ+PTYB·(β−1)],

highlighting that the Bismarckian factor is in fact a normalization of the INC.

is also independent of the generosity τof the pension system. Accordingly, α

is not only a pure measure of intragenerational redistribution but also allows

for cross-country comparisons of public pension systems of diﬀerent size.

2.2 The Data

For the empirical analysis, we used microdata taken from the Luxembourg

Income Study (LIS, 2008). It provides internationally comparable and reli-

able data on income distributions (see Atkinson, 2004). The LIS data were

employed for computing the values of the Bismarckian factor αand the gen-

erosity index τ. Furthermore, LIS data were used in order to compute the

ﬁrst three central moments of the income distribution (mean, variance, skew-

ness).

The following countries were included in our data set: Austria (ﬁrst year:

1985, last year: 2003), Australia (1987, 2000), Belgium (1985, 2000), Canada

(1987, 2000), Denmark (1987, 2004), Finland (1987, 2000), France (1984,

2000), Germany (1984, 2000), Greece (1995, 2000), Ireland (1987, 2000), Italy

(1986, 2000), Luxembourg (1985, 2000), Mexico (1984, 2002), the Nether-

lands (1983, 1999), Norway (1986, 2002), Spain (1990, 2000), Sweden (1987,

2000), Switzerland (1982, 2002), the United Kingdom (1986, 1999) and the

United States (1986, 2000). Apart from rare exemptions in terms of radical

system changes, large aggregates such as pension systems transform them-

selves only gradually. Hence, we looked at the longest available time period

for each country. In most cases, the earliest wave providing us with the nec-

essary data was Wave II (around 1985), the latest wave available at the time

this research was conducted was Wave V and Wave VI in some cases (around

2000).

All LIS data employed in our analysis refer to the household level. The

household is the most natural economic unit to focus on as its members

jointly plan on earning and spending income. At the household level, we

have to distinguish between “raw” and equivalized household net income. In

order to compute the Bismarckian factor as well as the generosity of the pen-

sion system, we used “raw” household net income. That is, αand τmeasure

the legal status of the pension system as it was reﬂected in the respective

income distribution. In the subsequent empirical analysis, we shall explore

correlations between changes in the income distribution and the pension sys-

tem. Therefore, when computing the moments of the income distribution,

we employed equivalized household net income. By adjusting the income dis-

tribution for the diﬀerent needs of diﬀerent household types, we based this

part of our analysis on household welfare. Note that we used the household

weights provided by LIS in order to weight cases. Furthermore, if necessary,

income data was adjusted for inﬂation using the consumer price index with

the last LIS year of the respective country as the base year (data source:

OECD Main Indicators).

LIS reports household net income in an aggregate variable (LIS variable:

DPI). For computing the moments of the income distribution, it was ad-

justed for diﬀering needs by using the square root of household size (D4) as

weights. As recommended by LIS, the equivalized income data were bottom-

and top-coded.7The variable “state old-age and survivors beneﬁts” (V19)

7Equivalized household net incomes smaller than 1% of the mean equivalized income

were recoded as 1% of the mean equivalized income. Household net incomes larger than 10

provided us with the data required for computing the (unequivalized) mean

pension beneﬁts of the income quintiles.8Note that V19 provides a broad

measure of intragenerational redistribution, including — in addition to min-

imum pensions — diﬀerent non-insurance beneﬁts such as beneﬁts due to

education, unemployment, maternity etc. (see, for example, B¨

orsch-Supan

and Reil-Held, 2001).9

2.3 Changes of αand τ

Figure 1 displays the changes of the Bismarckian factor and the generosity

index. Each of the 20 countries considered has two markers referring to the

ﬁrst and the last year of analysis (country codes are listed in Table 1). The

horizontal axis refers to the generosity index, while the vertical axis states

the Bismarckian factor. Sweden kept the most generous pension system: in

1987 the share of pensions in total household net income was about 27.9%.

As compared to this, Mexico’s 1984 pension system was virtually negligi-

ble (2.9%). In the year 2000 France maintained the pension system with

the lowest degree of intragenerational redistribution (α= 0.764). In some

cases, the Bismarckian factor turned out to be outside the theoretical [0,1]

times the median unequivalized incomes were recoded as ten times the median household

net income weighted by the square root of household size.

8Note that the disaggregated variables V19S1a (“universal old-age pensions”) and

V19S1b (“employment-related old-age pensions”) would have been more suitable but were

available only from Wave IV on.

9Some of these beneﬁts, such as beneﬁts due to education times, may be regressive.

Therefore, there may be a slight downward bias in the Bismarckian factor as compared to

a scenario where only minimum pensions are considered.

interval: for Australia and Finland in the past and for Denmark and Nor-

way in the present, we recorded slightly negative values for the Bismarckian

factor. These countries had or still have an almost purely Beveridgean pen-

sion system, where additionally high-income earners do not receive the (full)

minimum pension.

Table 1 classiﬁes the countries according to their changes in the Bismarck-

ian factor and the generosity index. The generosity of most pension systems

seems to have increased. Likewise, for six countries only (Denmark, Greece,

Luxembourg, the Netherlands, Norway and Sweden), we had to record a

decreasing Bismarckian factor.10

One might wonder whether the observed changes of αand τare signiﬁ-

cant. In order to test this, we computed the annual changes of both variables

in percentage points (this proceeding allows for country-speciﬁc time spans).

Table 2 shows that neither the change in α(about 0.5 percentage points per

year) nor the change in τ(about 0.1 percentage points per year) are signif-

icant at conventional terms. Another interesting result, which can be taken

from Table 3, is that both, the absolute levels and the changes of Bismarckian

factor and generosity of the pension system were highly correlated. As dis-

cussed in the literature, less redistributive pension systems are usually larger

(though this relationship seems to have diminished a bit). Furthermore, most

pension reforms have taken place on the main diagonal of Figure 1 (or Table

10Note that in Lindbeck and Persson (2003) and Werding (2003), Denmark, Greece and

Sweden are considered as countries that reduced intragenerational redistribution in their

pension systems. At least in Sweden the relevant major pension reforms were enacted at

the time of collecting the data for Wave V. Hence, these reforms are not covered by our

data.

Figure 1: Changes of Bismarckian factor and generosity index

Table 1: Classiﬁcation of countries

Generosity Bismarckian factor

index increased decreased

increased Austria (AU) Denmark (DK)

Belgium (BE) Greece (GR)

Canada (CA) Luxembourg (LU)

Finland (FI) Norway (NO)

France (FR)

Ireland (IE)

Italy (IT)

Mexico (MX)

Spain (ES)

Switzerland (CH)

United States (US)

decreased Australia (AT) Netherlands (NL)

Germany (DE) Sweden (SE)

United Kingdom (UK)

Table 2: Changes of Bismarckian factor and generosity index

Variable Mean Level of

signiﬁcance

Bismarckian factor (α) 0.524 0.106

Generosity index (τ) 0.109 0.165

Table notes. n= 20. Annual changes in percent-

age points.

Table 3: Correlation between Bismarckian factor and generosity of the pen-

sion system

Variables Coeﬃcient Level of

signiﬁcance

αpast,τpast 0.673 0.001

αpresent,τpr esent 0.499 0.025

∆α, ∆τ0.578 0.008

Table notes. Pearson correlations. n= 20.

1), that is, increases of αcame along with increases of τ.

Technically, the reduction of the level of intragenerational redistribution

came about in very diﬀerent forms in the latest pension reforms (see, for

example, Casey et al., 2003). The most fundamental change certainly being

the switch from a deﬁned-beneﬁt to a notional deﬁned-contribution system as

in Italy and Sweden, which is the“most Bismarckian”speciﬁcation among the

diﬀerent types of pay-as-you-go systems. Other reforms included altering the

reference earnings with respect to which pensions are calculated, for example,

by moving from“best years” to a “period average”, by increasing the number

of contribution years required to obtain a full pension, or by changing the

method of calculating the reference earnings. All these reforms tightened the

link between individual earnings and future pension beneﬁts.

Summarizing this part of the analyis, we are inclined to answer the ques-

tion: “Back to Bismarck?” in the aﬃrmative as most countries have un-

dertaken pension reforms that reduced intragenerational redistribution. It

should be kept in mind, however, that as far as the average trend is con-

cerned the result is rather weak in statistical terms.

2.4 Changes in the Shape of the Income Distribution

and Life Expectancy

In this subsection, we consider changes of the mean, the coeﬃcient of vari-

ation11 and the skewness (the standardized third central moment) of the

income distribution. Furthermore, we consider changes in life expectancy.

11We employ the coeﬃcient of variation (standard deviation divided by mean) instead

of the variance because it is scale invariant.

These are captured by the residual life expectancy of a male aged 65. The

respective data is taken from the 2005 OECD Health Data set. Table 4

presents the changes in the shape of the income distribution and life ex-

pectancy over time. Additionally, we report the changes in household size

and the Gini coeﬃcient (the Gini coeﬃcient is listed in the LIS “keyﬁgures”).

Figures are stated in terms of annual changes in percent (income, household

size), percentage points (Gini coeﬃcient12) or expected life years, allowing

for the diﬀerent time spans that we had available for the 20 OECD countries.

Mean equivalized household net income on average increased by 1.6% per

year. Note that the average household size shrunk signiﬁcantly by more than

0.5% per year and thus contributed to the increase of equivalized household

income. Without this demographic eﬀect, the annual increase in mean in-

come would have been only 1.23% which still is signiﬁcant at the 1% level.

The coeﬃcient of variation annually rose a bit more than 0.6% (signiﬁcant

at the 10% level). Much more pronounced was the change of the income

distribution is terms of its skewness. On average, OECD countries’ income

distributions have become signiﬁcantly more positively skewed, as the an-

nual increase of the skewness coeﬃcient indicates (3.1%). Both coeﬃcient

of variance and skewness reﬂect a strong and signiﬁcant increase in income

inequality. As can be taken from the table, the additionally reported Gini

coeﬃcient increased by a bit less than 0.1 percentage points per year. The

relatively modest increase in the Gini coeﬃcient as compared to coeﬃcient of

variation and the skewness is easily explained. Due to the transfer principle,

12Since the Gini coeﬃcient is normalized on the [0,1] interval, we consider absolute

instead of relative changes.

Table 4: Change of income distribution and life expectancy over time

Variable Mean Level

of Signiﬁcance

Mean incomea1.609 0.001

Coeﬃcient of variationa0.641 0.063

Skewnessa3.104 0.008

Life expectancyc0.127 0.000

Household sizea-0.547 0.000

Gini coeﬃcientb0.092 0.052

Table notes. n= 20. Annual changes. aPercent.

bPercentage points. cExpected life years.

a rise in the variance of the income distribution unambiguously makes the

Gini coeﬃcient larger. This increase, however, may be counteracted by a

left shift of the median income: if the skewness of the income distribution

increases, a measurement of lower inequality at the bottom of the income

distribution is implied.

Table 4 also reports the annual change in life expectancy in terms of

expected life years of a person at the age of 65 is given. The coeﬃcient of

0.127 corresponds to a strong increase in life expectancy of about 1.5 months

per year.

2.5 Correlation Analysis

Table 5 gives the correlations between the variables listed in Table 4 and Bis-

marckian factor αand generosity index τ, respectively. The Bismarckian fac-

tor exhibits positive correlations with mean income, coeﬃcient of variation,

and life expectancy. We observe negative correlations between Bismarckian

factor and skewness as well as household size. However, only life expectancy

is signiﬁcant at the 5% level. Interestingly, mean, coeﬃcient of variation,

and skewness exhibit positive bivariate correlations, where the correlation

between the coeﬃcient of variation and skewness is particularly strong and

signiﬁcant at the 1% level. Hence, although coeﬃcient of variation and skew-

ness both increased and were highly correlated, their bivariate relationship

with the Bismarckian factor was oppositional.

In contrast to the Bismarckian factor, the generosity of the pension sys-

tem shows a negative but insigniﬁcant correlation with mean, coeﬃcient of

variation, and skewness. Only life expectancy shows a slightly positive corre-

lation. Life expectancy exhibits positive correlations with mean income and

coeﬃcient of variance and a negative correlation with skewness. As expected,

(the change in) household size is negatively correlated with (the change in)

mean income which is, to some extent, an eﬀect of using equivalized house-

hold income.

Of course, bivariate correlations can only give a ﬁrst impression of the

complex empirical relationships between these variables. In order to learn

more about the change of the pension system in OECD countries, we per-

formed OLS estimations with the Bismarckian factor and the generosity of

the pension system as the left-hand variables and the income variables as

Table 5: Bivariate correlations

Coeﬃcient Life HH

τMean of variation Skewness Expectancy Size

α0.578** 0.021 0.179 −0.266 0.472* −0.184

τ—−0.175 −0.150 −0.199 0.076 −0.159

Mean — 0.326 0.243 0.186 −0.352

C. of Var. — 0.745** 0.089 −0.285

Skewness — −0.131 −0.046

Life Exp. — −0.007

Table notes. n= 20. Pearson correlations. Annual changes of the variables

in percentage points (α,τ), percent (mean, coeﬃcient of variance, skewness,

household size) or expected life years. **p≤.01, *p≤.05.

well as life expectancy as the right-hand variables.13 We avoid the terms en-

dogenous and exogenous variables since we do not claim at this point of the

analysis that there is any clear causal relationship among them. Note that

we used regression through the origin because zero changes in the right-hand

variables should be associated with zero changes in the pension systems.14

Table 6 presents the results of the OLS regressions. In this section, we

only give a brief account of the ﬁgures stated in the table. We will discuss

them in detail together with the results of the experiment in the Section 4.

First, we comment on the change of the Bismarckian factor. The overall ﬁt

of the regression is satisfactorily high for a cross-section of only 20 countries.

The coeﬃcient for the annual change in mean income is close to zero and

insigniﬁcant. As it seems there was no direct relationship between the sig-

niﬁcant rise of household welfare in terms of equivalized household income

and the level of intragenerational distribution in the pension system. Like-

wise, the change of household size did not have an impact on changes of

α. A change of the coeﬃcient of variation by 1 percent came along with an

increase of the Bismarckian factor by about 0.7 percentage points, while a

rise in the skewness of the income distribution by 1 percent was associated

with a decrease of the Bismarckian factor of almost 0.25 percentage points.

Both variables are highly signiﬁcant. The relationship between increased life

expectancy and change of the Bismarckian factor is particularly strong, ex-

hibiting a coeﬃcient of 7.61 percentage points per additional expected life

13Since this is a case of seemingly unrelated regression with identical regressors, we

estimated both equations separately.

14As a robustness check, we also performed regressions including an intercept. In none

of the regressions, the intercept was signiﬁcant.

year. Unfortunately, no data was available as to the correlation between life

expectancy and income. This relationship will be explored in the experiment.

The regression for the generosity index is insigniﬁcant. None of the vari-

ables entering the regression exhibits a signiﬁcant correlation with τ. Hence,

the main channel through which changes in the income distribution and life

expectancy may have exerted an inﬂuence on the design of the pension sys-

tem is the level of intragenerational redistribution rather than its generos-

ity. This ﬁnding contrasts with Conde-Ruiz and Profeta (2007) who in their

model used both channels contemporaneously (yet, issue-by-issue) in order

to determine the design of the pension system.

The correlation analysis presented in this section has a number of draw-

backs. First, as already mentioned, it remains unclear whether the change in

intragenerational redistribution is actually caused by changes in the shape of

the income distribution and life expectancy or vice versa. Second, the num-

ber of observations is too low in order to run more sophisticated regressions,

taking into account the countries’ heterogeneity and other factors of inﬂu-

ence. Hence, in the next section, we shall present a laboratory experiment

that avoids both problems.

3 The Experiment

3.1 Experimental design

The experiment was fully computerized. It consisted of two parts. In the ﬁrst

part, the subjects were presented the decision task. In the second part, we

collected their sociodemographics and attitudes. Each subject had only one

Table 6: Results of OLS regression

Bismarckian factor Generosity index

Variable Coeﬀ. pCoeﬀ. p

Mean −0.075 0.654 −0.045 0.387

Coeﬃcient of Varia-

tion

0.733 0.016 −0.027 0.761

Skewness −0.248 0.009 −0.004 0.877

Life Expectancy 7.610 0.024 1.063 0.278

Household Size 0.009 0.984 0.145 0.314

F= 3.399, p= 0.030 F= 0.803, p= 0.565

R2= 0.531 R2= 0.211

Table notes. Variables entered the regression in terms of annual changes in per-

centage points (Bismarckian factor and generosity of the pension system), per-

cent (mean, coeﬃcient of variation, skewness), or life years (life expectancy).

n= 20. Regression through the origin.

Abbildung 1: Änderung des Bismarckfaktors in 20 OECD-Ländern

-0,200 0,000 0,200 0,400 0,600 0,800

Bismarckfaktor Vergangenheit

-0,200

0,000

0,200

0,400

0,600

0,800

Bismarckfaktor Gegenwart

Denmark

Greece

Luxembourg

Netherlands

Norway

Sweden

Australia

Austria

Belgium

Canada

Finland

France

Germany

Ireland

Italy

Mexico

Spain

Switzerland

United Kingdom

United States

Abbildung 2: Beispiel einer Entscheidungsaufgabe

Figure 2: Sample screen of the decision task

decision problem to solve. Figure 2 presents a sample screen of the decision

problem. There were ﬁve columns on the screen. Under each column the

number of“winning points”(labelled “Punkte”) was displayed, corresponding

to the height of the column. The subjects were told that the points resemble

“Mr./Mrs. A to E’s” claims against some (non-speciﬁed) social insurance sys-

tem. Below the winning points, the row labeled “Euro” gave the information

how the winning points would be exchanged for money at the end of the

experiment. This payoﬀ should be interpreted as the actual pension beneﬁt

after redistribution.

By using the control on the lower part of the screen, the participants could

choose the degree of redistribution of (pension) claims of ﬁve hypothetical

persons, ranging from zero to 100 percent. Initially, the control was set at

zero; this is equivalent to a Bismarckian factor of one, that is, there is a per-

fectly proportional relation between winning points and payoﬀs. By turning

the control to the right, the share of winning points which are redistributed

among individuals increases to up to 100 percent, implying a Bismarckian

factor of zero, that is, a pure Beveridgean pension system. The redistribution

of winning points was highlighted by spotted areas within the columns.

The participants were asked to choose the distribution of payoﬀs they

“liked best”by setting the control appropriately. Thereby the following payoﬀ

rules had to be taken into account: at the end of the experiment two groups,

each including ﬁve subjects, were randomly picked. Each of the selected

subjects was randomly assigned to an income position in its group (denoted

by“Mr./Mrs. A to E”) and given the respective cash payment. The payment

resulted from the income position and the median αof the group. That

is, the Bismarckian factor of each small society was determined by majority

vote from the ﬁve individual values set by the control. This simple incentive

structure is preference-revealing. Strategic considerations, such as coalition

building, could not play a role for the subjects’ decisions due to anonymous

data collection and randomized sampling of groups.

Remember that the index of generosity, the coeﬃcient of variation, the

skewness of the income distribution, and life expectancy were signiﬁcantly

correlated with the Bismarckian factor (in terms of changes of the respective

variables). The mean of the income distribution was not correlated with the

Bismarckian factor and will therefore be neglected in the following. Hence,

the experiment involved 18 treatments. We varied

– the factor, τ, at which winning points were exchanged into cash pay-

ments in order to test for the eﬀects of increasing the generosity of the pension

system; {low generosity, high generosity }

– the inequality of the distribution of winning points with respect to vari-

ance, σ, and skewness, λ, in order to test the eﬀect of increasing income

inequality; {low variance and symmetric income distribution, high variance

and symmetric income distribution, low variance and positively skewed in-

come distribution}

- the risk of not receiving a payment (beneﬁt), π, in order to test for the

life-expectancy eﬀect. {no risk, symmetric risk, risk negatively correlated

with number of winning points }

The “no-risk” scenario was conducted according to the previously de-

scribed rules. In the risk scenario, we use the fact that dying after a cer-

tain fraction πof the (ﬁxed) retirement period is equivalent in terms of the

expected value of pension beneﬁts to not receiving the maximum retirement

income with the same probability (see, for example, Diamond, 2003). Hence,

in the case of symmetric risk one out of ﬁve subjects did not receive a pay-

ment, implying a lower average life expectancy of the entire group.15

When risk was negatively correlated with income, the probability of not

receiving a beneﬁt was – as before – on average 20 percent; however, the indi-

vidual probability was calculated according to the formula πi=i/ P5

j=1 j,

where iis the rank in a descending ordering of the distribution of winning

points. In Figure 2, this scenario is indicated by the “Risiko” (risk) row.16 A

complete list of all treatments and the chosen parameters is given in Table

15Since the focus of our analysis is on intragenerational redistribution rather than the

intertemporal aspects of the pensions system, this approach appeared us to be a reasonable

short-cut. It also avoids problems as to the experimental design linked to intertemporal

choice such as discounting.

16Note that probabilities are rounded.

After the decision task, a standardized questionnaire had to be answered

by the subjects. In the questionnaire, we asked for the ﬁeld of study (ordered

by schools) as well as some knowledge and attitude questions, which will be

explained in detail in Section 3.3. Furthermore, at the beginning of the

experiment subjects had to indicate their sex.

3.2 Procedure

The experiment took place in the cafeteria of the University of Bremen on

July 2nd and 3rd, 2007. Interested students were informed about a scientiﬁc

study on social insurances. Furthermore, they were told that a show-up fee

of e5 was to be paid, that participation would take about 10 minutes, and

that there was a chance of winning up to e1,050. The subjects drew a

lot with a ﬁve-digit number. The ﬁrst three digits determined the treatment

according to Table 7, the fourth and ﬁfth digit gave the group number within

a treatment and the individual income position within the group, respectively.

However, the subjects were not given any information about the meaning of

the number, which was – together with the sex – the initial input necessary

to start the experiment.

At the end of the experiment, that is, after answering the questionnaire, a

lottery started. The ten lot numbers of the winning groups were selected by

an umpire (our secretary) before the experiment and saved on the computers.

Whenever the lot number of the subject coincided with a predetermined

number, the subject was informed that he or she is a winner. In case of a

risk scenario the winning subjects were reminded that due to a second lottery

Table 7: Parametrization of treatments

Risk of dying

None Symmetric Correlated

Distribution Generosity (π= 0) (π= 0.2) (πi=i/ P5

j=1 j)

Low variance, low 111 121 131

symmetric (τ= 0.1)

distribution high 211 221 231

(vY= 0.35, λ= 0) (τ= 0.3)

High variance, low 112 122 132

symmetric (τ= 0.1)

distribution high 212 222 232

(vY= 0.53, λ= 0) (τ= 0.3)

Low variance, low 113 123 133

positively skewed (τ= 0.1)

distribution high 213 223 233

(vY= 0.35, λ= 0.97) (τ= 0.3)

Table note. The ﬁgures in the table refer to the treatment number.

=coeﬃcient of variation of income distribution, λ.

=skewness.

there may not be a payment despite being a winner. After collecting all data,

the median of the control values (or individual preferences for redistribution,

respectively) of all ﬁve group members was determined. Based on this median

value, the individual payments were calculated. All subjects participating in

the experiment received, independent of whether being a winner or not, the

show-up fee. Per treatment two groups of ﬁve subjects each took part. Hence,

in total 180 students participated in the experiment. Show-up fees summed

up to 900 Euros. The ten winners received a total of e3,379, although one

subject in a risk-treatment did not receive a payment.

3.3 Results

On average, the individually preferred Bismarckian factor αof all subjects

was 0.61.17 There was no signiﬁcant diﬀerence (t-test: p= 0.334) between

men (61 percent of the sample, α= 0.60) and women (39%, α= 0.63). There

was a strong correlation between the average individual Bismarckian factor

chosen by the subjects and the expected average αof the other subjects,

which also had a value of 0.61 (Pearson correlation: ρ= 0.441, p < 0.01).

The correlation with the level of general basic income support, considered as

necessary for Germany by the subjects (mean: 643 Euros, standard deviation:

247 Euros), and αwas as expected negative but insigniﬁcant (ρ=−0.028,

p= 0.711). Between schools there were no signiﬁcant diﬀerences (F-test: p=

0.303) although some schools had a tendency for a below-average α(social

sciences: 0.53, education: 0.50) or above-average α(production engineering:

0.69).

17Germany’s present Bismarckian factor is 0.56 according to our microdata analysis.

Because in the beginning of the experiment, subjects were only told that it

deals with some non-speciﬁed social insurance system, we asked what subjects

believed to be “the social insurance”. The results were mixed: 29% thought

of the pension insurance (mean Bismarckian factor: 0.60) followed by health

insurance (26%, 0.62), unemployment insurance (18%, 0.61), long-term care

insurance (6%, 0.66) and accident insurance (4%, 0.54). Only few subjects

chose systems which – in a narrow sense – are not related to social insurance,

such as social aid (11%, 0.65) or“others” (6%, 0.48). There were no signiﬁcant

diﬀerences between answer groups with respect to the Bismarckian factor

(F-test, p= 0.301). Neither was there a signiﬁcant diﬀerence (F-test, p=

0.921) between the answers on the question whether responsibility for old-age

provision should be private (8%, 0.62), public (14%, 0.62) or jointly (78%,

0.60). The same insigniﬁcance (F-test, p= 0.437) could be found for the

self-assessment regarding risk attitude with the categories risk-averse (50%,

0.62), risk-neutral (32%, 0.58) and risk-loving (18%, 0.62). On average, the

subjects estimated the employees’ share in social insurance contributions in

Germany to be 24 percent.18 Again, there was no signiﬁcant correlation

with the Bismarckian factor (ρ=−0.106, p= 0.158), although there was

a slight tendency that subjects who estimated a high share preferred more

redistribution.19

The descriptive results presented so far are neither representative for the

entire population nor is it permissible to refer to the absolute level of the

18The actual employee’s share was about 21% in Germany in 2007.

19We did not ask for age, religion etc. because in our student population, the variation

of these variables is very low (most Bremen students are protestants and between 20 and

25 years old).

Bismarckian factor. In fact, homogeneity of the random sample allows com-

paring the diﬀerent treatments. Any diﬀerences in the preferred level of

redistribution result exclusively from the exogenous variation of the stimuli.

Whenever individuals respond to monetary incentives, these treatment eﬀects

in the laboratory turn out to be evidence for analogous eﬀects of changes of

the income distribution and life expectancy in the real-world.

Because the public pension system and thus the level of intragenerational

redistribution within the pension system are based on democratic majority

votes, we do not use the individual α’s but – in line with the incentive struc-

ture of the experiment – the median α’s of diﬀerent groups. As noted in

the Introduction, the individual decisions regarding the Bismarckian factor

were made from behind a Rawlsian “veil of ignorance”. This implies that

subjects were assigned their positions in the society only at the end of the

experiment. This should guarantee that subjects take on a neutral position

and choose only societally beneﬁcial levels of economic inequality. Further-

more, in reality there exists a substantial degree of uncertainty about one’s

own future economic situation. Another advantage of the “veil of ignorance”

is that subjects had no information about their fellow group members. This

anonymity allows to construct by permutation from the ten group members of

each of the 18 treatments a total of 252 group observations. Accordingly, we

had 4,536 independent observations of homogenous groups in our regression

analysis.

Table 8 shows the results of two OLS regressions in which the Bismar-

ckian factor (in percent) is the endogenous variable. Since the Bismarckian

factor is constrained to the closed [0,1]-interval, we report estimates both for

Table 8: Treatment eﬀects (robust OLS regression)

Untransformed Logit transformed

Variable Coeﬃcient Level of Coeﬃcient Level of

signiﬁcance signiﬁcance

Constant 57.414 0.000 0.302 0.000

High generosity 3.610 0.000 0.178 0.000

High variance 1.980 0.000 0.128 0.000

Positively skewed dis-

tribution

-5.286 0.000 -0.253 0.000

High life expectancy -0.583 0.204 -0.018 0.394

Life expectancy posi-

tively correlated with

income

2.683 0.000 0.140 0.000

n,¯

R24,536 0.075 4,536 0.089

F,p75.05 0.000 89.66 0.000

Breusch-Pagan χ2,p627.22 0.000 659.15 0.000

Table note. Dependent variable: Bismarckian factor (in percent). In the

logit-transformed model, 21 observations with a Bismarckian factor of 0 were

assigned a logit of −3. White’s heteroscedasticity corrected covariance matrix

was used to compute standard errors.

the original variable and the logit-transformed variable. We also control for

heteroscedasticity by computing standard errors using White’s robust covari-

ance matrix. Exogenous were the treatment dummies high generosity, high

variance, positively skewed income distribution, high life expectancy, and

life expectancy positively correlated with income. The benchmark case is a

treatment with low generosity, low variance and symmetric income distribu-

tion, and low life expectancy uncorrelated with income. While the regression

explains only a small part of total variance in the data (in experiments the

noise is usually quite large), it is nevertheless highly signiﬁcant.

Since both regressions yield almost identical results, we comment only on

the untransformed regression, which has an easier interpretation with regard

to the estimated coeﬃcients.20 The average Bismarckian factor (in percent)

was about 58% in the benchmark case. Increasing the exchange rate of

payments for winning points (that is, moving to a “high generosity” scenario)

raised the Bismarckian factor signiﬁcantly by approximately 3.6 percentage

points. Increasing the variance of the income distribution, increased the

Bismarckian factor signiﬁcantly by approximately two percentage points. In

the treatments with a more positively skewed income distribution, αwas

signiﬁcantly smaller (by about 5.3 percentage points). In contrast to the

empirical analysis, our experiment allows to diﬀerentiate between symmetric

and asymmetric changes of life expectancy. While the symmetric increase

of life expectancy had no signiﬁcant eﬀect on the Bismarckian factor, an

asymmetric change had a signiﬁcant positive eﬀect on the Bismarckian factor

20In the logit-transformed model, the coeﬃcients state the change in the log of the

odds-ratio with respect to an absolute change in the exogenous variables.

Table 9: Summary of results

Factor LIS Experiment

Generosity ↑ ↑

Mean ↔—

Variance ↑ ↑

Skewness ↓ ↓

Life expectancy ↑symmetric: ↔

correlated with income: ↑

Table note. Impact of changes in the left-hand variables on α.

(2.1 percentage points).

4 Discussion

In this section we compare and interpret the results gained from the analysis

of the LIS data and the experiment. Table 9 provides a stylized overview.

The analysis of the LIS data showed a strong positive correlation be-

tween the size of the pension system in terms of the generosity index and

the Bismarckian factor. A more generous pension system came along with

less intragenerational redistribution. Such a negative correlation between

the level of intragenerational redistribution and the size of the pension sys-

tem also has been investigated by, for example, Cremer and Pestieau (1998),

Casamatta et al. (2000a, 2000b), K¨

othenb¨

urger et al. (2008), and Rossignol

and Taugourdeau (2006). Our empirical result is clearly conﬁrmed by the

experiment, where the change of the generosity index was a pure treatment

eﬀect, that is, fully exogenous. How does this result relate to Boulding’s

(1962) hypothesis of a “modest table” and a “high table”?

According to the LIS data the pension beneﬁt of the bottom quintile

increased by 1.935% (std. error: 0.577) per year – that is less than the mean

pension beneﬁt of all pensioners (2.624%, s.e.: 0.972) but a bit more then

the mean of (equivalized) household net income (1.609%, s.e.: 0.405; see

Table 4). This highlights – with the caveat that the mean diﬀerences are not

statistically signiﬁcant (the former mean diﬀerence is 0.689 with p= 0.230;

the latter one is 0.327 with p= 0.594)21 – that Boulding’s “modest table”

grows with increasing wealth of the society rather than being absolutely ﬁxed.

It points to the fact that in OECD countries poverty is more a relative than an

absolute concept. In other words: income support is given in order to allow

people to participate in a society’s usual activities (although at a lower level),

and less as a means of avoiding poverty in the sense of famine, malnutrition,

and homelessness. On the other hand, the minimum pension seems to have

grown less than the mean pension such that the relative distance between

minimum pension and mean pension beneﬁt has signiﬁcantly increased by

4.3% (p=0.084).

Result 1: The minimum pension (Boulding’s “modest” table) is

a weakly superior good.

Both parts of the analysis brought about a negative relationship between

21One should keep in mind that the LIS analysis considers intragenerational redistri-

bution in a broader sense than just minimum pensions, some of the transfers even being

regressive, such that the pure minimum pension change may be stronger.

intragenerational redistribution and changes in the variance of the income

distribution. It is straightforward to show that the degree of inequality in

the pension system (in terms of the coeﬃcient of variation) is related to the

degree of inequality in the income distribution as follows:

vP=αvY.(5)

Accordingly, vPis linear homogenous in vY, and an increase in αthat is

caused by an increase in vYunambiguously leads to an increase in inequality

of the pension system. For example, consider the experiment’s benchmark

case with Y= (1,000; 1,500; 2,000; 2,500; 3,000) and vY= 0.3536. The

high-variance scenario with Y0= (500; 1,250; 2,000; 2,250; 3,500) increased

the variation coeﬃcient by 50% (vY0= 0.5303) and induced our subjects to

state α’s about 1.980 percentage points higher. As a consequence of this,

inequality in the pension system rose from vP= 0.2051 to vP0= 0.3181, that

is, by about 55%.

However, this observation seems to contradict the literature. Conde-Ruiz

and Profeta (2007) recently argued that a winning coalition of the rich and

the poor (high inequality condition) could implement a Beveridgean pen-

sion system, while a low degree of inequality would allow the middle-class

to introduce a Bismarckian system. Also from the perspective of the im-

partial observer model, this result comes as a surprise. An isolated increase

in the variance is equivalent with a regressive transfer. Consequently, with

a concave welfare function one would expect to see lower instead of higher

Bismarckian factors.

Although we neglect the usual equity-eﬃciency trade-oﬀ in our experi-

ment, social planners and real world politicians will have to take this trade

oﬀ into account, naturally limiting the level of intragenerational redistribu-

tion in a society. In an experiment with considerable ﬁnancial incentives,

Traub et al. (2008) showed that subjects put relatively high weight on eﬃ-

ciency consideration (in terms of Pareto dominance) as compared to equity

consideration (in terms of transfer and Lorenz dominance) when faced with

an equity-eﬃciency trade-oﬀ. Note also that from an empirical point of view

the transfer principle is highly controversial. Using questionnaire-based ex-

periments, Amiel and Cowell (1999) showed that between 53 to 74 percent

of their subjects rejected the transfer principle in the diﬀerent contexts of

inequality measurement, poverty, and social welfare.

Amiel and Cowell (1999) concluded that people tend to judge inequality in

terms of income diﬀerence (rather than income levels directly) and the overall

shape of the income distributions (rather than only the incomes involved in

the transfer). We explain the violation of the transfer principle by randomiza-

tion preferences (for experimental evidence see Bernasconi, 2002; and Traub

et al., 2008): self-interested social planners might perceive a trade-oﬀ be-

tween“fairness” in terms of inequality and the chance of excelling the others.

Randomization preferences imply that a probability mixture of two original

(very promising but unequal) income distributions between which the social

planner is indiﬀerent from under the veil of ignorance is preferred over the

two original income distributions. Such preferences violate the betweeness

axiom of expected utility theory (see Chew, 1989) but may be opportune if

the social planner looks for a “fair” procedure (for example, tossing a coin) to

justify “unfair” outcomes. Our empirical and experimental results suggests,

that the weight that is given to the chance of being among the top recip-

ients of pension beneﬁts increases disproportionately high with increasing

background inequality of the income distribution.

Result 2: Increasing inequality of the income distribution in-

duces the social planner to put less weight on outcome fairness

and more weight on procedural fairness. Hence the level of in-

tragenerational redistribution in the pension system is decreased

and, thus, the Bismarckian factor increased.

Table 9 shows that an increase of the skewness of the income distribu-

tion unambiguously decreased the Bismarckian factor. We believe that this

results can be traced back to an increase of relative deprivation. Seidl et

al. (2006) experimentally estimated Parducci’s (1965, 1968, 1974, 1982) so-

called judgement equation which forms part of his range-frequency theory.

Their estimate was given by

Ji=−0.028 + 0.855 ×Ri+ 0.187 ×Fi,(6)

where Jiis the judgement of stimulus i,Ri= (Si−minj{Sj}/(maxj{Sj} −

minj{Sj}) is the range component, Fi= (ri−1)/(N−1) is the frequency

component, riis the rank of stimulus i, and Nis the total number of stim-

uli. Using equation (6), we can easily illustrate how the evaluation of in-

come distributions changes if they become more skewed. We consider the

benchmark case with S= (1,000; 1,500; 2,000; 2,500; 3,000), ﬁrst. Here,

we obtain an average judgement of 0.493. For the income distribution that

is positively skewed (but has the same mean income and variance), S00 =

(1,250; 1,600; 1,750; 2,100; 3,300), we compute only 0.378.22 Temkin’s work

22In principle, the judgement should be constrained to the [0,1]-interval and the weighs

(1986, 1993) suggests that the latter society is less happy. Inequality aver-

sion that is driven by the complaints of the poor about their situation in

relation to that of the rich should therefore be higher in the treatment with

a positively skewed income distribution. Note that Devooght (2003) found

experimental support for Temkin’s model, too.

If inequality aversion – an thus the size of the Bismarckian factor – is

driven by relative deprivation, this creates a paradox: though we actually

observe lower α’s, the amount of relative deprivation that is felt by the social

planner does not decrease. It is straightforward to show, that the judgement

given in equation (6) is independent of α. An intuitive explanation for that

is, that a change of the Bismarckian factor does only inﬂuence the dispersion

of pension beneﬁts but not the skewness of their distribution (the skewness of

the positively skewed distribution is 0.97 as stated in Table 7). In other words:

though relative deprivation might induce redistribution policy, changing the

level of intragenerational distribution is an insuﬃcient measure to decrease

relative deprivation in the society.

Result 3: Increasing the skewness of the income distribution

deepens relative deprivation in the society, augments the social

planner’s preference for intragenerational redistribution and, thus,

decreases the Bismarckian factor. However, increasing the level

of intragenerational redistribution does not change the skewness

of the distribution of pension beneﬁts.

should add up to one. However, the empirical estimate brought about slight deviations

from the theoretical model. For more details on range-frequency theory, we refer to the

article by Seidl et al. (2006).

Finally, we comment on the relationship between life expectancy and Bis-

marckian factor. The LIS data did not let us diﬀerentiate between symmetric

and asymmetric increases of life expectancy. However, there is suﬃcient evi-

dence in the literature for a positive eﬀect of income on life expectancy (see,

for example, Deaton and Paxson, 2001; and Attanasio and Emerson, 2003)

to assume that the LIS data, too, reﬂect this asymmetry. Hence, concerning

the eﬀect of an asymmetric increase of life expectancy on the Bismarckian

factor, empirical and experimental results again are perfectly in line with

each other.

From an economic perspective, increasing life expectancy at a given re-

tirement age implies a higher risk of being poor and having to rely on state

transfers during retirement. The pension system covers the income risks in-

volved with unknown life expectancy. If life expectancy increases across the

board, that is, symmetrically, one would expect the generosity of the pension

system τto increase in order to guarantee the same replacement income as

before. However, though the respective coeﬃcients were positive, the cor-

relation analysis in Section 2.5 showed no signiﬁcant relationship between

generosity index and life expectancy. In case of an increase of life expectancy

which favors the rich, there is an obvious, rational explanation for lowering

the degree of intragenerational redistribution. To work out our argument,

let us assume that life expectancy of the average citizen remains unchanged

and that life expectancy of the rich (poor) increases (decreases). On the one

hand, this implies an increase of the pension system’s generosity because the

rich receive higher beneﬁts than the poor. On the other hand, this eﬀect is

counteracted by a Harsanyi-Friedman type social planner who realizes that

the expected value of pension beneﬁts of the rich (poor) has increased (fallen).

From the social planner’s perspective, it is relatively more proﬁtable to share

“the cake”among the rich pensioners. The ﬂat component of the pension sys-

tem will be reduced in favor of the earnings-related component such that we

observe an increase of the Bismarckian factor and a corresponding reduction

of intragenerational redistribution. Although viewed from a diﬀerent angle,

this is similar to the life expectancy eﬀect described by Borck (2007) and

Gorski et al. (2007).

Result 4: Asymmetric changes in life expectancy in favor of the

rich diminish the expected utility of pensions at the lower end of

the income distribution, reduce the social planner’s preference for

intragenerational redistribution and, thus, increase the Bismarck-

ian factor.

This result is nicely reﬂected in the experimental results reported in Table

8. Comparing the benchmark group with the group that was treated with a

symmetric increase in life expectancy shows that the treatment eﬀect was in-

signiﬁcant (a slightly negative value of −0.583 percentage points is recorded).

There is obviously no reason to change αas the expected utility of the pension

system is independent of it. We do not know the subjects’ utility functions,

but we can directly compare the expected value of the pension beneﬁt under

both treatments. When introducing risk, it dropped from e200 to e160.

The generosity of the pension system was given, so the social planner could

not go against it by increasing “the cake”. Under the asymmetric treatment,

the expected value of the pension system was dependent on the Bismarckian

factor: e6.67×α+e160. Hence, the subjects had to balance the eﬃciency

gain of a higher αwith its side eﬀect of higher inequality. In the experiment,

this yielded a Bismarckian factor more than 2 percentage points higher than

in the symmetric scenario.

5 Conclusions

In this paper, we presented an analysis of the long-term change in OECD

countries’ pension systems. In a ﬁrst step, using microdata drawn from the

Luxembourg Income Study (LIS), we showed that there is some empirical

evidence for a reduction of intragenerational redistribution in public pension

systems, as suggested (but not yet empirically conﬁrmed) in the recent liter-

ature. As a measurement for the level of intragenerational redistribution in

the pension system, we employed the Bismarckian factor. Though the major-

ity of countries decreased intragenerational redistribution, accompanied by

a rise in the generosity of the pension system, the change proved to be in-

signiﬁcant at conventional statistical terms. We would answer the question:

“Back to Bismarck?” in the aﬃrmative. It should be kept in mind, however,

that the empirical evidence is only weak.

The main focus of our analysis was on the factors determining societal

preferences for intragenerational redistribution in public social insurance sys-

tems, the ﬁrst pillar of the public pension system in particular. Our proceed-

ing comprised two diﬀerent analytical steps: a cross-country study based on

LIS data and an economic laboratory experiment. While the empirical anal-

ysis brought about interesting correlations between the Bismarckian factor

and some variables of interest, it had some limitations due to low sample size

and unclear causal relationships. The laboratory experiment had the clear

advantage of enabling us to model changes in potential explanatory variables

as exogenous treatment variables. While it is hardly possible to translate

the subjects’ absolute distributional preferences from the experiment into

the real world, this approach allowed to study causal relations. Furthermore,

while the empirical analysis was naturally limited to 20 observations, the

experiment was conducted with a sample of 180 student subjects from which

we generated by permutation more than 4,500 independent observations.

Our results can be summarized as follows: Empirical analysis and labora-

tory experiment produced identical marginal eﬀects. Factors that increased

the Bismarckian factor (decreased intragenerational redistribution) were in-

creases in the generosity of the pension system and the variance of the income

distribution as well as asymmetric increases of life expectancy in favor of the

rich. Higher skewness of the income distributions decreased the Bismarckian

factor.

We explain our results in the following ways. In OECD countries poverty

is more a relative than an absolute concept, that is, income support is given

in order to allow people to participate in society’s usual activities rather

than just as a means of providing for subsistence. The minimum pension (or

Boulding’s“modest table”) therefore increases with society’s wealth, however,

at a less than proportional rate. The relative importance of intragenerational

redistribution falls and the Bismarckian factor increases (Result 1).

Falling intragenerational redistribution following increasing variance of

the income distribution is somewhat unexpected and can be explained by a

violation of the transfer principle in people’s preferences. Because of ran-

domization preferences the social planner in cases of increasing inequality

puts less weight on outcome fairness and more weight on procedural fairness

which causes the Bismarckian factor to increase (Result 2). Increasing skew-

ness, on the other hand, increases the level of intragenerational redistribution

because relative deprivation in the society becomes more pronounced which

then augments the social planner’s preference for redistribution. Notwith-

standing changing the Bismarckian factor is an inappropriate means of alle-

viating relative deprivation because it reduces only the dispersion of beneﬁts

but not the skewness of their distribution (Result 3).

Increasing life expectancy implies a higher risk of drifting into old-age

poverty and having to rely on state transfers. However, a symmetric increase

of life expectancy for all (income) groups in society would rather increase

generosity than change the level of intragenerational redistribution. Hence,

only an asymmetric change of life expectancy in favor of the rich had a

signiﬁcant impact on the Bismarckian factor. Under these circumstances,

the social planner realizes that the expected value of the pension beneﬁts of

the rich has increased and therefore ﬁnds it more proﬁtable to direct transfers

towards them by strengthening the earnings-related element of the pension

system (Result 4).

Taking fundamental societal developments – in particular globalization

and demographic change – into consideration our results yield the follow-

ing important insights. Both developments tend to strengthen the earnings-

related component of the public pension systems’ ﬁrst pillar, unless they

induce a substantial change of the skewness of the distribution of retirement

incomes. In the process of globalization barriers to trade and factor mobil-

ity are removed. Factor price equalization generates welfare gains for the

countries involved, although the gains may not be shared equally within a

country. In terms of our analysis we expect mean income and (possibly) the

variance of the income distribution to rise which – according to Results 1

and 2 – decreases the preferred level of intragenerational redistribution. The

ageing of societies may move the pension system into the same direction,

depending on whether life expectancy increases asymmetrically or in favor of

the rich, as Result 3 points out. Evidence from past decades shows that life

expectancy increased the Bismarckian factor as if life expectancy changed

asymmetrically.

However, there is one major counterforce to the reduction of intragenera-

tional redistribution which comes from an increase in the skewness of income

distributions. Past data indicates that skewness increased substantially in

OECD countries, and there are few signs that this development will not con-

tinue in the future. Globalization, for example, tends to reduce the labor

share in national income. Upcoming old-age poverty is a major concern in

many countries’ political debate nowadays.

Globalization and ageing may even interact in some respects. The LIS

analysis showed that while changes of the income distribution were insignif-

icant with respect to changes of the level of redistribution, changes of gen-

erosity had a signiﬁcant impact. Globalization leads – in the ﬁrst place –

to an increase of mean income, but not necessarily generosity (although Re-

sult 1 indicates that preferences change accordingly). In an ageing society,

however, we expect the power of the old generation to increase which allows

them to vote in favor of a more generous pension system, as argued in Brown-

ing’s (1975) seminal paper. When times of globalization and ageing coincide

(as projected for the next decades), mean income and generosity increase,

leading to a higher Bismarckian factor. It should be noted that the voting

power (or gerontocracy) argument introduces an element of intergenerational

redistribution into our reasoning.

Under these circumstances the design of future pension systems remains

an unresolved question. Whether pension system will move even further

“back to Bismarck” depends on the strength of the skewness eﬀect relative

to the other impact factors shown to be relevant in our analysis. But even

if pension systems take a turn – induced by society’s concern about relative

deprivation – to becoming more Beveridgean in the future, this will not solve

the problem of relative deprivation due to the inability of the Bismarckian

factor to change the skewness of the distribution of pension beneﬁts.

Acknowledgments

Financial support by the German Federal Pension Insurance’s Research Net-

work on Pensions (FNA) is gratefully acknowledged. Earlier versions of the

paper were presented at the IIPF conference in Warwick, at the FNA work-

shop “Wohlstandsverteilung und gesetzliche Rentenversicherung” in Berlin,

at the annual conference of the Verein f¨

ur Socialpolitik in Graz, at the

EON Ruhrgas conference on “Demographic Change and Public Policy”, at

the EPCS conference in Jena, at the IEF workshop “Modelling the eﬀects of

pensions and other welfare state transfers in an ageing world” in Madrid, at

the Journ´ees d’Economie Publique Louis-Andr´e G´erard-Varet in Marseille,

at the “Beyond Basic Questions” workshop in G¨

ottingen and at the “Finanz-

wissenschaftliches Kolloquium” in Paderborn. The authors would like to

thank Rolf Aaberge, J¨

urgen Faik, Christophe Hachon, Stefan Hupfeld, Tim

K¨

ohler-Rama, Rupert Sausgruber, Diana Sonntag, Sven St¨

owhase, Christian

Traxler, Bj¨

orn Tyrefors, and Andreas Wagener, as well as conference and

seminar participants for helpful comments and discussions.

References

Amiel, Y.; Cowell, F.A. (1992): Measurement of income inequality: Exper-

imental test by questionnaire. Journal of Public Economics 47, 3–26.

Amiel, Y.; Cowell, F.A. (1999): Thinking about inequality. Cambridge:

Cambridge University Press.

Amiel, Y.; Cowell, F.A. (2000): Attitudes to risk and inequality: A new

twist on the transfer principle. Distributional Analysis Discussion Pa-

per No. 56, STICERD, London School of Economics, London.

Atkinson, A.B. (2004): The Luxembourg Income Study (LIS): Past, present

and future. Socio-Economic Review 2, 165–190.

Attanasio, O.; Emmerson, C. (2003): Diﬀerential mortality in the UK. Jour-

nal of the European Economic Association 1, 821–850.

Ballano, C.; Ruiz-Castillo, J. (1993): Searching by questionnaire for the

meaning of income inequality. Revista Espaola de Economa 10, 233–

259.

Bernasconi, M. (2002): How should income be divided? Questionnaire evi-

dence from the theory of ‘impartial preferences’. In: Moyes, P.; Seidl,

C.; Shorrocks, A. (Eds.): Inequalities: Theory, experiments and appli-

cations. Journal of Economics/Zeitschrift f¨

ur National¨

okonomie, Sup-

plement 9, Vienna-New York: Springer, 163–195.

Borck, R. (2007): On the choice of public pensions when income and life

expectancy are correlated. Journal of Public Economic Theory 9, 711–

725.

B¨

orsch-Supan, A.; Reil-Held, A. (2001): How much is transfer and how

much is insurance in a pay-as-you-go system? The German case. Scan-

dinavian Journal of Economics 103, 505–524.

Boulding, K. (1962): Social justice in social dynamics. In: Brandt, R. (ed.):

Social justice, Englewood Cliﬀs: Prentice-Hall, 73–92.

Bosmans, K.; Schokkaert, E. (2004): Social welfare, the veil of ignorance

and purely individual risk: An empirical examination. Research on

Inequality 11, 85–114.

Browning, E.K. (1975): Why the social insurance budget is too large in a

democracy. Economic Inquiry 13, 373–388.

Casamatta, G.; Cremer, H.; Pestieau, P. (2000a): The political economy of

social security. Scandinavian Journal of Economics 102, 503–522.

Casamatta, G.; Cremer, H.; Pestieau, P. (2000b): Political sustainability

and the design of social insurance. Journal of Public Economics 75,

341–364.

Casey, B.; Oxley, H.; Whitehouse, E.; Antolin, P.; Duval, R.; Leibfritz,

W. (2003): Policies for an ageing society: Recent measures and areas

for further reform. OECD Economics Department Working Paper No.

369, OECD Publishing, Paris.

Chew, S.H. (1989): Axiomatic utility theories with the betweeness property.

Annals of Operations Research 19, 273–298.

Conde-Ruiz, I.; Profeta, P. (2007): The redistributive design of social secu-

rity systems. Economic Journal 117, 686–712.

Coronado, J.L.; Fullerton, D.; Glass, T. (2000): The progressivity of social

security. NBER Working Paper No. 7520.

Cowell, F.A.; Schokkaert, E. (2001): Risk perceptions and distributional

judgments. European Economic Review 45, 941–952.

Cremer, H.; Pestieau, P. (2003): Social insurance competition between Bis-

marck and Beveridge. Journal of Urban Economics 54, 181–196.

Deaton, A.; Paxson, C. (2001): Mortality, education, income and inequality

among American cohorts. In: Wise, D. (ed.): Themes in the economics

of aging. Chicago: Chicago University Press for NBER, 129–70.

Diamond, P.A. (1967): Cardinal welfare, individual ethics, and interper-

sonal comparison of utility: Comment, Journal of Political Economy

75, 765–766.

Diamond, P.A. (2003): Taxation, incomplete markets, and social Security.

Cambridge: MIT Press.

Devooght, K. (2003): Measuring inequality by counting “complaints”: The-

ory and empirics. Economics and Philosophy 19, 241–263.

Epstein, L.G.; Segal, U. (1992): Quadratic social welfare functions. Journal

of Political Economy 100, 691–712.

Easterlin, R.A. (2001): Income and happiness: Towards a uniﬁed theory.

Economic Journal 111, 465–484.

Fenge, R.; Gebauer, A.; Holzner, C.; Meier, V.; Werding, M. (2003): Al-

terssicherungssysteme im internationalen Vergleich: Finanzierung, Leis-

tungen, Besteuerung. ifo Beitr¨

age zur Wirtschaftsforschung, Band 10,

Ifo Institute Munich.

Frohlich, N.; Oppenheimer, J. A. (1994): Preferences for income distri-

bution and distributive justice: A window on the problems of using

experimental data in economics and ethics. Eastern Economic Journal

20, 147–155.

Friedman, M. (1953): Choice, chance, and the personal distribution of in-

come. Journal of Political Economy 61, 277–290.

Gil, J.; Lopez-Casasnovas, G. (1998): Life-time redistribution eﬀects of the

Spanish public pension system. Economics Working Paper No. 242,

Universitat Pompeu Fabra.

Gorski, M.; Krieger, T.; Lange, T. (2007): Pensions, education and life

expectancy. Discussion paper No. 07/04 of the DFG Research Group

“Heterogeneous Labor: Positive and Normative Aspects of the Skill

Structure of Labor”, University of Konstanz.

Harrison, E.; Seidl, C. (1994a): Acceptance of distributional axioms: Ex-

perimental ﬁndings. In: Eichhorn, W. (Ed.), Models and measurement

of welfare and inequality. Heidelberg: Springer, 67–99.

Harrison, E.; Seidl, C. (1994b): Perceptual inequality and preferential judg-

ments: An empirical examination of distributional judgments. Public

Choice 19, 61–81.

Harsanyi, J.C. (1953): Cardinal utility and the ethics of welfare. Journal

of Political Economy 61, 434–435.

Harsanyi, J.C. (1955): Cardinal welfare, individualistic ethics, and interper-

sonal comparisons of utility. Journal of Political Economy 63, 309–321.

Hassler, J.; Lindbeck, A. (1997): Optimal actuarial fairness in pension sys-

tems: A note. Economics Letters 55, 251–255.

Korpi, W.; Palme, J. (1998): The paradox of redistribution and strategies of

equality: Welfare states institutions, inequality, and poverty in Western

countries. American Sociological Review 63, 661–687.

K¨

othenb¨

urger, M.; Poutvaara, P.; Profeta, P. (2008): Why are redistributive

social security systems smaller? A median voter approach. Oxford

Economic Papers 60, 275–292.

Lef`ebvre, M.; Pestieau, P. (2006): The generosity of the welfare state to-

wards the elderly. Empirica 33, 351–360.

Lef`ebvre, M. (2007): The redistributive eﬀects of pension systems in Europe:

A survey of evidence. Luxembourg Income Study Working Paper Series

No. 457.

Lindbeck, A.; Persson, M. (2003): The gains from pension reform. Journal

of Economic Literature 41, 74–112.

Luxembourg Income Study (LIS) Micro database, (2008); harmonization

of original surveys conducted by the Luxembourg Income Study asbl.

Luxembourg, periodic updating.

Mongin, P. (2001): The impartial observer theorem of social ethics. Eco-

nomics and Philosophy 71, 147–179.

Oswald, A.J. (1997): Happiness and economic performance. Economic

Journal 107, 1815–1831.

Parducci, A. (1965): Category judgement: A range-frequency model. Psy-

chological Review 72, 407–18.

Parducci, A. (1968): The relativism of absolute judgements. Scientiﬁc

American 219/6, 84–90.

Parducci, A. (1974): Contextual eﬀects: A range-frequency analysis. In:

Carterette, E.C.; Friedman, P. (eds.): Handbook of perception, Vol.

II: Psychological judgement and measurement. New York/San Fran-

cisco/London: Academic Press, 127-41.

Parducci, A. (1982): Category ratings: Still more contextual eﬀects! In:

Wegener, B. (ed.): Social attitudes and psychophysical measurement.

Hillsdale, NY: Lawrence Erlbaum Associates, 88–105.

Queisser, M. (2000): Pension reform and international organizations: From

conﬂict to convergence. International Social Security Review 53, 31–45.

Rawls, J. (1971): A theory of justice. Oxford: Oxford University Press.

Reil-Held, A. (2000): Einkommen und Sterblichkeit in Deutschland: Leben

Reiche l¨

anger? SFB 504 Discussion Paper No. 00-14, University of

Mannheim.

Rossignol, S., Taugourdeau, E. (2006): Asymmetric social protection sys-

tems with migration. Journal of Population Economics 19, 481-505.

Runciman, W.G. (1966): Relative deprivation and social justice. Henley:

Routledge & Kegan Paul.

Seidl, C.; Traub, S.; Morone, A. (2006): Relative deprivation, personal in-

come satisfaction, and average well-being under diﬀerent income distri-

butions. In: McGillivray, M. (ed.): Inequality, poverty and well-being.

Basingstoke: Palgrave Macmillan, 66-90.

Stouﬀer, S.A.; Suckman, E.A.; Devinney, L.C.; Star, S.A.; Williams, R.M.

(1949): The American soldier: Adjustment during army life. Princeton:

Princeton University Press.

Temkin, L.S. (1986): Inequality. Philosophy and Public Aﬀairs 15, 99-121.

Temkin, L.S. (1993): Inequality. Oxford University Press, Oxford.

Traub, S.; Seidl, C.; Schmidt, U.; Levati, V. (2005): Friedman, Harsanyi,

Rawls, Boulding - or somebody else? An experimental investigation of

distributive justice. Social Choice and Welfare 24, 283-309.

Traub, S.; Seidl, C.; Schmidt, U. (2008): An experimental study on individ-

ual choice, social welfare, and social preferences. European Economic

Review (forthcoming).

Werding, M. (2003): After another decade of reform: Do pension systems in

Europe converge? DICE Report: Journal of Institutional Comparisons

1, 11–16.

More Beveridgean or Bismarckian? A comparative analysis of pension systems in selected CEE countries

Chapter

Full-text available

Feb 2016

Edyta Marcinkiewicz

The main aim of this study is to compare and assess tendencies in the pension systems in the Czech Republic, Estonia, Hungary, Poland, Slovakia and Slovenia in respect to the level of intra-generational redistribution. They are examined form the current workers' perspective as well as from the current pensioners' perspective. First, the paper discusses institutional features that classify these pension systems as more Beveridgean or Bismarckian. As shown, today's pension system designs tend to link more closely pension benefits and earnings. These findings are supported by the OECD estimates that prove that the theoretical replacement rates for current workers involve very little redistribution, except for the Czech Republic. Second, using aggregated data obtained from the OECD and Eurostat databases covering the time span 2005-2013 the situation of current pensioners is examined. The results of the analysis based on three indicators of intra-generational redistribution show that pension systems in most studied CEE countries contain a strong redistributive component. Thus, they can be described as Beveridgean. However, Slovenia can be perceived as an exception.

Fiscal decentralization and income (re)distribution in OECD countries’ regions

Article

Full-text available

Jul 2023
Struct Change Econ Dynam

Measuring intra-generational redistribution in PAYG pension schemes

Article

Full-text available

Jan 2022
PUBLIC CHOICE

Voters in ageing societies expect pension reforms to be both inter-generationally and intra-generationally fair. In this paper, we propose a global measure of intra-generational redistribution in pay-as-you-go pension schemes as a basis for voters’ evaluations of reforms. Our novel index only requires information on contributions by and pension benefits paid to retirees, enabling us to measure intra-generational redistribution isolated from possible inter-generational redistribution. We rely on the contribution records of approximately 100,000 Germans, who progressed into retirement in 2007–2015, to measure the level of intra-generational redistribution in the German statutory pension scheme (GRV). A recent reform of the childcare benefit provision, which became effective in 2014, confirms the predictions of our index. The reform introduced additional benefits for a substantial subgroup of German mothers, owing to which the index value for women, but not for men, jumps up. Our findings suggests that GRV fulfills the ideal of a Bismarckian pension system without intra-generational redistribution for men, while women benefit significantly from intra-generational redistribution.

The Retirement Age and the Pension System, the Labor Market and the Economy (Wiek emerytalny a system emerytalny, rynek pracy i gospodarka)

Article

Jan 2021

Measuring Intra-Generational Redistribution in PAYG Pension Schemes

Preprint

Full-text available

Apr 2020

In this paper, we propose a novel index for measuring intra-generational redistribution in pay-as-you-go pension schemes. Our index solely requires information on contributions and pension benefits of retirees, enabling us to measure intra-generational redistribution isolated from possible inter-generational redistribution. We use contribution records of approx. 100,000 German individuals, who progressed into retirement in 2007-2015, to measure the level of intra-generational redistribution in the German statutory pension scheme (GRV). A recent reform of childcare benefit provision, which became effective in 2014, confirms the predictions of our index. The reform introduced additional benefits for a subgroup of substantial size of German mothers, due to which the index value for women, but not for men jumps up. Our findings suggests that GRV fulfils the ideal of a Bismarckian pension system without intra-generational redistribution for men, while women benefit from intra-generational redistribution.

Pension Systems and Labour Market Participation in European Countries

Chapter

Mar 2020

Filip Chybalski

Many countries all over the world have been facing the problem of deteriorating demographics, forcing reforms of the pension system model as well as its main parameters. An important perspective of the assessment of the financial sustainability of a pension system is the division of current GDP among generations. This results from the macroeconomic rules that determine the pension system functioning. The main factor affecting the sustainability and, therefore, the income adequacy of pensioners in the long run is the relationship between the pension system and the labour market. The timing of retirement determines the border between the working age generation and the generation of pensioners. In this chapter, the author attempts to answer the research question whether empirical data covering European countries confirm that there is a linkage between pension systems design or their selected parameters and the condition of labour markets. This question seems to be crucial for policymakers, especially in countries refraining from a systematic increase in the retirement age or that are even planning to decrease it. The method employed to analyse empirical data is mainly based on selected statistical as well as econometric tools.

Working life, gender pension inequality and redistribution - an empirical exanimation of the contribution pension link in Germany

Thesis

Jan 2019

Jonas Klos

This thesis examines important aspects of the connection between contributions and pension benefits in Germany. It encompasses the application and development of new empirical methods and models. This thesis begins by discussing the gap in own old-age incomes of men and women and exploring the causes for these differences by means of decomposition methods using German micro-data. It finds that this gap is mainly explained by differences in labor market experience and education. The gap is especially high at the lower end of the pension income distribution, while the contribution of differing labor market experiences to the explained gap is particularly pronounced for retirees with low pensions. The thesis continues by providing a micro-simulation study on the long-run effects of career interruptions in Germany. Using data of the German Socio-Economic Panel, it finds that career interruptions will, for the average individual, have lifelong effects on incomes and labor-force participation. It quantifies these effects for the average affected individual as well as on the entire society. It therefore provides additional information on the total cost of career interruptions. Finally, this thesis proposes a new index for measuring intra-generational redistribution in PAYG pension schemes. This index solely requires information on contributions and pension benefits of retirees, enabling the measurement of intra-generational redistribution isolated from possible inter-generational redistribution. As an application, contribution records of approx. 100,000 German individuals, who progressed into retirement in 2007-2015, are used to measure intra-generational redistribution in the German statutory pension scheme (GRV). A reform of childcare benefit provision, which became effective in 2014, is utilized to show that the index responds in the predicted direction.

Terrorism in the Worlds of Welfare Capitalism

Article

Full-text available

Jan 2009

This is an outdated working paper version of the following paper: Krieger, T., and D. Meierrieks (2010), Terrorism in the Worlds of Welfare Capitalism. Journal of Conflict Resolution 54(6), pp. 902-939. Please check the published paper and cite accordingly.

A new proposal of pension regimes typology: empirical analysis of the OECD countries

Article

Full-text available

Feb 2017

In the paper, we present a new typology of pension regimes based on two main dimensions: the extent of involvement of the state and the market, and the role of voluntary schemes. We propose three theoretical pension regimes. The study proves that our theoretical typology is also consistent with empirical pension systems in OECD countries. In order to group them into the similar pension models, we employ multivariate statistical analysis. As a result of our empirical research, the regimes distinguished in the theoretical framework have found their counterparts in the clusters of real pension systems operating in 30 countries.

Has Intragenerational Redistribution Become Less Important in Pension Systems' Public Pillar? An International Comparison Based on LIS Microdata

Article

Full-text available

Apr 2011

We empirically investigate whether the significance of intragenerational redistribution in the public pillar of pension systems in 20 OECD countries has changed systematically since the 1980s and whether international convergence of the degree of intragenerational redistribution can be observed. Intragenerational redistribution is measured by the Bismarckian factor which provides information about the relative importance of the earnings-benefit link in the pension formula (as compared to a flat-benefit Beveridgean pension system). Based on micro data from the Luxembourg Income Study, we find both, a trend towards (more Bismarckian) pension systems which obey the principle of participation equivalence and an international convergence of pension systems. The reduced variation of pension systems (sigma convergence) is driven by countries with a high degree of intragenerational redistribution catching up with more traditional Bismarckian countries (beta convergence). Both, fundamental pension reforms as Sweden's and Italy's move to "notional defined contribution" systems, and parametric reforms ranging from the removal of group-specific benefits to alternative calculations of contribution history, such as changing from "best years" to the entire worklife, underlie this development.

Why are More Redistributive Social Security Systems Smaller? A Median Voter Approach

Article

Full-text available

Dec 2005

We suggest a political economy explanation for the stylized fact that intragenerationally more redistributive social security systems are smaller. We relate the stylized fact to an "efficiencyredistribution" trade-off to be resolved by political process. The inefficiency of social security financing is due to endogenous labor supply. Using data on eight European countries, we find that the stylized fact and a considerable degree of cross-country variation in contribution rates can be explained by the median voter model.

Relative Deprivation, Personal Income Satisfaction, and Average Well–Being under Different Income Distributions *

Working Paper

Full-text available

Nov 2002

Searching by questionnaire for the meaning of income inequality

Article

Full-text available

Jan 1993

A Theory of Justice

Book

Jun 2009

JOHN RAWLS

The Gains from Pension Reform

Article

Mar 2003
J ECON LIT

We classify social security pension systems in three dimensions: actuarial versus non-actuarial, funded versus unfunded, and defined-benefit versus defined-contribution systems. Recent pension reforms are discussed in terms of these dimensions. Shifting to a more actuarial system reduces labor-market distortions, although limiting the scope for redistribution. Shifting to a funded system may increase saving, redistribute income to future generations and distort contemporary labor supply. A partial shift to a funded system helps individuals diversify their pension assets. A shift from a defined-benefit to a defined-contribution system means that income risk will be shifted from workers to pensioners.

The American Soldier: Adjustment During Army Life and Combat and Its Aftermath.

Article

Jan 1949
Mil Aff

Acceptance of Distributional Axioms: Experimental Findings

Article

Jan 1994

This paper is an empirical study of the acceptability of distributional axioms. Data from five German universities were used to analyze attitudes with respect to the anonymity postulate, relative and absolute income inequality aversion (tests of the acceptability of scale and translation invariance), specific transfer affinities, and replication aversion (the population principle). This paper represents an innovation in the analysis of perception of distributional equality in that it offers more comprehensive and conclusive results as to the statistical consistency of test responses, introduces a test of the anonymity postulate and, finally, uses combinations of some of these responses to express more complex perceptions of income distributional inequalities in the form of inequality attitudes. Although the acceptance rate of the anonymity principle was not as high as we expected, anonymity was confirmed the strongest when compared with all other axioms. By contrast, other results show a lower acceptance rate, such as in the case of constant relative inequality aversion and decreasing absolute inequality aversion (41 % and 53 %, respectively). Combinations of these attitudes showed the prevelance of centrist inequality attitudes among our test populations (18 %), especially when an income-shock effect was introduced (27 %). Further analysis revealed the rather strong rejection of the transfer principle, (72 %), although not among supporters of traditional views of scale invariance. The most surprising finding in our study, however, is the rather strong rejection of the population principle (61 %).

Happiness and Economic Performance

Article

Nov 1997

Andrew J. Oswald

Optimal Actuarial Fairness in Pension Systems

Article

Aug 1997
ECON LETT

Abstract Arationale for a compulsory ,pension system is that the government wants to correct supposedly myopic behavior by the individuals. Given the existence of such a system, we calculate the optimal relation between marginal contributions and benefits, i.e., the optimal degree of marginal actuarial fairness, as seen from the point of view of the individuals or of the government. The following is shown,to hold under general assumptions ,of individual ,utility: The optimal ,degree of marginal actuarial fairness increases in the rate of return in the social security system ,and ,decreases in the ,government’s rate of time preference. If the government’s rate of time preference is lower than the individual’s, the government gains more than the individuals by making the system more ,actuarially fair. It is also shown ,that labor supply always increases when ,the link between ,marginal contributions and benefits is strengthened. ,,,,,,,,,,,, Assar Lindbeck is also associated fellow of IUI, Stockholm. Address of authors: Institute for International Economic Studies, Stockholm University, S- 106 91 Stockholm, Sweden. Telephone:+46-8-162000, Fax: +46-8-161443, Email: John.Hassler@iies.su.se and Assar@iies.su.se, homepage: http:/WWW.IIES.SU.SE/ 2

Contextual effects: A range-frequency analysis

Article

Dec 1974

ALLEN PARDUCCI

Back to Bismarck: Shifting Preferences for Intragenerational Redistribution in OECD Pension Systems

Figures

Recommended publications

Why Funding is Not a Solution to the Social Security Crisis

Fairness of Public Pensions and Old-Age Poverty

Integration of Pension, Assistance and Taxation: How to Balance Insurance Role with Redistribution R...

Are Contributions to Public Pension Programmes a Tax on Employment?