The Immigrant / Native Wealth Gap in Germany, Italy and Luxembourg

Transcription

1 The Immigrant / Native Wealth Gap in Germany, Italy and Luxembourg Thomas Y. MATHÄ Alessandro PORPIGLIA Eva SIERMINSKA Economics & Research Dept. Economics & Research Dept. CEPS/INSTEAD Central Bank of Luxembourg Central Bank of Luxembourg, DIW, Berlin University of Tor Vergata, Rome IZA, Bonn thomas.mathae@bcl.lu alessandro.porpiglia@bcl.lu eva.sierminska@ceps.lu Version: April 2013 Abstract: This paper analyses the existence of an immigrant/native wealth gap by using household survey data for Germany, Italy and Luxembourg. The results show that, in all three countries, a sizeable wealth gap exists between natives and immigrants. Towards the upper tail of the wealth distribution the gap narrows to a small extent. This gap persists even after controlling for demographic characteristics, country of origin, cohort and age at migration, although cross-country differences exist in the immigration penalty. The Oaxaca-Blinder decomposition shows, despite sizeable differences in immigrant and native characteristics at the mean, that the largest share is accounted for by differences in coefficients as opposed to endowments. Keywords: household, survey data, wealth gap, immigrants, distribution JEL Codes: D31, F22 This is a modified version of BCL WP 57 / ECB WP We thank Vincent Hildebrand for his valuable suggestions concerning the empirical implementation, our discussants Jirka Slacalek and Anna Tavarez, as well as participants at the BCL and ECB joint conference on Household Finance and Consumption, held on October 25-26, 2010 in Luxembourg, the participants at the annual ESRA conference in Lausanne in July 2011, annual conference of the Swedish Network for European Studies in Economics and Business (SNEE) in Mölle in May 2012, and the participants at the fourth NORFACE conference held in London in April 2013 for helpful comments received. The views expressed in this paper are personal views of the authors and do not necessarily reflect those of the Banque centrale du Luxembourg or the Eurosystem. For Sierminska, this work is part of the WealthPort project (Household Wealth Portfolios in a Comparative Perspective) supported by the Luxembourg Fonds National de la Recherche (contract C09/LM/04) and by core funding for CEPS/INSTEAD by the Ministry of Culture, Higher Education and Research of Luxembourg.

2 1 Introduction Wealth is a natural candidate to measure economic household wellbeing. It is the outcome of an accumulation process and enhances the quality of life by providing the option to smoothen consumption despite possible income volatilities. It is therefore frequently used as a measure to analyse how various economic groups in the society are performing economically, notably how well the immigrant population is doing relative to the native population in a country (e.g. Gittleman and Wolff, 2004, Bauer et al., 2011, Sinning, 2007). This is also, as unlike differences in other measures of immigrant economic integration, such as the income gap, wealth disparities are more likely to be transferred to successive generations, thereby threatening the immigrant assimilation process (Gibson et al., 2007). In view of industrialised countries increasingly ageing populations, the means to which accumulated wealth provides for consumption smoothing and the upkeep of acquired living standards in retirement is becoming ever important; an issue that is linked to the upkeep of current social welfare systems. More immigration is a commonly advocated solution discussed in this context. Indeed, the continued deepening and enlargement of the European Union has increased labour mobility, and together with the effects of globalisation more and more people nowadays live and work outside their country of birth. However, at current labour force participation and fertility rates, it is reported that a yearly million immigrants into the EU25 are needed to keep the labour force constant (Holzmann, 2005). It is thus clear that, in the long run, the population aging problem cannot be solved solely by the increase of the immigration flows. Nevertheless, at least in the short term, immigration may help to successfully smoothen the effects of population ageing. The success of which strongly depends on the extent to which immigrants contribute to the social welfare system, which is linked to their economic success and wealth accumulation. Several scholars have risen that a gap of immigrant to native wealth may threaten the sustainability of pension systems (e.g. Cobb-Clark and Hildebrand 2006b; Sinning 2007; Bauer, 2011 and Doiron and Guttmann, 2009). For studies specifically analysing the existence of this gap and what it entails for the old age subpopulation see Lewin-Epstein and Semyonov (2011) and Amuedo-Dorantes and Pozo (2012). This paper relates, in particular, to Shamsuddin and DeVoretz (1998) and Cobb- Clark and Hildebrand (2006b), that use methodologies analogous to ours to identify the immigrant to native wealth gap in using respectively Canadian and US data, to Sinning (2007) that analysed the same topic using an older vintage of the German SOEP, and to Bauer et al. (2011) that is, at the best of our 2

3 knowledge, the only study that analyses the immigrant to native wealth gap in a cross country comparative framework. Our study contributes to this field of research; we provide a cross-country investigation of immigrant to native household net wealth gap in a pure European comparative context and focus on three countries that experienced different immigration dynamics over their history: Germany, Italy and Luxembourg. In this respect, this study differs from Bauer et al. (2011) who focus on three countries with relatively similar recent immigration histories (at least with respect to the size of the immigration flows). By studying these specifically selected countries, we hope to uncover whether the existence of a wealth gap remains universally valid and how its determinants (at mean level) vary. Germany is considered a traditional immigration country accepting just temporary, mainly unskilled workers (Bauer et al. 2011) whereas Italy is a traditional emigration country that, in recent years, has evolved into an immigration country (e.g. Del Boca and Venturini, 2005). In this context, Luxembourg is a unique case as it attracts both skilled and unskilled workers due to its high wages and high living standards (see Cordeiro, 2001 for an overview on Luxembourg immigration history). At the moment, it is the country with the highest foreign population share in the EU having experienced a high level of immigration since the beginning of the last century. Nonnationals presently account for about 44% of the Luxembourg resident population. Furthermore, this is the first paper conducting an analysis of the immigrant/native wealth gap in Italy and Luxembourg in a comparative context. 1 We show that there is a sizeable wealth gap between natives and immigrants in all three countries considered. At the 75th percentile, the immigrant/native wealth ratio is 30%, 14% and 60%, respectively. While the relative gap narrows at increasing percentiles, it is robust across the entire net wealth distribution. Furthermore, it persists even after controlling for relevant household characteristics and is not affected by different economic structures or migration situations, although the estimated effects vary across countries. A decomposition à la Oaxaca-Blinder shows that the largest share of immigrant to native wealth gap is related to the differences in the coefficients of household 1 In Luxembourg, the integration of immigrants has previously been analysed from the income perspective, but not for the wealth perspective. Ametepé and Hartmann-Hirsch (2008) find no relevant differences in income between natives and immigrants in Luxembourg, especially among highly qualified individuals. Furthermore, integration of Portuguese immigrant household in Luxembourg have been analysed by Hildebrand et al (2012). Büchel and Frick (2005) provide evidence on immigrants economic performance across Europe thereby focussing how differences immigration policy can affect the outcome across countries. 3

4 and individual characteristics as opposed to the differences in the characteristics themselves. This paper proceeds as follows: section 4.2 provides a review of relevant literature on wealth differences between native and immigrant households. Section 4.3 presents and describes the used databases. Section 4.5 discusses wealth levels and net wealth components. Section 4.5 provides the methodological framework and presents the results. Section 4.6 concludes this paper. 2 Relevant Literature The relevance of immigration has steadily been increasing since WWII and with it began the debate on the socioeconomic integration of immigrants (see Schmidt and Zimmermann, 1992, and Bauer et al., 2005 for a review of immigration history Germany, Del Boca and Venturini, 2005 for Italy and Cordeiro, 2001 for Luxembourg). The increase of immigrants duration of stay in the hosting countries lead to a shift away from analysing immigrant assimilation based on labour market outcomes and the potential impact on the hosting countries, as was the case in much of the early literature (see among others Chiswick, 1978 and Borjas, 1994). Instead, researchers interests turned to the analysis of wealth position of immigrants and natives. Total household net wealth can be seen as the outcome of an accumulation process resulting from: resources to be invested (i.e. inheritance and savings resulting from income and consumption patterns) and rate of returns (influenced by the portfolio choices)(cobb-clark and Hildebrand, 2006b). In this perspective, it is interesting to review the contributions analysing: a) the household total net wealth gap between immigrants and natives b) the differences in the single components of this accumulation process as saving patterns and portfolio composition (with a particular focus on the real estate component), c) aspects of particular policy relevance as the effect of the wealth gap for old age cohorts or wealth gap international comparisons. In recent years, household total net wealth disparities between natives and immigrants or ethnic and racial groups have been analysed for various countries, with most of the evidence stemming from the U.S. Blau and Graham (1990) report large wealth differences between Blacks and Whites, which aside of a non negligible influence of income differences and other sociodemographic characteristics., appear to be mainly related to the effect of ethnicity. Looking beyond the racial wealth gap debate, Campbell and Kaufman (2006) for example analyse the wealth accumulation of the five largest racial ethnic groups in the U.S. They report sizable wealth difference between 4

5 the different ethnic groups, with Blacks and Whites marking the two end points of the distribution. In a study of Mexican Americans, Cobb-Clark and Hildebrand (2006a) report that racial and ethnic wealth differences are much larger than corresponding income differences and that much of the wealth disadvantage of Mexican American households is attributable to them having more children and younger household heads (see Scholz and Levine, 2004 for a review of the U.S literature on the wealth inequality by race). Most available evidence points to the existence of a wealth gap between immigrants and natives (e.g. Hao, 2004; Cobb-Clark and Hildebrand, 2006b for the U.S.; Shamsuddin and DeVoretz, 1998 and Zhang, 2003 for Canada; Gibson et al., 2007 for New Zealand; Sinning, 2007 for Germany; Semyonov and Lewin- Epstein, 2010 for Israel; Doiron and Guttmann, 2009 for Australia and Bauer et al., 2011 for Australia, Germany and the US). In this context, different homeownership rates between immigrants and natives have been shown to contribute to observed wealth differences (e.g. Borjas, 2002 and Sinning, 2010). Cobb-Clark and Sinning (2011) emphasise the existence of an immigrant to native gap in the housing appreciation patterns. A further issue of interest is the possible convergence of immigrant and native wealth levels over time - it is debated whether or not immigrants catch up over time with natives. Hao (2004) for example studies the wealth of immigrants in the U.S. and native-born Americans and reports that, immigrants tend to converge in wealth accumulation to natives. Shamsuddin and DeVoretz (1998) analyse wealth differences between immigrants and natives in Canada. They report differences in wealth accumulation and estimate that immigrant households need on average about 15 years to catch up to the mean wealth level of Canadian-born household with similar socio-economic characteristics. Zhang (2003) reports that at the lower end of the distribution immigrants are less wealthy than their Canadian-born counterparts, whereas in the 40 th to 90 th percentile they outpace them. Others factors contributing to observed wealth differences between natives and immigrants may be related different saving patterns or risk taking behaviour. First, household wealth is the result of an accumulation process, and hence different saving patterns among immigrant and natives can strongly influence observed net wealth among native and immigrant households. Carrol et al. (1994) report that country of origin plays a relevant role in determining the saving behaviour of Canadian immigrants. Zhu and Piracha (2010) report that changes in the immigration legal framework in Germany modified immigrant household saving behaviour, as they adapted their expectations concerning the 5

6 length of stay in the hosting country. Second, the portfolio composition of assets reflects the risk attitude of the investor and determines the returns and savings that accrue, and thus household wealth in more general in the long run. In this context, differences in the portfolio allocation between native and immigrant household shed light onto differences risk-taking behaviour and exposure to economic fluctuations betweens those two groups. Again, empirical studies report that such differences exist between ethnic groups (e.g. Blau and Graham, 1990; Cobb Clark and Hildebrand 2006a). Cobb-Clark and Hildebrand (2006b) report differences between immigrants and natives to be influenced by cohort and country of origin. Osili and Pauson (2007) report this difference to be influenced by educational attainment, income, and as being persistent over time. Interestingly, ethnic concentration in specific metropolitan areas appears to be a factor reducing immigrant access to financial markets. Sinning (2007) stresses that immigrants to Germany tend to have, not just lower wealth levels, but even a lesser diversified portfolio than natives and attributes the most relevant part of the reported differences to the difference in educational attainments. Turning to the analysis of aspects of particular policy relevance, it is useful to remark that wealth is a crucial variable to be taken into account to evaluate living standards and vulnerability, especially of people who quit the job market. In this context, a growing literature is devoted to the analysis of immigrant to native wealth gap in old age. Lewin-Epstein and Semyonov (2011) found that in Israel the immigrant to native wealth gap in old age grows along household wealth distribution, and tends to be reduced for immigrants who arrived in Israel in earlier periods. Being outside the job market, retired immigrant households are possibly more vulnerable to the current financial crisis, especially if their wealth level were insufficient or their portfolio composition were be badly allocated; Auedo-Dorantes and Pozo (2012) analysed the impact of the recession on the wealth of older immigrants and native households in the U.S. and report that mixed households experienced the most relevant downturn in their total wealth, even when taking into account the fact that their portfolio composition, on average, included a higher share of financial assets (i.e., the most affected by the crisis). Furthermore, effects on the top and the bottom of the wealth distribution appear to be more severe than in the interquartile range. Bauer et al. (2011) was the first paper to put the immigrant to native wealth gap in a cross-country perspective. Analysing data from three traditionally immigration countries (Australia, Germany and Unites States) the study contributes to assess the determinants of immigrant to native wealth gap under 6

7 different immigration settings, finding that while in Germany and US the wealth gap arises as an outcome of different education and demographic patterns, in Australia, a country that experienced a selective and permanent immigration, the immigrant wealth disadvantages appear to be linked to the difficulty of immigrants to take advantage of their relatively higher education skills. The contribution of our paper is in the field of assessing the presence of an immigrant to native wealth gap conditional on socioeconomic variables, which is performed not just in a framework of traditional immigration countries, but even in a set of countries with a very different relevance of the immigration phenomenon. 3 Data description The analysis presented in this paper is based on three different household surveys, constructed to be country representative and including (ex-post) harmonised wealth aggregates. German data are from the 2007 release of German Socio-Economic Panel (SOEP), as included in the Luxembourg Wealth Study (LWS). 2 This is a representative longitudinal survey that includes almost 12,000 households resident in Germany. Italian data are from the 2008 wave of the Bank of Italy Survey of Household Income and Wealth (SHIW) (for more details this survey see for example Brandolini and Cannari, 1994), which includes almost 8,000 households resident in Italy. For comparability purposes Italian data employed in this paper are harmonised using a methodology consistent with the LWS definitions. The Luxembourg data are from a small wealth module included in the 2008 PSELL-3/EU-SILC (EU Survey on Income and Living Conditions), as included in the Luxembourg Wealth Study (LWS). The PSELL-3/EU-SILC is a representative household panel survey. All variables in value terms are expressed in current euro. Table 1: Availability of wealth components in the data Germany Italy Luxembourg Principal residence x x x Investments in Real Estate x x x Total Financial Assets x x x Principal Residence Mortgages x x Other Property Mortgages Home secured Debt x Non-Home secured Debt x x x Net Wealth x x x x 2 The LWS database contains ex-port harmonised wealth micro-datasets from a number of rich countries. We focus on three countries with most recent data. For more details on LWS, see Sierminska et al. (2006). 7

8 The household net wealth is calculated as the sum of the different available wealth components; depending on the questionnaire structure of the three analysed surveys, each component could have been asked as a household aggregate or disaggregated at individual level. Table 1 provides a description of the different household wealth components in each country. Surveys differ across countries and therefore the availability of specific wealth components also differs. To increase the comparability of net wealth, we will use a measure of net wealth that includes financial assets, the value of the principal residence and investment real estate net of mortgages on both type of properties and net of other home secured and non-home secured debt. This measure excludes business equities and pension assets, as these are not available in all three countries. Despite our attempts to harmonise the net wealth value, difficulties remain (for a discussion see Sierminska et al., 2006) and components commonly used for the calculation of the aggregate may vary, resulting in differences in the definition in each country. The components for the value of the principal residence, total financial assets and investment in real estate are available for all three countries. German data reports separately principal residence mortgages and mortgages on other real estate property, while for Italy, an aggregate for home secured debt is available. Luxembourg database reports solely principal residence mortgages and investment real estate is reported net of loans. Consequently, the household liability figures reported for Luxembourg are likely to be somewhat underestimated in the paper. Non-home secured debt is present in all the three considered countries, although the composition is not homogeneous; the relatively small share of non-home secured liabilities suggests that this heterogeneity is not particularly relevant. 3 An important omission in all of these surveys is pension and business equities. As their importance differs across countries, cross-national comparisons are bound to reflect these omissions. 4 All databases contain edited and imputed values. In the Italian case missing data of the principal variables are stochastically imputed. 5 3 Non-home secured debt is present as an aggregate in the German data, while is provided in a disaggregated form in the Italian and Luxembourg databases ; It includes vehicle loans, total installment debt and informal debt for Italy, and total installment debt, educational loans and other loans from financial institutions for Luxembourg. 4 See Frick & Headey (2009) for a comparison of wealth inequality that includes pension entitlements among the elderly in Australia and Germany. 5 For a comprehensive description of the database see Banca d Italia (2010) 8

9 In the German and Luxembourg case just income and wealth components have been imputed using a multiple imputation procedure. 6 In the following section we provide a number of basic statistics highlighting the differences between immigrant and native headed household in our sampled countries (Table 2). All reported values are weighted and country representative unless otherwise stated. All monetary values are either aggregated or reported at the household level. We classify a household as immigrant if the household head is born outside the country in question, regardless of his/her nationality. Thus, naturalised household heads are considered as immigrants to reflect the cultural background rather than the citizenship status. The household head is considered the household member who identifies himself as the financial knowledgeable person answering the household level questions on behalf of the whole household. 7 With a share of over 40% of total households headed by a non-native person, the share of immigrant household heads is substantially higher in Luxembourg than in Germany (10%) or Italy (7%). The share of men heading households is substantially higher for immigrant households than the country average in all three countries considered. The ageing of the population is a pressing issue in all Western European countries. In the three countries, the majority of households do not have any children younger than 18 years of age. Immigration is often discussed as the cure to an ageing population. The statistics explain why: immigrant household heads tend to be both younger and to have a higher number of children. The age distribution is much more left skewed for immigrant than for native household heads. In Italy, more than 88% of immigrant households heads, but only 36% of native households heads fall into the years of age category. In Luxembourg, the share is 58% and 43%, respectively. At 75%, the share of immigrant households in pre-retirement age (less than 65 years of age) is 7 percentage points higher than for natives in Germany. The share of elderly people (over 65 years of age) is much smaller among immigrant than among native household heads. The respective shares for native household heads are 32%, 37% and 31% in Germany, Italy and Luxembourg but only 25%, 5% and 6 Financial assets in the wealth module for Luxembourg are reported in categories. After multiple imputations we use information on interest income to calculate monetary values within each category. This is a first such attempt using Luxembourg data. 7 Thus even in mixed marriages (one immigrant, one native) the reference person is the most knowledgeable about finances, thus the household member having the largest impact on wealth accumulation. 9

10 14% for immigrant household heads. The latter are also not only more likely to have children but they also tend to have more children. The educational pattern of immigrant and native household heads is of particular interest. In Germany, both tend to be concentrated in the secondary education category. With a share of 56% for native and 49% for immigrant household heads, the mode is the completion of secondary education. Interestingly, we observe a slightly higher share of less educated immigrants than natives. 15% of native household heads as compared to 28% of immigrant household heads have not completed a higher education level than primary education. In contrast, the respective share of those having completed tertiary education is quite similar (29% and 24%). 10

11 Table 2: Descriptive statistics Immigration status Gender Age Education Net wealth Income Number of children Marital status Germany Italy Luxembourg Native Immigrant Total Native Immigrant Total Native Immigrant Total Non Weighted Number of obs. 10,425 1,117 11,542 7, ,977 1,645 2,134 3,779 Weighted Share Male Female Over No Edu/Primary Secondary/Post Secondary Tertiary Mean 115,711 61, , ,528 60, , , , ,508 Median 20, , ,000 2, , , , ,000 Mean 26,172 23,818 25,934 33,666 20,982 32,759 61,223 56,882 59,451 Median 21,564 19,602 21,399 27,170 17,009 26,411 54,534 45,958 50, > Never Married Married Sseparated/Divorced Widowed Source: Authors calculation. Data from LWS and Bank of Italy Notes: All statistics weighted and country representative unless otherwise stated. Wealth and income in EUR. Personal characteristics refer to the household head. 11

12 In Italy, for native and immigrant household heads, the education mode is having completed secondary education. Also, for the latter, the share of having completed tertiary education is slightly higher than for native household heads (11% vs. 10%). The education distribution for native household heads presents a more pronounced left tail with 30% of observations having completed just the primary level of education (compared to 9% of the immigrants). These numbers omit illegal immigration that would possibly inflate the share of households headed by a person holding a low education. 8 In Luxembourg, for native household heads the mode is to have completed secondary education as is the case for immigrant household heads. The distribution of this variable for the two groups is quite different however; the education distribution of immigrant headed households presents much wider tails than their native counterparts. Among immigrant household heads, 24% have completed either primary or no education, compared to less the 12% of native household heads; 30% of immigrant household heads have completed tertiary education while only 19% of native household heads achieved this educational level. Thus, a relatively high share of immigrant household heads are considered to be either low or highly educated, probably reflecting the fact that a relevant share of the immigrant population is either employed in low specialization sectors or in the financial sector and European intuitions. 4 Descriptive statistics of wealth levels and wealth components Next, we turn to income and wealth gaps and compare them across countries. It is worthwhile to emphasise that the wealth gap between immigrant and native households is wide (in the range of 50% or more) in all three countries and substantially wider than the income gap. This is shown Figure 1, which presents the mean income and the mean wealth of immigrants as a percentage of the respective values for natives and shows that immigrant households mean income tends to be relatively close to that of native households. This is particularly the case for, Germany and Luxembourg. However, the net wealth held by immigrant households is just 48% of the net wealth of native households in Luxembourg, 53% in Germany and 26% in Italy. 8 It needs to be noted that all data is at the household level. In Italy, there are few young household heads since many adult children still live with their parents. As a result the household structure needs to be taken into account when making conclusions regarding wealth distribution. See also Bover (2010) on this point. 12

13 Figure 1: The Income and Wealth Gap Mean income & wealth of immigrant relative to native households, in % 0 Germany Italy Luxembourg Income Wealth Source: Authors calculation. Data from LWS and Bank of Italy Notes: All statistics weighted and country representative. The difference in the income and wealth gap is not surprising. While the first is a flow variable the latter is a stock variable and reflects the whole process of accumulation during the lifetime of the individuals, incorporating a number of factors such as family background, inheritances, portfolio allocations etc. Comparing the central tendency measures of the distribution of net wealth for immigrant and native households only gives a partial view of the wealth gap issue, given the large skewness of the data. Table 3 shows the values and immigrant/native headed households net wealth ratios at various points of the wealth distribution. In all three countries, net wealth turns positive at an earlier stage of the distribution for native than for immigrant households. Across the wealth distribution native households are always wealthier than immigrant households. Furthermore, net wealth among immigrant headed households is more asymmetrically distributed. The relative wealth gap narrows at higher percentiles of the wealth distribution in all countries, at the 75 th percentile the net wealth of immigrant headed households is 60% of the net wealth of native headed households for Luxembourg, 30% for Germany and 14% for Italy, while at 95 th percentile of wealth distribution the share is 67% for Luxembourg, 62% for Germany and 47% for Italy, the wealth gap slightly increases in all three countries in the latter percentile of net wealth distribution. 13

14 Table 3: The distribution of net wealth and the wealth gap (ratio of immigrant to native wealth). Percentile Total -55,000-11,000-2, , , , ,000 1,004,636 Native -55,000-10,135-2, , , , ,381 1,026,558 Immigrant -60,000-15,000-5, , , , ,854 Wealth Gap Total -9, , , , , ,500 1,405,816 Native -7, , , , , ,000 1,460,906 Immigrant -27,000-5,000-1, ,677 41, , , ,000 Wealth Gap n.a Total -1, , , ,677 1,074,090 1,427,095 2,757,096 Native , , , ,834 1,208,354 1,600,000 3,500,000 Immigrant -3, , , , ,219 1,074,090 2,151,494 Wealth Gap 1560 n.a Luxembourg Italy Germany Source: Authors calculation. Data from LWS and Bank of Italy Notes: All statistics weighted and country representative. Wealth in EUR

15 Germany Italy Luxembourg Table 4: Asset participation rates of households Native Immigrant Total Main Residence Financial Assets Investment in Real Estate Home Secured Debt n.a. n.a. n.a. Mortgage* Non Home Secured Debt Main Residence Financial Assets Investment in Real Estate Home Secured Debt * Mortgage n.a. n.a. n.a. Non Home Secured Debt Main Residence Financial Assets Net investment in Real Estate Home Secured Debt n.a. n.a. n.a. Mortgage* Non Home Secured Debt Source: Authors calculation. Data from LWS and Bank of Italy Note: Weighted shares, country representative; * calculated for homeowners only The distribution of the main net wealth components elicits further differences between native and immigrant headed households. Table 4 describes the ownership rate of each component of household wealth. Consistent with the empirical findings in the U.S. (e.g. Borjas, 2002), homeownership rates are lower among immigrant households than among native households in Luxembourg, Germany and Italy. For home owners, having a mortgage is more common among immigrant households, reflecting the fact native households are more likely to receive property as inheritance or intergenerational transfer, for instance. Financial asset participation rate differences among the two groups is more pronounced for Germany compared to Luxembourg and Italy where the difference in favour of native household heads is still 17% and 13%. Investments in real estate are less common than holding financial assets. This is true for all households in all three countries. Slightly more than 27% of the overall population declared to have invested in real estate in Luxembourg, 13% in Germany and 22% in Italy. The difference in participation rate between native and immigrant households for this component is more pronounced in Italy and Germany than Luxembourg. Non-home secured debt is more likely to be owned by immigrant headed household for Luxembourg and Italy, while the 15

16 participation rates seem to be quite equally distributed among native and immigrant headed household both in Germany. 16

17 Main Residence Total Financial Assets Investment in Real Estate Home Secured Debt Mortgages Non Home Secured Debt Table 5: Conditional mean and median of net wealth components Germany Italy Luxembourg Native Immigrant Total Native Immigrant Total Native Immigrant Total Mean 210, , , , , , , , ,231 Median 180, , , , , , , , ,000 Mean 37,909 42,905 38,234 29,773 11,028 28,649 51,063 45,459 49,111 Median 15,000 8,000 14,661 9,531 3,000 8,353 17,497 17,382 17,462 Mean 253, , , , , , , , ,002 Median 100, , , ,000 50, , , , ,000 Mean n.a. n.a. n.a. 68,781 77,205 69,381 n.a. n.a. n.a. Median n.a. n.a. n.a. 55,000 70,000 56,000 n.a. n.a. n.a. Mean 84,557 96,767 85,552 n.a. n.a. n.a. 136, , ,627 Median 70,000 80,278 70,000 n.a. n.a. n.a. 95, , ,000 Mean 19,071 19,677 19,136 7,231 6,011 7,122 4,058 2,205 3,031 Median 8,686 9,600 8,949 4,800 4,600 4,800 1,330 1,155 1,187 Source: Authors calculation. Data from LWS and Bank of Italy Note: Weighted statistics, country representative. Figures in EUR. 17

18 To complete the picture of the wealth distribution of immigrant and native households we look at the conditional central tendency measures of each component. The mean and median in Table 5 are calculated for those households only that have declared to hold the respective asset. Across the positive wealth components, with the relevant exception of the main residence value for Germany, both the conditional mean and median are higher for native households. For mortgages it is the opposite. The average value of the principal residence, which is the largest component of the wealth portfolio for immigrant and native homeowners, is quite similar across countries. However, as shown in Table 4 the homeownership rates vary, and hence the big differences in the wealth gap are present among countries. Investments in real estate are more common and of higher value among native headed households in all the considered countries. Italy exhibits a higher relative immigrant/native gap compared to Luxembourg and Germany in the conditional mean value of total financial assets. Non-home secured debt at the median level is quite equally distributed among immigrant and native headed households in all three countries considered. Cross-country differences in homeownership rates and mortgage take-up can be to a large extent be explained by differences in institutions across countries (see Bicakova and Sierminska, 2008 for an overview). The descriptive statistics presented in this and in the previous section corroborate the existence of a clear immigrant/native wealth gap in Luxembourg, Germany and Italy, despite the different economic and migration situations in the three countries. In the next sections we will analyse whether the immigration status of the household head has still a negative effect on wealth accumulation after controlling for relevant household characteristics. 5 Empirical Methodology and results In order to describe the immigrant to native wealth gap, we will follow a twofold approach. As a first step, we identify the existence and size of a conditional immigrant to native wealth gap at different points of the wealth distribution in Germany, Italy and Luxembourg (simultaneous quantile regression). In a second step, we will concentrate on the mean of the household net wealth distribution and assess how much of the gap is due to the differences in the mean level of the covariates and how much is due to the difference in the covariates coefficients (the Oaxaca-Blinder decomposition). For Germany and Luxembourg coefficients and standard errors of the generalised quantile regression and the Oaxaca decomposition are adjusted for the variability between imputations according to the combination rules by Rubin (1987). 18

19 The distribution of net wealth is usually skewed to the right. As a result, the empirical model is commonly estimated in logarithmic terms (see among others Shamuddin and DeVoertz, 1988), however, given the high number of zeros and the presence of negative values the use of a logarithmic transformation is not appropriate, leading to reduction of the number of observation and to nondesirable censoring of the database. Therefore, we use the inverse hyperbolic sinus transformation (IHS), with θ equal to 1, which approximates the logarithm for positive values of the underlying variable, is defined at zero and approximates the negative logarithm for negative values of the underline variable (Johnson, 1949; Burbridge at al., 1988). For applications in similar studies see among others Cobb-Clark and Hildebrand (2006b), Pence (2006) and Gale and Pence (2006). 5.1 Quantile regression In order to identify the conditional immigrant to native wealth gap, we estimate an IHS simultaneous quantile regression model of the determinants of net household wealth for Germany, Italy and Luxembourg. Simultaneous quantile regressions have widely been used in literature to analyse the immigrant to native wealth gap (e.g. Zhang, 2003, Cobb-Clark and Hildebrand 2006b; Doiron and Guttmann, 2009; Gibson et al., 2007, Sinning, 2007; Lewin-Epstein and Semyonoy, 2011 and Bauer et al., 2011). Our approach closely relates to that of Cobb-Clark and Hildebrand (2006b) who used the IHS simultaneous quantile regression model to identify the immigrant to native wealth gap in the U.S., to Sinning (2007) who analysed the same topic using an older vintage of the German SOEP, as well as to Bauer et al. (2011). Following this approach the q th -quintile of the IHS wealth distribution can therefore be expressed as: 1 sinh ( ) = α + β X + β I + ε (1) w q 1 2 where I is an indicator function that takes the value 1 if the household head is an immigrant and zero if (s)he is a native (in other specifications of the empirical model, we also distinguish between different nationalities of immigrants) and X is a vector of household characteristics, including life cycle theory related variables (age and age squared of the household head), household composition related variables (marital status and number of children aged less than 18), education attainments of the household head and IHS 19

20 transformation of disposable income. 9. Estimations for Germany and Italy include regional dummy variables; in Germany, regional dummies represent the Bundesländer, in Italy the North, Centre and South macro regions. In order to explore different dimensions of the impact of immigration on household net wealth and to take into consideration country specific characteristics, different model specifications are estimated. These include cohort of immigration to control for different waves of immigration (Shamuddind and DeVoertz, 1988, Zhang, 2003, Cobb-Clark and Hildebrand 2006, Doiron and Guttmann, 2009, Bauer et al., 2011, Sinning 2007 and Lewin- Epstein and Semyonoy, 2011) and age at immigration of the household head to control for assimilation. Table 6 reports the results of the simultaneous quantile regression at the median level (q=50). These results corroborate the descriptive results provided in the previous section. Even after controlling for household characteristics, columns 1, 3 and 4 show as, the median net wealth of immigrant households is estimated 10 to be about 95%, 94% and 70% lower the corresponding aggregate for native-born households in Germany, Italy and Luxembourg, respectively. This indicates that immigrant households with similar characteristics to nativeborn households have a wealth disadvantage; this result is in line with the findings of the previous literature (with the relevant exception of Zhang 2003). The results for Germany are consistent with those of Sinning (2007). In all three countries, household median net wealth exhibits a positive elasticity with respect to disposable income (as in Cob-Clark and Hildebrand, 2006b). Age and age squared of the household head present significant coefficients whose signs are in line with life cycle theory. However, the coefficient of age squared is particularly low, probably reflecting lower dis-saving necessities due to the presence of a strong welfare system, such as a public or state pension as is the case in all three countries considered. Secondary and tertiary education of the household head has a positive and significant effect on net wealth with the strongest effect of tertiary education. The result for Germany is consistent with comparable findings of Bauer et al. (2011) and Sinning (2007). A household with a separated or divorced household head has a lower median wealth than the basic category in all countries considered. In Germany and Luxembourg, being single has a clear negative effect on household median net wealth with respect 9 According to the definition of the Luxembourg Income Study, the definition of disposable income is equal to gross income minus income taxes and contributions. 10 Starting with the coefficients of equation (1) in table 6, this percentage is calculated, as follows: % 2 (w) = 100 * (exp( β ) -1) (Wooldridge, 2009) 20

21 to the base category (being married). Additionally, if the household head is widowed the median net wealth is reduced significantly in Germany while the effect seem to be positive in Italy. Columns 2 and 5 present the estimation results, which include controls for the 5 principal foreign countries of birth plus indicator variables representing immigrants born in other EU-15 countries and other non EU-15 countries for Luxembourg and Germany. In the Italian dataset, the information of the country of origin is not available. The results are robust to this alternative specification with the exception of secondary education of household head dummy for Luxembourg whose coefficient becomes non-significant. Being of foreign nationality at birth has a negative effect on median net wealth regardless of the country of birth in question in both countries, with the sole relevant exception of Italian origin immigrants in Germany. 21

22 Table 6: Quantile regression, Q=50 Dependent variable IHS net wealth Germany Italy Luxembourg Age *** *** *** 0.21 *** *** (13.595) (13.287) (6.47) Age-squared *** *** *** *** *** ( ) ( ) (-5.25) (-8.941) (-6.980) Secondary educ *** *** *** ** 0.15 (5.133) (4.581) (5.30) Tertiary educ *** *** *** *** ** (8.764) (7.637) (3.58) Number of children *** *** * (4.055) (3.968) (1.87) (-1.176) (-0.373) Disposable income *** *** *** *** *** (3.377) (3.118) (18.63) Never Married *** *** * *** (-7.276) (-7.579) (1.87) (-0.523) (-2.867) Separated/divorced *** *** * *** *** (-7.943) (-8.457) (-1.74) (-3.880) (-4.001) Widowed *** *** *** (-4.842) (-4.918) (2.94) Gender *** *** * (-2.584) (-2.763) (0.20) (-1.619) (-1.869) Immigrant *** *** *** (-6.216) (-12.62) ( ) Country of birth Country *** ** (-4.565) (-2.131) Country ** ** (-1.983) (-2.549) Country *** *** (-5.562) (-6.878) Country *** *** (-3.196) (-3.955) Country *** (-1.476) (-2.608) Other EU *** *** (-2.874) (-5.258) Other non EU *** *** (-3.263) (-2.718) Constant *** *** *** *** *** (-6.127) (-5.033) (-5.81) (-4.848) (-2.578) Regional Dummies yes yes yes no no Pseudo R-squared Number of obs Source: Authors calculation. Data from LWS and Bank of Italy Note: Simultaneous quantile regression, bootstrapped std. errors (500 reps), T-statistics in parentheses. Country 1-5 refer to Portugal, Belgium, France, Germany and Italy for the Luxembourg regression and Turkey, Poland, Russia, Kazakhstan and Italy for the German regression. The base category is native, married with primary education. Regional dummies represent the Bundesländer for Germany, and north/centre/south marco regions for Italy. * for p<.10, ** for p<.05, and *** for p<.01 22

23 Table 7: Quantile regression for the 75 th and 90 th percentile Dependent variable IHS net wealth Quantile regression, Q=75 Quantile regression, Q=90 Germany Italy Luxembourg Germany Italy Luxembourg Age *** *** *** *** *** *** (21.640) (7.14) (13.602) (15.830) (7.34) (8.772) Age-squared *** *** *** *** *** *** ( ) (-4.96) ( ) ( ) (-4.93) (-7.089) Secondary educ *** *** *** *** *** *** (7.620) (4.58) (3.050) (5.812) (4.10) (3.697) Tertiary educ *** *** *** *** *** *** (15.182) (5.67) (3.698) (13.736) (4.22) (4.189) Number of children *** *** (5.073) (0.72) (0.748) (4.199) (1.33) (1.169) Disposable income *** *** *** *** *** (3.357) (23.74) (4.842) (2.773) (9.66) (1.364) Never Married *** * ** *** * ( ) (1.79) (-2.161) (-5.244) (1.49) (-1.914) Separated/divorced *** *** *** *** ( ) (-0.40) (-4.317) ( ) (-0.31) (-5.262) Widowed *** *** *** (-6.614) (2.66) (0.067) (-4.824) (0.15) (-1.350) Gender * (-1.049) (-0.82) (-0.986) (-1.837) (0.16) (-0.075) Immigrant *** *** (-7.72) (-7.04) Country of birth Country *** *** *** *** (-6.112) (-5.960) (-3.649) (-5.668) Country *** *** *** *** (-4.033) (-2.918) (-4.931) (-3.097) Country ** *** *** *** (-2.371) (-7.683) (-5.476) (-4.833) Country *** *** ** (-1.622) (-4.748) (-4.093) (-2.442) Country ** ** * ** (-2.423) (-2.275) (-1.938) (-1.977) Other EU *** *** *** (-3.033) (-3.490) (-1.242) (-4.243) Other non EU *** *** *** *** (-3.659) (-6.264) (-4.329) (-5.047) Constant *** *** *** *** (12.927) (-1.04) (1.422) (24.262) (2.73) (4.945) Regional Dummies yes yes no yes yes no Pseudo R-squared Number of obs Source: Authors calculation. Data from LWS and Bank of Italy Note: Simultaneous quantile regression, bootstrapped std. errors (500 reps), T-statistics in parentheses. Country 1-5 refer to Portugal, Belgium, France, Germany and Italy for the Luxembourg regression and Turkey, Poland, Russia, Kazakhstan and Italy for the German regression. The base category is native, married with primary education. Regional dummies represent the Bundesländer for Germany, and north/centre/south marco regions for Italy. * for p<.10, ** for p<.05, and *** for p<.01 As mentioned previously, in order to take into account the varying effects of immigrant status across the wealth distribution, simultaneously to the correspondent median regression we estimated quantile regressions at the top of the household net wealth distribution, for the 75 th and 90 th quantiles. The results are reported in Table 7. The immigrant/native wealth gap is wide and 23