Wealth and Asset Holdings of Immigrants in Germany

Abstract

International migration of people is a momentous and complex phenomenon. Research on its causes and consequences, requires sufficient data. While some datasets are available, the nature of migration complicates their scientific use. Virtually no existing dataset captures international migration trajectories. To alleviate these difficulties, we suggest: (i) the international coordination of data collection methodologies and standardization of immigrant identifiers; (ii) a longitudinal approach to data collection; (iii) the inclusion of adequate information about relevant characteristics of migrants, including retrospective information, in surveys; (iv) minimal anonymization; (v) immigrant boosters in existing surveys; (vi) the use of modern technologies and facilitation of data service centers; and (vii) making data access a priority of data collection.

Full text

Page 1

Wealth and Asset Holdings of Immigrants in
Germany
Mathias Sinning
RWI Essen and IZA Bonn
February 2008
Abstract.
choices of immigrants in Germany. The empirical findings reveal significant dif-
ferences in overall wealth and various wealth components between German natives
and immigrants. Differences in real estate constitute the major part of different lev-
els of net worth, indicating that disparities in home-ownership rates are responsible
for the main part of the overall wealth gap. Moreover, migrants’ degree of portfolio
diversification is significantly lower than that of comparable natives. The results
of a decomposition analysis suggest that differences in wealth and asset holdings
may be explained by disparity in educational attainment to a sizable extent, while
the effects of income differentials and differences in demographic characteristics are
insignificant.
This paper examines the relative wealth position and the portfolio
JEL-Classification: F22, D31
Keywords: International migration, wealth accumulation, decomposition analysis,
multiple imputation
The author would like to thank Juan Bar´ on, Thomas Bauer, Deborah Cobb-Clark,
Markus Grabka and John Haisken-DeNew for very helpful comments. All corre-
spondence to Mathias Sinning, Rheinisch-Westf¨ alisches Institut f¨ ur Wirtschafts-
forschung (RWI Essen), Hohenzollernstr. 1-3, 45128 Essen, Germany, Phone: +49-
201-8149214, Fax: +49-201-8149284, Email: sinning@rwi-essen.de.

Page 2

1Introduction
As a result of the increasing relevance of international migration, the economic
and societal integration of immigrant minorities into the society of their host coun-
try has become a matter of intense debate among economists and policy makers
in many immigration countries worldwide. Following the seminal contribution by
Chiswick (1978), the literature on the economic performance of immigrants has
largely concentrated on the extent to which labor market outcomes (e.g., earnings
and employment status) of immigrants vary over the settlement process (Borjas,
1994; Zimmermann, 2005). Recent studies have started to examine how the relative
wealth position of immigrants enhances as their duration of residence in the host
country increases (Shamsuddin and DeVoretz, 1998; Zhang, 2002; Cobb-Clark and
Hildebrand, 2006a,b; Bauer et al., 2007).
An investigation of the nativity wealth gap allows inferences about the overall
economic well-being of immigrants. Studies that focus exclusively on income will
underestimate differences in economic well-being between natives and immigrants if
wealth disparities are even more pronounced. Moreover, policies that seek to reduce
income differences do not necessarily alleviate wealth inequalities, because wealth
may be distributed quite differently from income (Blau and Graham, 1990; Gibson
et al., 2007). At the same time, wealth represents an important measure of the eco-
nomic integration of immigrant minorities. Wealthier families have access to better
schools and enhanced health facilities and live in neighborhoods characterized by
lower levels of crime (Gittleman and Wolff, 2004). Wealth further provides liquidity
in times of economic hardship, access to the credit market, and the resources to
1

Page 3

maintain living standards in retirement (Cobb-Clark and Hildebrand, 2006b; Bauer
et al., 2007).
Previous studies provide a quite consistent picture regarding the existence of an
overall nativity wealth gap in different countries (Carroll et al., 1994; Cobb-Clark
and Hildebrand, 2006b; Bauer et al., 2007). Unfortunately, very little is known
about differences in the portfolio decisions of native-born and foreign-born individ-
uals, although it may be expected that both wealth levels and portfolio allocations
depend on nativity (Amuedo-Dorantes and Pozo, 2002; Cobb-Clark and Hildebrand,
2006a,b). This paper aims at filling this gap by investigating differences in the mag-
nitude and composition of wealth between German natives and immigrants and
examines the reasons for these differences. Since portfolio allocations may be re-
sponsible for a sizeable part of the nativity wealth gap, particular attention will
be paid to differences in composition and diversification of asset portfolios between
native-born and foreign-born individuals in Germany. In the empirical analysis,
which is based on cross-sectional data drawn from the German Socio-Economic
Panel (SOEP), two research questions will be addressed: Are there differences in in-
dividual wealth and asset holdings between natives and immigrants? Which part of
these differences can be attributed to disparities in socioeconomic and demographic
characteristics between the two groups?
Germany provides an interesting case study for the analysis of wealth and asset
holdings of immigrant minorities. During the 1960s, “temporary” guest workers
from Southern Europe were encouraged to migrate to Germany to fill an low-skilled
labor shortage. Many of them, however, decided to stay in Germany permanently
2

Page 4

(Schmidt and Zimmermann, 1992; Bauer et al., 2005). These immigrants were typi-
cally very different in education, cultural background and motivation to their higher-
skilled European counterparts that migrated to the United States after the Second
World War. Further restrictions limiting dual-nationality and complicating applica-
tion for German citizenship may have restrained potential assimilation, in contrast to
the integrative policies of typically immigration countries such as Australia, Canada
and the United States (Antecol et al., 2003). The wealth accumulation behavior of
this group of immigrants may become an important factor for the German pension
system, because because a sizeable part of the immigrant population in Germany
will reach retirement age in the coming decades.
The results of the empirical analysis reveal considerable differences in wealth
and asset holdings between natives and immigrants, indicating substantial disparity
in the economic well-being of the two groups. Moreover, differences in real estate
constitute the major part of different levels of net worth, suggesting that disparities
in home-ownership rates are responsible for the main part of the overall wealth gap.
Furthermore, migrants’ degree of portfolio diversification is significantly lower than
that of comparable natives. The results of a decomposition analysis suggest that
differences in wealth and asset holdings may be explained by disparity in educa-
tional attainment to a sizable extent, while the effects of income differentials and
differences in demographic characteristics are insignificant. The estimates of the
single components of wealth reveal that educational attainment is highly relevant
for the investment in financial and other assets as well as private insurances but
relatively less important for the accumulation of real estate. Finally, in most cases,
3

Page 5

more than half of the gap in wealth and asset holdings remains unexplained by dif-
ferences in income, education, and demographic characteristics between natives and
immigrants.
The paper proceeds as follows. Section 2 provides a short survey of the existing
literature on wealth and asset holdings. Section 3 describes the data used for the
empirical analysis and provides some descriptive statistics. The empirical strategy
and the estimation results are presented in Section 4. Section 5 concludes.
2Wealth and asset holdings of immigrants
From a theoretical perspective, there are several ways in which wealth levels and
portfolio choices may differ between natives and immigrants. Due to self-selection
and selective immigration policies of the receiving countries, immigrants are typically
non-representative of both the sending and receiving country populations. Conse-
quently, different observable and unobservable characteristics may be responsible
for differences in the magnitude and composition of wealth. For instance, wealth
disparities may be a result of differences in the economic performance of natives and
immigrants that were caused by different skill levels (Chiswick, 1978; Borjas, 1987).
Wealth levels and portfolio choices may differ between similar natives and immi-
grants for a number of other reasons. First, the nativity wealth gap may be the result
of different portfolio compositions. In particular, the higher ability of immigrants to
diversify portfolios across countries may allow them to hold different asset portfolios
that reduce income risk and lower the need for precautionary savings. Supporting
this hypothesis, Amuedo-Dorantes and Pozo (2002) argue that the apparent lower
4

Page 6

precautionary savings observed for immigrants in the US may be caused by the
fact that they engage in saving by remitting parts of their income to their home
countries. Amuedo-Dorantes and Pozo (2006) find that a higher income risk leads
to increased remittances of immigrants. Sinning (2007) demonstrates that return
intentions have a significant influence on migrants’ savings in their home country.
Second, different preferences or risk aversion may explain portfolio choices with
different rates of return and consequently variation in overall wealth levels. In par-
ticular, both immigrants’ preferences and risk aversion may be affected by social
norms in the sending country that are likely to influence not only intergenerational
transfers and inheritances, but also asset allocation, rates of return and in turn
wealth accumulation (Bauer et al., 2007). Empirical evidence suggests that inter-
generational transmission processes exist for both portfolio choice decisions (Chiteji
and Stafford, 1999) and attitudes towards risk and trust (Dohmen et al., 2006).
Moreover, Bonin et al. (2006) find that immigrants to Germany are significantly
more risk averse than native-born Germans, indicating that attitudes towards risk
may depend on nativity status.
Third, immigrants’ portfolio choice decisions and the resulting relative wealth
position may depend on expectations regarding retirement and return migration.
Cobb-Clark and Stillman (2006), for example, demonstrate that immigrants to Aus-
tralia are more uncertain about their retirement age than natives. This uncertainty
may be explained by migrants’ location choices after retirement to a sizeable extent
(De Coulon and Wolff, 2006). Theoretical models suggest that interactions between
relative economic conditions in home and host countries and expectations regarding
5

Page 7

return migration may affect the wealth accumulation behavior of immigrants. Galor
and Stark (1990), for example, demonstrate that the positive remigration probabil-
ity of immigrants increases their labor supply in the host country and consequently
their saving propensity. Djajic and Milbourne (1988) and Djajic (1989) show that
temporary migrants accumulate more wealth than natives and permanent migrants
if commodity prices in the host country are higher than in the home country. Finally,
Dustmann (1997) demonstrates that immigrants accumulate more wealth than na-
tives if they face greater income risk and argues that the amount of migrants’ savings
is a function of the correlation in labor-market shocks in home and host countries.
Previous studies provide a quite consistent picture regarding the existence of an
overall nativity wealth gap in different countries. Carroll et al. (1994), for example,
find differences in the saving patterns of immigrants to Canada across countries of
origin. They demonstrate that these patterns do not resemble the national saving
patterns in the sending countries because of immigrant selectivity variations across
sending regions, indicating that savings disparities within the immigrant population
do not reflect cultural differences. Cobb-Clark and Hildebrand (2006b) discover that
entry-cohorts do not affect overall wealth levels and demonstrate that the year of
arrival is significantly related to the portfolio choices of the foreign-born population
in the United States. Bauer et al. (2007) investigate the source of the relative
wealth position of immigrants in Australia, Germany and the United States at the
household level. Their findings reveal substantial wealth disparities between native
and immigrant households in Germany. Moreover, they provide empirical evidence
for the relevance of income, educational attainment and demographic characteristics
6

Page 8

in explaining wealth differentials between native and immigrant households.
3Data and descriptive analysis
3.1Data
In the empirical analysis, data from the German Socio-Economic Panel (SOEP) for
the year 2002 is utilized.1The SOEP is a representative longitudinal study includ-
ing German and immigrant households that started in 1984. In 2002, about 24,000
persons in nearly 13,000 households were sampled. The SOEP includes information
about socioeconomic and demographic characteristics, household composition, oc-
cupational biographies, etc. The empirical analysis is restricted to the year 2002,
because information about wealth is only available for this wave. As less than 2%
of the foreign-born population lives in East Germany, the analysis focuses on house-
holds residing in West Germany. Immigrants are defined as foreign-born individuals
who immigrated to Germany after 1948 (including foreign-born persons with Ger-
man citizenship).
The empirical analysis is performed at the individual level because wealth ques-
tions were included in the individual questionnaire of the SOEP, permitting an
explicit consideration of the distribution of wealth between spouses within house-
1The data used in this paper were extracted from the GSOEP Database provided by the DIW
Berlin (http://www.diw.de/GSOEP) using the Add-On package PanelWhiz v1.0 (Oct 2006) for
Stata(R). PanelWhiz was written by Dr. John P. Haisken-DeNew (john@panelwhiz.eu). The
PanelWhiz generated DO file to retrieve the GSOEP data used here and any Panelwhiz Plugins
are available upon request. Any data or computational errors in this paper are my own. Haisken-
DeNew and Hahn (2006) describe PanelWhiz in detail.
7

Page 9

holds. The estimation sample is restricted to include only native and foreign-born
couple-headed household heads and spouses who are between 25 years and 75 years
old. Since a substantial share (25.6%) of the households in the resulting sub-sample
of immigrants lives in mixed households (in which one partner is native-born and
the other is foreign-born), a separate consideration of spouses within households at
the individual level is particularly interesting. After excluding all observations with
missing values on one or more of the variables used in the analysis, the data set
contains 3,308 native-born and 587 foreign-born individuals.
3.2 Multiple imputation of wealth components and repeated-
imputation inference
In 2002, the individual SOEP questionnaire surveys seven components of wealth,
including owner-occupied housing (including mortgage debt), other property (in-
cluding mortgage debt), financial assets, business assets, tangible assets, private
pensions (including life insurance) and consumer credits (Frick et al., 2007). Based
on the individual share of the net market value of these components, four categories
are derived for the empirical analysis: (i) overall net worth, (ii) owner-occupied and
other property, (iii) financial and other assets, (iv) private insurances. Appendix-
Table A.1 includes a detailed description of the definition of these outcome measures.
Survey data – especially questions on wealth – typically suffer from measurement
error that have to be addressed by editing and imputation of item-non-response. A
revised version of the 2002 wealth module of the SOEP that accounts for measure-
ment errors was made available in 2007. Frick et al. (2007) provide an extensive
8

Page 10

description of editing and imputation procedures that were applied to obtain the
revised wealth information. In particular, missing values were imputed by regression-
based multiple imputation in the revised data. The advantage of this approach is
that it provides information that can be used to estimate the uncertainty that is
prevalent due to missing values, providing a basis for more valid inference and tests
of significance (Montalto and Sung, 1996).
The main idea of multiple imputation is to replace missing values by estimates
derived from a regression of the outcome measure on a set of explanatory vari-
ables. To simulate the sampling distribution of the missing values appropriately,
each missing value is replaced by five generated values that are imputed by the pro-
cess of randomly drawing a residual five times to obtain five different imputations,
referred to as “implicates”. Due to the generation of more imputed values, this pro-
cedure improves the approximation to the true sampling distribution. In practice,
the average of these values is calculated to produce the best estimate of what the
results would have been if the missing data had been observed (Rubin, 1987).
Generally, the best point estimates and estimates of variance for parameters of in-
terest based on the available information is achieved by simply combining the results
across the five implicates. This method, which is referred to as “repeated-imputation
inference” (Rubin, 1987), is applicable to both linear and nonlinear models. Given
the five point estimates of a parameter vector of interest, Q1,Q2,Q3,Q4,Q5, and
the corresponding variance estimates, U1,U2,U3,U4,U5, the best point estimate of
the parameter is simply the average of the five separate point estimates:
Qm
=
?m
i=1Qi
m
,i = 1,...,m,(1)
9

Page 11

where m is the number of implicates. The total variance Tmof the point estimate
consists of two components. The first component (the “within” imputation variance)
may be estimated by the average of the five separate variance estimates,
Um =
?m
i=1Ui
m
,i = 1,...,m.(2)
The estimate of the second component (the “between” imputation variance) is
Bm =
?m
i=1(Qi− Qm)t(Qi− Qm)
m − 1
.(3)
The total variance of the point estimate is the sum of the “within” imputation
variance and the “between” imputation variance, whereas the latter is weighted by
an adjustment factor for the use of a finite number of implicates:
Tm = Um+ (1 +1
m)Bm.(4)
Finally, the standard deviation of the point estimate is defined as the square root
of the total variance.
In the following empirical analysis, repeated-imputation inference is applied to
obtain the point estimates of the parameters of interest and the corresponding vari-
ance estimates by combining the estimation results across the five implicates. The
underlying separate point estimates of the different implicates are available from the
author upon request.
3.3 Descriptive statistics
Table 1 includes information about the level of wealth held by natives and immi-
grants. The numbers indicate that the overall net worth of natives is consider-
ably higher than that of immigrants. Specifically, immigrants to Germany hold
10

Page 12

only about 55% (e 65,071) of the overall net worth of natives. Immigrants are
also much less likely to report positive net worth than natives.However, this
lower propensity to hold positive net worth explains the overall nativity wealth
gap only partially. Conditional on having positive net worth, immigrants still hold
less than 60% (e 71,133) of the net worth of natives.
The numbers of the different wealth components indicate that the major part of
the nativity wealth gap is attributable to differences in real estate. While immigrants
hold about 51% (e 41,766) of the net market value of owner-occupied and other
property, the corresponding ratio of financial and other assets and private insur-
ances amounts to 57% (e 12,212) and 71% (e 13,894), respectively. Given positive
amounts of the respective wealth component, immigrants hold about 69% (e 83,305)
of the net market value of owner-occupied and other property. Since immigrants are
on average much less likely to hold financial and other assets or private insurances
than natives, the corresponding shares of the conditional market values of these
components are above 80%. These numbers are supported by the number of assets
held by natives and immigrants. While natives hold on average about 2.3 different
assets, immigrants hold only about 1.7 assets.
Table 1 further describes the relevant socioeconomic and demographic charac-
teristics of natives and immigrants.2Immigrants have a lower income, are younger
and less educated and have more children than natives. There are also differences
in the distribution of the foreign-born population across entry cohorts and regions
of origin. The majority of the immigrant population arrived either before 1974 or
2Appendix-Table A.2 includes a definition of these variables.
11

Page 13

after 1989. Immigrants to Germany primarily stem from OECD member countries,
Central and Eastern Europe or Ex-Yugoslavia.
Figures 1-4 display the unconditional gaps in the overall wealth level and the
three wealth components between natives and immigrants and the correspond-
ing 95% confidence interval over the entire distribution. These figures reveal signifi-
cant differences at most points of the overall wealth distribution and the distribution
of the respective wealth components between natives and immigrants. While the
overall wealth gap is significantly negative along the entire distribution, differences
in the wealth components are insignificant at most points below the median but
steadily increasing along the distribution above the median. At the 25thpercentile,
the overall wealth gap is e 18,313. This gap amounts to e 57,661 at the median
and increases to e 76,144 at the 75thpercentile (see Figure 1). The differences in
real estate and financial and other assets between natives and immigrants are zero
at the 25thpercentile but positive at the median. While the median gap in real es-
tate amounts to e 59,461, it declines slightly to e 59,318 at the 75thpercentile (see
Figure 2). Differences in financial and other assets between natives and immigrants
are only e 5,000 at the median and add up to e 10,185 at the 75thpercentile (see
Figure 3). Finally, the gap in private insurances between natives and immigrants
is e 2,000 at the 25thpercentile and amounts to e 4,624 at the median. This gap
increases to e 8,011 at the 75thpercentile (see Figure 4).
12

Page 14

3.4Determinants of net worth components and diversifica-
tion
To assess the relative importance of the factors affecting the overall net worth and its
components, the determinants of the different outcome measures are being investi-
gated. As wealth distributions are usually skewed to the right, the existing literature
typically relies on log-linear regression models (Shamsuddin and DeVoretz, 1998).
However, a log transformation is inappropriate for individuals with zero or negative
net worth. Consequently, a quantile regression model is estimated to analyze the
determinants of net worth and its components at the median of the distribution.
Specifically, the following cross-sectional quantile regression model is estimated for
native and foreign-born individuals (i),
mik
= βq
0k+?Xiβq
1k
(5)
+ Ii
?βq
2k+ βq
3kZi+ βq
4kMiHi+ βq
5kMi(1 − Hi) + Diβq
6k+ Riβq
7k
?+ εq
ik
= Xiβq
k+ εq
ik,i = 1,...,N,k = 1,...,K,
where mikis the net market value of outcome measure k and q reflects a specific
percentile of the distribution. Four outcome measures are considered in the empir-
ical analysis: overall net worth, owner-occupied and other property, financial and
other assets and private insurances.?Xicontains information about income (i.e. cur-
rent net income), education (in years) and demographic characteristics (number of
children younger than 18 in the household, age and age squared). To distinguish
between immigrants residing in mixed households and those who do not, several
indicator variables are considered. Specifically, Ii reflects the immigrant status,
13

Page 15

including immigrants who reside in a mixed household, while Zi is an indicator
variable for the sample of immigrants with foreign-born partners. Mi is an indi-
cator variable for mixed households and Hidenotes whether the observed person
is considered as head of the household. The model is identified by imposing the
restriction βq
3k+ βq
4k+ βq
5k= 0. Moreover, Di is a vector of indicator variables
capturing immigration cohorts, and Riis a vector of indicator variables reflecting
immigrants’ regions of origin. Finally, the vector βqincludes the model parameters
to be estimated and εq
iis an error term with the usual properties.
The model contains the full set of immigration cohort and region of origin indica-
tors to facilitate interpretation of the estimation results. Identification of the overall
constant is achieved by restricting the estimated coefficients on these variables to
sum to zero, i.e. the restrictions?
m and n are the numbers of immigration cohorts and regions of origin respectively.
mβq
3km= 0 and?
nβq
4kn= 0 are imposed, where
Consequently, βq
2kmay be interpreted as the overall difference in the outcome mea-
sure between natives and immigrants given a set of characteristics, while βq
6kand βq
7k
comprise the deviations of specific immigration cohorts and regions of origin from
this outcome measure.
In addition to the analysis of the factors influencing the components of net worth,
the determinants of the degree of asset portfolio diversification are being investigated
by using the number of assets held by an individual as a dependent variable. To
account for the fact that the dependent variable is given by a count data variable, a
Poisson regression model is estimated. The Poisson regression model assumes that
the dependent variable conditional on the covariates is Poisson distributed with
14