Welfare migration in Europe

Labour Economics 16 (2009) 353–363

Contents lists available at ScienceDirect

Labour Economics j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / l a b e c o

Welfare migration in Europe☆ Giacomo De Giorgi a, Michele Pellizzari b,⁎ a b

Stanford University, United States Department of Economics, Bocconi University, via Roentgen 1, 20136 Milan, Italy

a r t i c l e

i n f o

Article history: Received 27 March 2008 Received in revised form 28 December 2008 Accepted 14 January 2009 Available online 3 February 2009 JEL classification: J61

a b s t r a c t The enlargement of the European Union has increased concerns about the role of generous welfare transfers in attracting migrants. This paper explores the issue of welfare migration across the countries of the preenlargement European Union and finds a significant but small effect of the generosity of welfare on migration decisions. This effect, however, is still large enough to distort the distribution of migration flows and, possibly, offset the potential benefits of migration as an inflow of mobile labour into countries with traditionally immobile native workers. © 2009 Elsevier B.V. All rights reserved.

Keywords: EU enlargement Migration Welfare state

1. Introduction The recent enlargement of the European Union has been followed by migration flows of around 250,000 persons per year from the new member countries into the Union. These flows were partly anticipated by statistical projections prepared at the outset of the enlargement, although transitional restrictions to migration were applied by almost all countries of the EU-15 (Alvarez-Plata et al., 2003; Dustman et al., 2003). The effects of such increased migration have been analysed by academics as well as policy makers (Boeri and Brücker, 2001; Sinn, 2001). Among other considerations, much concern has arisen about the role of generous welfare transfers in attracting migrants (see Kvist, 2004; Sinn, 2004a,b). This paper estimates the extent to which welfare generosity affects the location decisions of migrants in the 15 countries of the preenlargement Union. Migration, and particularly migration from Eastern to Western European countries, can, in fact, have both positive and negative effects on the economies of the receiving countries. On

☆ Our special thanks go to Giovanna Albano, Gaetano Basso, Francesco Fasani, Mauro Maggioni and Sara Pinoli who have substantially contributed to this project with skillful assistance and comments. We would also like to thank all seminar participants at the University of Verona, the University of Salerno, the Barcelona ENTER conference and the XX AIEL conference in Rome. Tito Boeri, Rodolfo Helg, Francesca Mazzolari and all participants to the EUfunded FLOWENLA Project provided important comments and suggestions. All errors are our own responsibility. Financial support from Bocconi University is gratefully acknowledged. ⁎ Corresponding author. Tel.: +39 02 58363070. E-mail address: [email protected] (M. Pellizzari). 0927-5371/$ – see front matter © 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.labeco.2009.01.005

the one hand, the enlargement, involving countries with younger and growing populations, is expected to alleviate the financial strain of many member states' public pension systems. On the other hand, since migrants use the welfare state relatively more than natives (Boeri et al., 2002), they may also increase pressure on its sustainability. Hence, immigration could be potentially driven, among other things, also by the generosity of the welfare system. If this is the case, the large variation in welfare arrangements across Europe may indeed distort the distribution of migration flows across member countries. An efficient distribution of such flows would be extremely beneficial to the European labour markets. In fact, contrary to the United States, European workers are very immobile, making it difficult for the economy to adjust to asymmetric changes in labour demand. This is particularly true after the introduction of the Euro, that eliminated exchange rate variation, an instrument that was normally used in the past for stabilizing the economy against asymmetric shocks. Migrants are, by definition, a very mobile form of labour and, as long as they move into areas where labour is scarce and wages are high, they can play a crucial role in counterbalancing the low mobility of native European workers. High wages and high employment probabilities, rather than the generosity of welfare, should be the main driving forces of migration. To the extent that migrants choose their destination on the basis of the generosity of welfare, the potential benefits of acquiring a more mobile labour force through a softening of border restrictions will be lost. Moreover, the costs (in terms of higher social expenditure) of larger migration flows may be unevenly distributed, at the expense of countries and regions where financial pressure on welfare is already high. The issue of welfare induced migration has already received a good deal of attention in the economic literature. Early studies were generally

354

G. De Giorgi, M. Pellizzari / Labour Economics 16 (2009) 353–363

based on aggregate data and showed that states with more generous benefits were also associated with slightly larger inflows of migrants. With aggregate data, however, it is impossible to focus on those subgroups of the population (large families, single parents, women, etc.) who are more likely to consider the generosity of welfare an important element in their migration decisions. Using micro-data other authors (Blank, 1988; Borjas, 1999; Gramlich and Laren, 1984; Meyer, 2000) still found a significant impact of welfare on migration. Even the most recent strand of papers (Enchautegui, 1997; Gelbach, 2004; Levine and Zimmerman, 1999; McKinnish, 2005; McKinnish, 2007; Walker, 1994), that use more sophisticated identification strategies, confirm the existence of some welfare migration, especially for the most vulnerable groups of the population. Nevertheless, these effects are generally considered too small to seriously affect policy making.1 All these studies, however, are based on the United States, where geographical mobility is relatively high among native workers and the role of migration as a stabilizer after asymmetric shocks is somewhat less important. The current paper focuses on Europe and shows that the effect of welfare generosity can be large enough to offset the changes in migration patterns that would arise from an asymmetric shock to unemployment. Besides its focus on geographical labour reallocation, this paper also contributes to the literature by providing the first estimates of welfare migration across European countries. Our empirical analysis, using data from the European Community Household Panel (ECHP), shows that migrants into the pre-enlargement European Union choose their destination on the basis, among other things, of the generosity of welfare. Identification comes from variation in aggregate (but time-varying) measures of welfare benefits at the time of migration within individuals and across potential destination countries. Our estimates rest on the assumption that other unobservable aggregate country characteristics that may affect location decisions (culture, natives' attitudes, social networks) vary at a lower than yearly frequency. This assumption is somewhat stronger than those required for identification in the most recent papers in this area (Gelbach, 2004; McKinnish, 2005, 2007). However, under such stronger identification requirements we are able to estimate what is perhaps the most interesting policy effect for the European debate, that is the effect of welfare generosity of migrants' location decisions across countries. In fact, more sophisticated identification strategies typically come at the cost of being able to identify only very marginal effects around state or regional borders. The estimates that could be produced with such an approach in Europe are probably even less generalizable than in the United States. Given the large variation in population and cultural characteristics, people living around the country borders in Europe are very unlikely to be representative of the entire population. The process of economic and social integration has led to harmonisation of several policy dimensions across Europe but very little has been done on the welfare side, most likely because of political impediments. From a policy perspective, our results highlight the potential benefits of reducing the heterogeneity of welfare provisions across European countries. For example, one intervention that has been discussed at several stages of the integration process is the introduction of a European minimum

1

Some of these papers (Borjas, 1999; McKinnish 2005) focus also on welfare recipiency, an issue that we do not address directly in this paper mostly due to data limitations. However, we believe that this is a relatively minor limitation since, at the time of migration, individuals typically do not know whether and when they will become welfare recipients. Moreover, most countries currently apply restrictions to welfare access by immigrants who can apply for benefits only after some years of residence in the country. Hence, welfare generosity is likely to play a role only for those individuals who are planning to reside in the destination country for a relatively long period of time.

income, a minimum of a level of assistance benefit of last resort that all countries in the Union should guarantee to those citizens whose household income falls below a given threshold.2 Without explicitly addressing the political economy of such an intervention, this paper informs about one particular channel through which welfare harmonisation might be beneficial. The paper is organised as follows. Section 2 briefly describes the data. Section 3 presents our main results from the empirical analysis of the correlation between welfare generosity and migration decisions. In Section 4 we report a series of robustness checks that corroborate our main findings. Section 5 presents a simple simulation exercise that shows how welfare-induced migration interacts with asymmetric shocks to unemployment. Finally, Section 6 concludes. 2. The data 2.1. The European Community Household Panel The European Community Household Panel is a panel dataset of households covering all the 15 pre-enlargement countries of the European Union. The ECHP started in 1994 and 8 waves of data have been released so far, covering the period from 1994 to 2001. Not all countries entered the survey at the same time and for three of them – Germany, Luxembourg and the United Kingdom – the original sample has been replaced after the first three waves with harmonised versions of household panels already been produced nationally: the German Socio-Economic Panel (GSOEP), the Luxembourg's Socio-Economic Panel (PSELL) and the British Household Panel Survey (BHPS). Identical sampling procedures are applied in all countries and individuals are administered the same set of questions, thus making the data highly comparable across countries. Respondents to the ECHP questionnaire are also asked various questions about migration and citizenship. In what follows migrants are identified as those who either indicated to be citizens of a non EU-15 country or were born abroad and lived in a different country before arriving in the place of current residence. We adopt this dual definition because it allows us to cover the largest set of countries (e.g. there is no information on citizenship for Germany but migration trajectories are reported) and also because it is the one that more closely approximates official migration data from the OECD (OECD, 2001), as shown in Table 1.3 There still exist discrepancies, which are sometimes large, between the official data and our own estimates of the stock of foreign-born population but, considering that the ECHP is not explicitly designed for the study of migration and that the sample of migrants in some countries is relatively small, the overall performance of the ECHP on this issue can be considered satisfactory. Another important piece of information available from the ECHP is the year of arrival in the current country of residence, which allows us to match each individual migrant with the economic conditions of all possible destination countries at the time the decision to migrate was taken. However, since data on welfare generosity and macroeconomic conditions are not easily available for the long past years, only migrants who arrived in the country of present residence in or after 1970 have been considered. In Section 4 we present additional results based on a restricted set of years and countries for which more detailed and disaggregated data on labour market conditions are available.

2 In a working paper version of this study we estimated the costs of a programme of this type under several alternative financing hypotheses. We find that a European^ wide minimum income would cost approximately two thirds of what is currently spent on similar schemes (social assistance and housing). 3 In an earlier version, we defined migrants simply as those who were born abroad and had lived abroad before coming to the country of current residence. The empirical estimates produced using this alternative definition were qualitatively similar to the ones presented here. However, the dual definition used in the current version of this paper allows us to identify a larger group of migrants and thus to produce more precise estimates. Moreover, it also replicates the official statistics more closely (see Table 1).

G. De Giorgi, M. Pellizzari / Labour Economics 16 (2009) 353–363 Table 1 Comparison of data sources. Stock of foreign population in 2001 Country Austria Belgium Denmark Finland France Germany Greece Ireland Italy Luxembourg Netherlands Portugal Spain Sweden United Kingdom

OECDa

ECHPb

9.40 8.21 5.00 1.90

6.48 8.63 3.46 3.51 7.52 13.05 4.41 3.73 1.57 35.58 1.61 2.47 1.64 3.87 3.53

8.88 7.00 3.90 2.36 37.55 4.26 2.17 2.74 5.34 4.39

a

Source: Trends in International Migration, OECD, 2003 edition. Table A.1.5. Foreign population defined as persons who declare to be citizens of a foreign country or that they were born in a foreign country (or both).

355

For many countries the early nineties have been the years of most intense immigration but the data indicate a large variation in the sequence of migration waves across countries, resulting in a rather even distribution of arrivals for the entire Union. It is worth mentioning here a couple of caveats of this sample. It is a known fact that some countries are often used as a port of entry (e.g. Italy, Spain, Greece) and the final destination of migration might not be the one observed in the data. The ECHP does not contain questions about the intention to leave the country in the future. We can only argue that most immigrants, who transit to one country with the intention to move to another one, typically do that illegally and, even if they register, they remain in the port-of-entry country only for a very short period. Thus, they are very unlikely to be sampled in the ECHP. This also implies that illegal migrants will generally not be covered by our analysis, a feature of the data that should not be particularly problematic. In fact, in most countries unregistered migrants cannot access the types of benefits that we consider, with some exceptions for minimum income schemes and some health related benefits.

b

The sample has also been restricted to individuals who were aged between 15 and 55 at the time of arrival. Unfortunately, the data on welfare generosity are not available for Luxembourg and this country has been dropped from the sample used in Sections 3 and 4. Ultimately, the empirical exercise of Sections 3 and 4 uses a sample of 3038 migrants from outside the pre-enlargement EU-15, distributed across 14 countries of the European Union according to Table 2. Summary statistics are also shown in Table 2, together with the distribution of the years of arrival for each destination country. The data show that migrants usually move in their late twenties and that they are rather evenly distributed by gender although in some countries, like Italy, the presence of women is particularly high (probably because they cluster into female-dominated occupations, like housekeeping and nursing). The distribution of education is more varied with the Nordic and the Anglo-Saxon countries attracting the most educated migrants (with some exceptions like Greece and Spain). However, as we discuss more in detail later in Section 4, education is only recorded at the time of the interview and therefore it might not reflect the skill level at the time when the migration decision was taken. Hence, we will not use it in our baseline specification and consider it only in the robustness checks.

2.2. OECD Data-base on Benefit Entitlements and Replacement Rates Data on welfare generosity come from the OECD Data-base on Benefit Entitlements and Replacement Rates, which has been often used in the literature to describe the generosity as well as other characteristics of the welfare systems in the countries of the OECD. In particular, this database measures the generosity of welfare by computing the ratio between income out of work – i.e. from welfare benefits – and income in work – i.e. some measure of the average wage. These figures, called replacement rates, are computed for several family types (single, couple with and without dependants), income levels (average earnings or 2/3 of the average earnings) and various durations of unemployment (from the first month up to the 60th month), as some benefits, like unemployment insurance, are paid only for a limited period of time. The OECD produces these figures taking the average wage in the manufacturing sector (what is usually called the wage of the “average production worker” or APW) as a measure of earnings and, on the basis of each country's regulations, computes the amount of benefits a typical worker with that level of earnings is entitled to in case of unemployment. The calculation considers all cash benefits, i.e. unemployment benefits, housing benefits, family benefits and minimum income programmes. The OECD Data-base on Benefit Entitlements and Replacement Rates exists in two versions. The first one contains gross replacement

Table 2 Summary statistics. Country of destination

Sample size

Age at arrival

Female

Austria Belgium Denmark Finland France Germany Greece Ireland Italy Netherlands Portugal Spain Sweden United Kingdom Total

324 105 127 184 309 950 273 30 85 64 120 114 139 214 3038

27.9 (8.89) 25.9 (8.07) 26.5 (8.66) 30.4 (7.77) 27.2 (9.07) 25.6 (7.87) 28.2 (8.64) 30.8 (7.41) 27.7 (7.68) 26.2 (8.08) 29.9 (11.33) 28.1 (8.75) 28.3 (8.72) 27.5 (8.97) 27.2 (8.63)

0.54 0.49 0.56 0.48 0.56 0.53 0.61 0.40 0.66 0.59 0.56 0.52 0.61 0.51 0.54

Source: ECHP 1994–2001.

Secondary education

Tertiary education

Means (and std.dev.) 0.46 0.31 0.28 0.43 0.13 0.28 0.40 0.40 0.52 0.08 0.28 0.35 0.49 0.34 0.33

0.17 0.36 0.28 0.36 0.24 0.07 0.33 0.40 0.12 0.05 0.18 0.39 0.28 0.41 0.21

Distribution of year of arrival (frequencies) b1975

b1980

b 1985

b 1990

0.14 0.23 0.13 0.10 0.28 0.45 0.05 0.20 0.09 0.14 0.09 0.10 0.03 0.19 0.24

0.22 0.41 0.20 0.19 0.49 0.59 0.16 0.40 0.25 0.27 0.54 0.22 0.10 0.34 0.38

0.30 0.57 0.34 0.41 0.67 0.68 0.29 0.53 0.41 0.34 0.69 0.39 0.18 0.47 0.50

0.49 0.82 0.67 0.61 0.82 0.81 0.43 0.77 0.66 0.64 0.85 0.67 0.37 0.66 0.68

356

G. De Giorgi, M. Pellizzari / Labour Economics 16 (2009) 353–363

Fig. 1. Gross welfare benefits by country.

rates, i.e. computed without considering the taxation of benefits, which are indeed taxed in many OECD countries. This version of the database goes back as far as 1960. The second version includes taxes in the calculations but, unfortunately it only covers the years from 1995 onwards. In our analysis we need to attach measures of welfare generosity to all migrants who arrived in the country of current residence since 1970 and we are then forced to use gross benefits from the first version of the database.4 Since our interest lies in the separate identification of the effects of the wage and of welfare benefits on the destination of migrants, it has been necessary to reconstruct the level of the benefit from the replacement rate by multiplying the ratios by the APW earnings. Moreover, given that our identification strategy is based on the comparison of the combinations of wages, employment possibilities and welfare generosity across the destinations countries, the benefits have been adjusted by purchasing power. Fig. 1 shows the evolution over time and across countries of the measure of welfare generosity that will be used for the empirical analysis of Sections 3 and 4. These numbers are computed averaging the total amount of welfare benefits (in PPP-weighted current euros)

4 The ECHP samples individuals in 1994 and then follows them over time. Hence, the very few migrants who arrived after 1995 are necessarily family reunions into sampled households.

over two income levels (APW and 2/3 of APW earnings), 3 family types (single, couple with dependent spouse, couple with working spouse) and several durations of unemployment (from 0 to 60 months). 5 The data in Fig. 1 confirm what is a well known fact, that welfare institutions vary considerably across countries and offer a good deal of variation that can be exploited for econometric identification. For the purpose of our analysis it would be extremely helpful to have information on the evolution of welfare benefits in the sending countries. Unfortunately this type of data are currently not available, although the OECD has recently extended most of its datasets to most Eastern European countries. Additionally, even if such information was available the ECHP does not allow identifying the exact country of origin.6 3. Welfare generosity and the choice of migration: an empirical model In this section we estimate a very standard conditional logit model that describes the location decisions of migrants. We are implicitly

5 Monetary amounts are expressed in PPP-adjusted ECUs, the pre-euro currency of ^ ^ the Union. 6 Although the survey contains questions on the country of origin, most of them were simply reaggregated at the level of the continent. Missing values prevent the use ^ of any more disaggregated indicator.

G. De Giorgi, M. Pellizzari / Labour Economics 16 (2009) 353–363

assuming that the decision to migrate has already been taken and the only choice to be made concerns the country of destination. Meyer (2000) also uses a similar model and compares it with other empirical strategies that have been implemented in the literature. Suppose that each individual i is faced with D alternative destination countries and utility obtainable from migrating into country d is: U ðxid Þ = V ðxid Þ + ηid

ð1Þ

where V(xid) is a deterministic function of a set of intrinsic characteristics of country d, possibly varying across individuals, xid (i.e. prevailing unemployment rate, average wages, etc.). ηid is a random utility term for individual i moving to country d . For simplicity, assume that V(xid) is linear: V(xid) = xidβ. Individual i will, therefore, choose destination d if: U ðxid ÞN U ðxik Þ for all k ≠ d

ð2Þ

Assuming that the random utility components ηid's are all independently and identically distributed (over both i and d) according to a type I extreme-value distribution, the probability that individual i chooses destination d can be rewritten as:7


Pr V ðxid Þ + ηid z V ðxik Þ + ηik

for all

 k≠d =

xid β

PDe

k = 1

exik β

= Pid ðxj βÞ: ð3Þ

Eq. (3) clearly indicates that the identification in this model comes from comparing the same individual faced with different alternative destinations. In fact, since each individual i is only observed taking one destination d, all individual characteristics that do not vary across countries (age, gender, etc.) cancel out in the specification of Pid. The log-likelihood function for a sample of N migrants facing D destinations can, then, be written as:

LðβÞ =

N D X X i=1

fid log Pid ðxj βÞ

ð4Þ

d=1

where fid = 1 if individual i chooses destination d and zero otherwise. The model in Eq. (4) is estimated under various specifications of the utility function U(xid). Our main interest lies in the identification of the coefficients on country-specific characteristics, such as the unemployment rate, the average real wage and the average benefit level in each destination country. These variables are measured for each country d in the year in which each individual i settled into his/ her country of current residence. Importantly, the set of destination attributes, xid, also includes a set of destination country dummies interacted with 4 time period dummies for arrivals before 1980, between 1980 and 1985, between 1985 and 1990 and after 1990, so that destination-specific effects are allowed to change at discrete time intervals. This is, in fact, the key identification assumption of our model. Many unobservable destination-specific characteristics (the strictness of migration laws, networks of migrants already present in the country, etc.) are likely to affect migration decisions. The set of destination country dummies interacted with dummies for discrete time intervals is

7

We will relax the independence assumptions later in Section 4.

357

meant to control for such characteristics. The underlying hypothesis is that these effects vary over time at a lower than yearly frequency. In the particular specification that we adopt here such frequency is defined at 5-year intervals but, as we show in Section 4, our results are robust to a finer specification, down to 3-year periods. Conditional on the full set of interacted country and time effects, the coefficients that we estimate on our variables of interest (the unemployment rate, the wage rate and the average benefit level) are identified by variation within individuals, across countries and over 5year time cells. Such variation is generated by all types of economic fluctuations as well as by changes in policy, which are particularly important for benefit levels. As we also stressed in the introduction, this set of assumptions differentiates our approach from the most recent analyses of welfare migration (Gelbach, 2004; McKinnish, 2005, 2007). Those papers adopt identification strategies that exploit variation across borders and subgroups of the population. In fact, the use of cross-border variation in welfare generosity within the United States seems to be the most popular strategy currently being used in the best known papers. We believe that such an approach could hardly help identifying welfare-induced migration across European countries. For obvious historical and cultural reasons the allocation of people in the vicinity of either sides of any given country border can hardly be considered exogenous in the European context. So, even if data with the necessary characteristics (large sample size and cross-country comparability) to carry out such analysis were available, we doubt that exploiting cross-border variation could be a fruitful approach in our specific application. Moreover, the estimates that could be produced with such approach in Europe are probably even less generalizable than in the United States. Given the large variation in population and cultural characteristics, people living around the country borders in Europe are very unlikely to be representative of the entire population. In fact, the object of our empirical investigation is really the extent of crosscountry rather than cross-region (or any other sub-national geographical entity) welfare migration. Under our identification requirements we are able to estimate what is perhaps the most interesting policy effect for the European debate, that is the effect of welfare generosity of migrants' location decisions across countries. In fact, more sophisticated identification strategies typically come at the cost of being able to identify only very local effects around state or regional borders. However, it is fair to acknowledge that, due to the strong identification assumptions, our estimates are still likely to suffer from omitted variable bias due to unobservable country characteristics. We return to this issue in Section 4 where we present a series of robustness checks suggesting that the extent of such bias should be limited. Table 3 reports the main results of our empirical exercise. In the first column of Table 3 we simply replicate the known result that employment possibilities and wages are the main determinants of the destination of migration. Our estimates indeed confirm that migrants tend to go into countries with lower unemployment rates and higher real wages. The second column of Table 3 repeats the same estimation adding to the set of explanatory variables the interactions between the log of the unemployment rate and of the wage with some individual characteristics, to check whether some groups are particularly sensitive to the conditions of the labour market. The results indicate that those who migrate early in their lives (before 25 years old) also appear to attach less importance to the current labour market conditions, possibly because their decision is also affected by longer term factors related to investments in human capital. Interestingly, there seems to be no detectable differences between men and women. The third column of Table 3 introduces our measure of welfare benefits associated with the year of arrival in the destination country

358

G. De Giorgi, M. Pellizzari / Labour Economics 16 (2009) 353–363

Table 3 14-countries conditional logit estimates for the destination of migration. Only migrants from outside the EU-15, arrived between 1970 and 1994 (log) unemployment rate [lURATE]a (log) wage [lWAGE]b (log) welfare benefit [B]c

[1]

[2]

[3]

[4]

− 0.517⁎⁎⁎ (0.093) 4.449⁎⁎⁎ (0.704) –

−0.499⁎⁎⁎ (0.100) 4.666⁎⁎⁎ (0.715) –

− 0.467⁎⁎⁎ (0.100) 4.692⁎⁎⁎ (0.725) 0.377⁎⁎⁎ (0.120)

− 0.443⁎⁎⁎ (0.100) 4.439⁎⁎⁎ (0.733) 0.479⁎⁎⁎ (0.127)

Interaction terms: individual characteristics (at the time of arrival) with… (Age at arrival − 25) ⁎ [lURATE] – 0.008⁎ 0.008⁎ (0.004) (0.004) Female ⁎ [lURATE] – − 0.074 − 0.076 (0.074) (0.073) (Age at arrival − 25) ⁎ [lWAGE] – − 0.064⁎⁎⁎ −0.064⁎⁎⁎ (0.013) (0.013) Female ⁎ [lWAGE] – −0.006 0.014 (0.239) (0.241) (Age at arrival − 25) ⁎ [B] – – – Female ⁎ [B]

–

–

–

Observations Individuals Log-likelihood

42058 3038 −6564.39

42058 3038 − 6548.44

42058 3038 − 6543.22

0.008⁎ (0.004) − 0.114 (0.075) − 0.069⁎⁎⁎ (0.016) 0.471 (0.293) 0.002 (0.004) −0.175⁎⁎⁎ (0.064) 42058 3038 6539.27

All specifications include a full set of destination country dummies, 4 5-year period dummies and the interactions between these two sets of dummies. Standard errors in parentheses. ⁎Significant at 10%; ⁎⁎significant at 5%; ⁎⁎⁎significant at 1%. a Source: OECD Economic Outlook. Values in percentage. b Annual compensation per employee in the private sector. Source: OECD Economic Outlook. All values are expressed in PPP/ECU. c Monthly benefit received a typical person aged 40 who has continuously worked and paid contributions since the age of 18, averaged over a period of 60 months of nonemployment, two earning levels (100% and 33% of the earnings of the average production worker) and three family types (single, one-earner couple, two-earners couple). Source: OECD Data-base on Benefit Entitlements and Replacement Rates. All values are expressed in PPP/ECU.

(in logs). The results show a positive and significant coefficient, indicating that, indeed, migrants are more likely to move into countries with more generous welfare benefits. This effect is reinforced when our measure of welfare generosity is interacted with the individual characteristics, as in column 4. Among the interaction coefficients, only the one with the gender dummy is significant, indicating that women are relatively less attracted by high-benefit countries. To get a first idea of the magnitude of these estimated effects, one could simply compare the coefficients of the log wage and the log benefit, which share the same unit of measurement. For example, from the figures in column 4 the effect of wages is more then 10 times larger than that of benefits. More formally, one could look at the effect on migration into one country of changes in unemployment, wages and benefits either in the same or in the other countries. In fact, according to Eq. (3), the effect of a change in the j-th characteristic of country d on the probability of migrating in country d is equal to: APid ðx jβÞ = Pid ðx jβÞ½1 − Pid ðxj βÞβj Axjid

ð5Þ

while the effect on the same probability of a marginal change in the same j-th characteristic of another country k ≠ d is: APid ðx jβÞ = − Pid ðxj βÞPkd ðxj βÞβj : Axjik

ð6Þ

Using Eq. (5), Table 4 reports the direct effects of a change of one standard deviation in each of the three variables of main interest – the unemployment rate, the average wage and the generosity of welfare – as implied by the estimates in column 4 of Table 3 and computed for a representative person (male, aged 25 at the time of migration and arrived in the country of current residence in 1990). The standard deviations changes in the unemployment rate, the level of wages and the generosity of welfare are computed across countries, given that this is the variation exploited by our identification strategy. Alternatively, we could consider changes measured in terms of the distributions of such variables over time and within countries. However, such an alternative approach would imply different changes in each country, thus making it difficult to compare. As indicated by the last two columns of Table 3, the effect of a onestandard deviation change in the unemployment rate (which is a huge change of more than 3 percentage points) is comparable to a similar change in welfare benefits only in a few countries. Perhaps, it is easier to interpret the ratio between the change in wages and benefits, since they share the same unit of measurement. The ratios in the last column of Table 4 show that increasing welfare benefits by onestandard deviation induces an increase in migration flows that is on average only 37% of what a similar change in wages would have generated. Only in Italy is the size of the two effects comparable but in this country benefits are extremely low and the simulated change of one-standard deviation leads to benefits more than 7 times higher. The full matrix of both direct and cross-marginal effects computed for the reference person according to Eqs. (5) and (6) is reported in Table A1 in the Appendix. These numbers are used to conduct the thought experiment reported in Table 6, which tries to relate these estimates to the initial issue of labour reallocation. Before moving to that part, however, let us show some simple robustness checks that corroborate the results presented in this section. 4. Robustness checks The first robustness check that we present in this section shows that our results do not rest on the specific measure of welfare generosity that we adopt in the main specification of the model in Table 3. The OECD average benefit level is a very precise indicator of the generosity of welfare provisions in the country because it includes information on several types of programmes, from unemployment

Table 4 Effects of one-standard deviation increases for the reference persona. Effect of an increase of 1 standard deviation in:

Country

Relative size of the own benefit effect

Own unemp. Rate

Own wages

Own benefits

[1]

[2]

[3]

Mean

6.7

18163.9

4986.0

Std.dev.

3.1

3412.8

2993.7

Austria Belgium Denmark Finland France Germany Greece Ireland Italy Netherlands Portugal Spain Sweden United Kingdom Average

− 0.0278 −0.0081 −0.0094 −0.0226 −0.0099 −0.0393 −0.0131 −0.0013 −0.0039 −0.0069 −0.0074 −0.0052 −0.0227 −0.0133 −0.0136

0.0840 0.0314 0.0415 0.0611 0.0541 0.1141 0.0756 0.0091 0.0221 0.0225 0.0435 0.0443 0.0422 0.0491 0.0496

0.0291 0.0070 0.0066 0.0213 0.0167 0.0381 0.0478 0.0035 0.0308 0.0047 0.0101 0.0119 0.0110 0.0194 0.0184

a

Male, aged 25, arrived in 1990.

[3]/[1]

[3]/[2]

1.05 0.86 0.70 0.94 1.69 0.97 3.66 2.65 7.94 0.68 1.37 2.31 0.48 1.46 1.35

0.35 0.22 0.16 0.35 0.31 0.33 0.63 0.38 1.39 0.21 0.23 0.27 0.26 0.39 0.37

G. De Giorgi, M. Pellizzari / Labour Economics 16 (2009) 353–363

insurance to family cash transfers, and across household compositions. Measures of this type that cover a relatively long time span and also allow reliable comparisons across countries are very hard to find. Perhaps the single most obvious alternative is information from national accounts about the welfare expenditure of the public administration. Using the Comparative Welfare State Dataset, compiled by researchers at various universities (Northwestern University, University of North Carolina, Duke University and Indiana University) and based on OECD and ILO publications, we construct a measure of public expenditure on unemployment related benefits as an alternative measure of welfare generosity.8 Table 5 repeats all the estimates of Table 3 by replacing the OECD average benefit level with this indicator. The results are very similar, both qualitatively and quantitatively. Next, we explore more in detail the role of our key identification assumptions. The first and most important of such assumptions requires unobservable labour market conditions, such as migrants' networks or immigration policies, to be uncorrelated with welfare generosity across countries. Obviously, there is no way we can test this assumption. However, we can look at the correlation between observable labour market conditions and welfare generosity. If there is no clear trend in such a correlation, then we might expect unobservable characteristics to behave similarly. To this purpose, Fig. 2 plots the regression coefficients obtained from a series of year-to-year OLS regressions of unemployment rate and average wage levels on our measure of welfare generosity. The dotted lines represent 95% confidence intervals. These simple correlations are very rarely significant at conventional statistical levels. For unemployment, the point estimates also move above and below zero, while they are a bit more consistent when we look at wages. However, also in this case the correlation with welfare generosity becomes significant only towards the mid-late 1990's, i.e. after the period considered in our data. Consistently with these results, our estimates of the welfare effect change only marginally when the unemployment rate and the wage level are excluded from the set of controls. For example, in a very simple model similar to those in Table 3, the welfare effect increases by only about 20% when the labour market controls are excluded from the specification. Moreover, the 90% confidence intervals of the estimates produced with and without such controls overlap substantially, thus suggesting that one could not easily reject the hypothesis of their being equal.9 Additional robustness checks are provided in Table 6. In column 1 we run the same model of Table 3, column 4 but we reduce from 5 to 3 years the time intervals within which country-specific characteristic are allowed to vary. In other words we replace the 5-year period dummies with 3-year period dummies, always fully interacted with the set of destination country fixed effects. As the comparison of Table 3 and Table 6 readily shows, the results are very similar. Further reducing the length of the time intervals impedes identification as it limits the variation in the main variables of interest. In column 2 we further explore heterogeneity between skilled and unskilled workers. As we already mentioned earlier, education is only measured at the time of the interview, whereas, for the purpose of our analysis, it would be better to have this variable recorded at the time of migration. For this reason we prefer not to include education in our main specification in Table 3. However, when we also consider it and define unskilled workers as those with less than secondary education, results show that the generosity of welfare benefits has a stronger effect on such workers, although in this specification the estimated coefficient is statistically indistinguish-

8 The raw correlation of these two measures across countries varies between 0.70 and 0.75 depending on the year. 9 Detailed results are available from the authors upon request.

359

Table 5 Public expenditure on unemployment related benefits as a measure of welfare generosity. Only migrants from outside the EU-15, arrived between 1970 and 1994 (log) unemployment rate [lURATE]a (log) wage [lWAGE]b Public expenditure on unemployment related benefits (% of GDP) [B]c

[1]

[2]

[3]

− 0.925⁎⁎⁎ (0.137) 0.875 (1.254) 0.354⁎⁎⁎ (0.134)

− 0.924⁎⁎⁎ (0.142) 0.685 (1.281) 0.346⁎⁎ (0.135)

− 0.928⁎⁎⁎ (0.142) 0.611 (1.292) 0.367⁎⁎ (0.144)

Interaction terms: individual characteristics (at the time of arrival) with… (Age at arrival − 25) ⁎ [lURATE] 0.004 (0.005) Female ⁎ [lURATE] − 0.049 (0.087) (Age at arrival − 25) ⁎ [lWAGE] − 0.129⁎⁎⁎ (0.030) Female ⁎ [lWAGE] 0.754 (0.512) (Age at arrival − 25) ⁎ [B] Female ⁎ [B] Observations Individuals Log-likelihood

18882 1683 − 3140.26

18882 1683 − 3129.48

0.004 (0.005) − 0.043 (0.088) − 0.125⁎⁎⁎ (0.036) 0.890 (0.621) − 0.001 (0.006) − 0.035 (0.092) 18882 1683 − 3129.39

All specifications include a full set of destination country dummies, 4 5-year period dummies and the interactions between these two sets of dummies. Standard errors in parentheses. ⁎Significant at 10%; ⁎⁎significant at 5%; ⁎⁎⁎significant at 1%. a Source: OECD Economic Outlook. Values in percentage. b Annual compensation per employee in the private sector. Source: OECD Economic Outlook. All values are expressed in PPP/ECU. c Source: Comparative Welfare State Dataset (2004) by Evelyne Huber, Charles Ragin, John D. Stephens, David Brady, and Jason Beckfield based on various OECD data sources.

able from zero. Unskilled workers, in fact, are those who are most likely to need welfare assistance. Furthermore, in column 3 we expand our set of labour market controls. For the most recent years the OECD produces official labour market statistics, such as the unemployment rate and the participation rate, disaggregated by age and gender.10 These are the data that we use in column 3 of Table 3, where we associate to each individual migrant the unemployment and participation rate corresponding to her age–gender cell at the time of arrival in each destination country. Table A2 in the Appendix describes the time coverage of the restricted sample that we use for this exercise. We also obtained wages by major industrial sectors (agriculture, manufacturing and services) from the ILO. However, these three measures are extremely correlated, especially since we are using cross-country variation, and including all of them in the specification prevents convergence of the optimization procedure. Hence, the results in column 3 still condition only on the average wage level but all results hold also when this is replaced with any of the three sectoral wages.11 Due to the much smaller set of data that can be used for this exercise, we were not able to reach convergence with the full set of country and time period interactions. Hence, in column 3 we only include country and time period dummies and the results are very similar to the previous models. In particular, unskilled workers now appear to be significantly more attracted by the generosity of welfare in the destination country.

10 Age is broken down into the following groups: 15–24, 25–34, 35–44, 45–54, 55–64 ^ ^ ^ ^ ^ and 65+. 11 Unfortunately, cross-country comparable data on wages disaggregated by age and/ ^ or gender are not currently available, especially if one needs a relatively long time series.

360

G. De Giorgi, M. Pellizzari / Labour Economics 16 (2009) 353–363

Fig. 2. Welfare benefits and labour market conditions.

In the last two columns of Table 3 we relax another important assumption that we have maintained so far. The conditional logit model of Eq. (3) assumes that the random utility components ηid are uncorrelated both across individuals and across choices. In column 4 we relax this second assumption and estimate a nested logit model where migrants are assumed to first choose between groups of countries and then choose one specific destination within the group. By imposing this slightly more complex structure on the decision process we can allow the random utility terms to be correlated across groups while the independence assumption is maintained only within each group. For this specification we classify countries into 4 groups: Nordic countries include Denmark, Finland and Sweden, Anglo-Saxon countries are Ireland and the United Kingdom, Mediterranean countries are Greece, Italy, Portugal and Spain and, finally, Continental countries are Austria, Belgium, Germany, France and the Netherlands. This grouping is often used to categorise European countries along some general features of their welfare systems. Very broadly, Nordic countries are characterised by universal and generous benefits, often financed by general taxation. The Anglo-Saxon welfare model is more explicitly directed at encouraging individuals' selfsufficiency so, for example, most benefits are subject to means testing and require an availability to work. In Continental countries stricter labour market regulations are coupled with relatively more insurance-based benefit schemes, somewhat based on the idea of primarily protecting the main breadwinner in the household. Finally, Mediterranean countries share many features of the Continental group but are often characterised by a predominant role of pensions in the distribution of welfare expenditure and by an even more pronounced reliance on the family as a provider of social insurance (Esping-Andersen, 1990). We apply the nested logit specification to both the full set of data using only aggregate labour market data (column 4) and to the restricted set that allows us to adopt the more complete labour market indicators (column 5). The results confirm our previous findings. Importantly, the more sophisticated nested logit now converges when we use the restricted set of data while still allowing for the full interaction of country and time period dummies.

5. Welfare migration and asymmetric unemployment shocks: a simple simulation In this section we go back to the results of Tables 3 and perform a very simple simulation of the effects of an asymmetric shock to unemployment. This exercise is carried out in a very simplified partial equilibrium setting where wages are assumed to be rigid to changes in unemployment. This assumption implies that our results are likely to be valid primarily in the short term, when wages cannot adjust freely due, for example, to the presence of unions that negotiate workers' pay at discrete time intervals. The only purpose of the simulation that we present in this section is to highlight the policy relevance of our results that, although quantitatively similar to previous findings for the United States, are still large enough to bear important implications for countries with very limited geographical mobility of native workers. Suppose that there are only two possible destination countries, A and B, and that they are initially identical, with unemployment equal to 6% and welfare benefits of 5000 euros per year (approximately the averages across the 14 countries used for the estimation). Migration flows will, then, be evenly distributed across the two countries, with the reference person migrating to any of them with a probability of 50%. This is the benchmark situation described at the top of Table 7. Suppose now – panel 2 in Table 7 – that the two countries are hit by an asymmetric shock to unemployment, which goes up to 7% in country A and down to 5% in country B. The estimates of Table 3 predicts that the asymmetric shock to unemployment will generate a reduction of about 3.7 percentage points in the flow of migrants to country A and a symmetric increase for country B.12 This change in migration flows would, then, help to absorb the shock even if no native worker would move from A to B. What would happen if benefits changed too? Panel 3 computes the effect on the migration probabilities of a 15% increase in country A's benefits associated with an identical and simultaneous reduction in country B. The results indicate that the effect would be very similar that of the change in unemployment simulated in panel 2. In fact, 12

The simulated effects are based on the estimates of the model in Table 3, column 4.

G. De Giorgi, M. Pellizzari / Labour Economics 16 (2009) 353–363 Table 6 Specification checks. Conditional logit [1]

[2]

Table 8 Simulated effects of increased dispersion in unemployment and welfare generosity — Unskilled.

Nested logit [3]

[4]

[5]

−0.376⁎⁎ − 0.310⁎⁎⁎ − 0.131⁎ − 0.184 (0.158) (0.107) (0.076) (0.133) 2.938⁎⁎⁎ 3.621⁎⁎⁎ 0.898⁎ 4.569⁎⁎⁎ (0.851) (0.747) (0.495) (1.197) 0.739⁎⁎⁎ 0.411⁎⁎⁎ 0.257⁎⁎⁎ 0.964⁎⁎⁎ (0.212) (0.128) (0.085) (0.210) – – 0.275⁎ – (0.166)

− 0.172⁎⁎ (0.084) 6.770⁎⁎⁎ (1.465) 0.486⁎⁎ (0.196) 0.760⁎⁎⁎ (0.284)

Interaction terms: individual characteristics (at the time of arrival) with… (Age at arrival 0.009⁎⁎ 0.010⁎⁎ 0.004 0.006 − 25) ⁎ [lURATE] (0.005) (0.004) (0.005) (0.005) Female ⁎ [lURATE] −0.157⁎⁎ −0.075 0.073 − 0.055 (0.077) (0.076) (0.081) (0.088) (Age at arrival −0.067⁎⁎⁎ −0.072⁎⁎⁎ − 0.067⁎⁎⁎ − 0.100⁎⁎⁎ − 25) ⁎ [lWAGE] (0.016) (0.017) (0.016) (0.026) 0.420 0.226 0.146 0.261 Female ⁎ [lWAGE] (0.293) (0.297) (0.287) (0.455) (Age at arrival 0.002 0.001 0.002 0.006 − 25) ⁎ [B] (0.004) (0.004) (0.004) (0.006) Female ⁎ [B] − 0.168⁎⁎⁎ − 0.178⁎⁎⁎ − 0.189⁎⁎⁎ −0.195⁎⁎ (0.063) (0.064) (0.070) (0.099)

0.001 (0.005) − 0.028 (0.081) −0.095⁎⁎⁎ (0.025) 0.460 (0.423) 0.008 (0.006) − 0.139 (0.105)

(log) unemployment rate [lURATE] (log) wage [lWAGE] (log) welfare benefit [B] (log) participation rate

Interaction terms with skill indicator (at the time of interview)a: Unskilled ⁎ [lURATE] – −0.248⁎⁎⁎ − 0.345⁎⁎⁎ – (0.078) (0.079) Unskilled ⁎ [lWAGE] – 1.762⁎⁎⁎ 1.575⁎⁎⁎ – (0.316) (0.307) Unskilled ⁎ [B] – 0.029 0.143⁎⁎ – (0.065) (0.070) Destination country Yes Yes Yes Yes fixed effects Time dummies Yes Yes Yes Yes (3 years) (5 years) (5 years) (5 years) Destination ⁎ time Yes Yes No Yes dummies Observations 42058 42058 25327 42058 Individuals 3038 3038 2429 3038 Log-likelihood −6469.86 −6502.54 − 4541.11 − 6485.87

361

− 0.155⁎ (0.090) − 0.932⁎⁎⁎ (0.349) 0.286⁎⁎⁎ (0.079) Yes Yes (5 years) Yes 25327 2429 − 4357.74

See text for details. Standard errors in parentheses. ⁎Significant at 10%; ⁎⁎significant at 5%; ⁎⁎⁎significant at 1%. a Unskilled = less than secondary education at the time of the first interview.

when one computes the joint effect (panel 4) of these two asymmetric changes – to unemployment and benefits – one finds that changing benefits by 15% (a rather feasible reform) would completely offsets

Migration probability

Absolute change

1. Benchmark Country A 0.500 Country B 0.500

Unemployment rate

Change (%)

Annual welfare benefits

Change (%)

– –

6.0 6.0

– –

5000 5000

0.17 − 0.17

5000 5000

0.00 0.00

0.00 0.00

5750 4250

0.15 − 0.15

2. Effect of increased dispersion in unemployment Country A 0.460 − 0.040 7.0 Country B 0.540 0.040 5.0 3. Effect of increased dispersion in welfare generosity Country A 0.530 0.030 6.0 Country B 0.470 − 0.030 6.0

– –

4. Cumulative effect of increased dispersion in unemployment and welfare generosity Country A 0.490 − 0.010 7.0 0.17 5750 0.15 Country B 0.510 0.010 5.0 − 0.17 4250 − 0.15 All effects are computed for the unskilled reference person (male, aged 25, arrived in 1990, with less than secondary education).

the induced variations in migration flows generated by a sizable, but not unreasonable, unemployment shock. Given the focus on unskilled migrants that characterises most of the previous literature, in Table 8 we repeat the simulation considering as a reference person someone with less than secondary education. In this case, we compute all the effects on the basis of the estimates of the model in Table 6, column 3 and consequently also on the disaggregated labour market data, that are available for a shorter time span. The results are qualitatively very similar to those in Table 7, although in this case the welfare magnet effect does not offset completely the shift in migration due to the asymmetric shock to unemployment (of the unskilled). This is due to the fact that unskilled workers are more responsive to both labour market conditions and welfare generosity, as it emerges clearly from the estimates in Table 6. Overall, the simulations in Tables 7 and 8 deliver an important message. It is true that welfare generosity is a lot less important than labour market conditions in determining the location decisions of migrants, nevertheless its effect is large enough to dramatically reduce, and possibly completely offset, the potential benefits of migration in terms of labour reallocation, especially in countries with traditionally low geographical mobility. In fact, while the direct effect of welfare benefits on migration choices is small, the cross-effects are sometimes surprisingly high (see Table A1 in the Appendix). 6. Conclusions

Table 7 Simulated effects of increased dispersion in unemployment and welfare. Migration probability 1. Benchmark Country A 0.500 Country B 0.500

Absolute change

Unemployment rate

Change (%)

Annual welfare benefits

Change (%)

– –

6.0 6.0

– –

5000 5000

– –

0.17 − 0.17

5000 5000

0.00 0.00

0.00 0.00

5750 4250

0.15 − 0.15

2. Effect of increased dispersion in unemployment Country A 0.463 −0.037 7.0 Country B 0.537 0.037 5.0 3. Effect of increased dispersion in welfare generosity Country A 0.536 0.036 6.0 Country B 0.464 −0.036 6.0

4. Cumulative effect of increased dispersion in unemployment and welfare generosity Country A 0.499 − 0.001 7.0 0.17 5750 0.15 Country B 0.501 0.001 5.0 − 0.17 4250 − 0.15 All effects are computed for the reference person (male, aged 25, arrived in 1990).

The results of this paper suggest that the generosity of the welfare state may act as a migration magnet across the countries of the European Union. This finding supports the concerns expressed by many observers of a potential threat to the welfare state posed by the enlargement of the Union. On the other hand, however, the same estimates indicate that the size of these welfare magnets is relatively low, compared to the role of labour market conditions, such as the unemployment rate and the level of wages. The issue, then, is not whether migrants will flood into countries with generous welfare benefits, but to what extent the variation in the welfare institutions across the countries of the Union will generate distortions in the flows of migration. The empirical analysis conducted in the last part of this paper shows that these distortions can be large enough to reduce, and possibly completely offset, the potential benefits of acquiring a more mobile labour force. This is, in fact, one of the most important advantages that increased migration can offer to the countries of the pre-enlargement Union, an area where native workers traditionally move a lot less than, for example, North-Americans.

362

G. De Giorgi, M. Pellizzari / Labour Economics 16 (2009) 353–363

Appendix A

Table A1 Cross-effects of one-standard deviation increases for the reference persona. BEL

DEN

FIN

FRA

GER

GRC

IRL

ITA

NLD

PRT

ESP

SWE

UKG

Unemployment rate AUT − 2.780 BEL 0.158 DEN 0.199 FIN 0.258 FRA 0.256 GER 0.681 GRC 0.281 IRL 0.044 ITA 0.098 NLD 0.123 PRT 0.112 ESP 0.165 SWE 0.162 UKG 0.243

AUT

0.109 −0.813 0.053 0.069 0.069 0.183 0.075 0.012 0.026 0.033 0.030 0.044 0.044 0.065

0.128 0.050 −0.937 0.081 0.080 0.214 0.088 0.014 0.031 0.039 0.035 0.052 0.051 0.076

0.314 0.122 0.153 −2.256 0.197 0.524 0.216 0.034 0.076 0.095 0.086 0.127 0.125 0.187

0.137 0.053 0.067 0.087 − 0.986 0.229 0.094 0.015 0.033 0.041 0.038 0.056 0.055 0.082

0.638 0.248 0.312 0.404 0.401 − 3.925 0.440 0.069 0.154 0.192 0.175 0.258 0.254 0.381

0.183 0.071 0.090 0.116 0.115 0.306 − 1.307 0.020 0.044 0.055 0.050 0.074 0.073 0.109

0.017 0.007 0.008 0.011 0.011 0.029 0.012 − 0.132 0.004 0.005 0.005 0.007 0.007 0.010

0.051 0.020 0.025 0.032 0.032 0.086 0.035 0.006 −0.388 0.015 0.014 0.021 0.020 0.031

0.092 0.036 0.045 0.058 0.058 0.153 0.063 0.010 0.022 − 0.690 0.025 0.037 0.037 0.055

0.098 0.038 0.048 0.062 0.062 0.164 0.068 0.011 0.024 0.030 −0.740 0.040 0.039 0.059

0.070 0.027 0.034 0.044 0.044 0.116 0.048 0.008 0.017 0.021 0.019 − 0.516 0.028 0.041

0.305 0.118 0.149 0.193 0.192 0.510 0.210 0.033 0.073 0.092 0.084 0.124 − 2.268 0.182

0.184 0.071 0.090 0.116 0.116 0.307 0.127 0.020 0.044 0.055 0.050 0.074 0.073 − 1.328

Average AUT BEL DEN FIN FRA GER GRC IRL ITA NLD PRT ESP SWE UKG

wages 8.395 −0.477 −0.602 − 0.779 −0.774 − 2.056 − 0.848 − 0.134 − 0.296 − 0.371 − 0.337 − 0.498 − 0.490 − 0.734

− 0.423 3.142 − 0.207 − 0.267 −0.266 −0.706 −0.291 −0.046 −0.102 − 0.127 − 0.116 − 0.171 − 0.168 − 0.252

−0.566 − 0.220 4.153 −0.358 − 0.356 − 0.946 − 0.390 − 0.061 − 0.136 −0.171 −0.155 −0.229 −0.225 −0.338

−0.849 − 0.330 −0.415 6.108 −0.534 − 1.420 − 0.585 − 0.092 − 0.204 − 0.256 − 0.233 − 0.344 − 0.338 − 0.507

−0.752 − 0.292 − 0.368 − 0.476 5.413 − 1.258 − 0.519 − 0.082 − 0.181 − 0.227 − 0.206 − 0.305 − 0.299 − 0.449

− 1.855 −0.720 − 0.907 −1.174 −1.167 11.413 −1.279 −0.201 −0.446 −0.560 − 0.508 − 0.751 − 0.738 − 1.107

−1.060 − 0.411 − 0.518 −0.671 − 0.667 − 1.772 7.562 − 0.115 − 0.255 −0.320 − 0.290 − 0.429 − 0.422 −0.633

− 0.118 − 0.046 − 0.058 − 0.075 − 0.074 − 0.197 − 0.081 0.912 − 0.028 − 0.036 − 0.032 − 0.048 − 0.047 − 0.071

− 0.292 −0.113 − 0.143 − 0.184 − 0.183 − 0.487 − 0.201 − 0.032 2.211 − 0.088 − 0.080 − 0.118 − 0.116 −0.174

− 0.300 − 0.116 − 0.147 − 0.190 − 0.188 − 0.501 − 0.207 − 0.033 − 0.072 2.254 − 0.082 − 0.121 − 0.119 − 0.179

− 0.576 − 0.224 − 0.282 − 0.365 − 0.362 − 0.963 − 0.397 − 0.063 − 0.139 − 0.174 4.350 − 0.233 − 0.229 − 0.344

− 0.597 − 0.232 − 0.292 −0.378 − 0.375 − 0.998 − 0.411 − 0.065 −0.144 − 0.180 − 0.164 4.429 − 0.238 − 0.356

− 0.568 − 0.220 − 0.278 − 0.359 − 0.357 − 0.949 − 0.391 − 0.062 − 0.137 − 0.171 − 0.156 − 0.230 4.218 −0.339

− 0.679 − 0.263 − 0.332 − 0.430 − 0.427 − 1.135 − 0.468 − 0.074 − 0.163 − 0.205 − 0.186 − 0.275 − 0.270 4.906

Welfare benefits AUT 2.909 BEL −0.165 DEN − 0.208 FIN − 0.270 FRA − 0.268 GER − 0.713 GRC − 0.294 IRL − 0.046 ITA − 0.103 NLD − 0.129 PRT − 0.117 ESP − 0.173 SWE − 0.170 UKG − 0.254

−0.094 0.700 − 0.046 − 0.060 − 0.059 − 0.157 − 0.065 − 0.010 − 0.023 − 0.028 − 0.026 − 0.038 − 0.037 − 0.056

− 0.090 − 0.035 0.660 − 0.057 − 0.057 − 0.150 − 0.062 − 0.010 − 0.022 − 0.027 − 0.025 − 0.036 − 0.036 − 0.054

− 0.296 −0.115 − 0.145 2.126 − 0.186 − 0.494 − 0.204 − 0.032 − 0.071 − 0.089 − 0.081 − 0.120 − 0.118 − 0.176

−0.232 − 0.090 − 0.113 − 0.147 1.666 − 0.387 − 0.160 − 0.025 − 0.056 − 0.070 − 0.063 − 0.094 − 0.092 − 0.138

− 0.619 − 0.240 − 0.303 − 0.392 − 0.389 3.809 − 0.427 − 0.067 − 0.149 − 0.187 − 0.170 − 0.251 − 0.246 − 0.369

−0.670 −0.260 − 0.328 − 0.424 − 0.421 − 1.120 4.780 − 0.073 − 0.161 − 0.202 − 0.184 − 0.271 − 0.267 − 0.400

− 0.045 − 0.018 − 0.022 − 0.029 − 0.029 − 0.076 − 0.031 0.351 − 0.011 − 0.014 − 0.012 − 0.018 − 0.018 − 0.027

−0.406 − 0.158 − 0.199 − 0.257 − 0.256 − 0.679 − 0.280 − 0.044 3.081 − 0.123 − 0.111 − 0.164 − 0.162 − 0.242

− 0.063 − 0.024 − 0.031 − 0.040 − 0.039 − 0.105 − 0.043 − 0.007 − 0.015 0.471 − 0.017 − 0.025 − 0.025 − 0.037

− 0.134 − 0.052 − 0.066 − 0.085 − 0.084 − 0.224 − 0.092 − 0.015 − 0.032 − 0.040 1.013 − 0.054 − 0.053 − 0.080

− 0.161 −0.062 − 0.079 − 0.102 − 0.101 − 0.268 − 0.111 − 0.017 − 0.039 − 0.048 − 0.044 1.191 − 0.064 − 0.096

−0.148 − 0.057 − 0.072 − 0.094 − 0.093 − 0.247 − 0.102 − 0.016 − 0.036 − 0.045 − 0.041 − 0.060 1.098 − 0.088

− 0.268 − 0.104 − 0.131 − 0.170 − 0.169 − 0.448 − 0.185 − 0.029 − 0.064 − 0.081 − 0.073 − 0.108 − 0.107 1.937

All effects are multiplied by 100. Male, aged 25, arrived in 1990.

a

Table A2 Country and time coverage of disaggregated labour market data. Destination country Austria Belgium Denmark Finland France Germany Greece Ireland Italy Netherlands Portugal Spain Sweden United Kingdom

Available years First

Last

1994 1983 1983 1970 1970 1970 1983 1971 1970 1971 1974 1972 1970 1984

1994 1994 1994 1994 1994 1994 1994 1994 1994 1994 1994 1994 1994 1994

G. De Giorgi, M. Pellizzari / Labour Economics 16 (2009) 353–363

References Alvarez-Plata, P., Brücker, H., Siliverstovs, B., 2003. Potential Migration from Central and Eastern Europe into the EU-15 — An Update. Report for European Commission. DG Employment and Social Affairs, Brussels. Blank, R.M., 1988. qThe effect of welfare and wage levels on the location decisions of female-headed households.q Journal of Urban Economics 24 (2), 186–211. Boeri, T., Hanson, G.H., McCormick, B., 2002. Immigration Policy and the Welfare System. Oxford University Press. Boeri, T., Brücker, H., 2001. The Impact of Eastern Enlargement on Employment and Labour Markets in the EU Member States. Report for European Commission. DG Employment and Social Affairs, Brussels. Borjas, G., 1999. Immigration and welfare magnets. Journal of Labour Economics 17 (4), 607–637. Dustman, C., Casanova, M., Preston, I., Fertig, M., Schmidt, C.M., 2003. The Impact of EU Enlargement on Migration Flows. Report for the Immigration and Nationality Directorate of the UK Home Office. Esping-Adersen, G., 1990. Three Worlds of Welfare Capitalism. Princeton University Press. Gelbach, J.B., 2004. Migration, the life cycle, and state benefits: how low is the bottom? Journal of Political Economy 112 (5), 1091–1130. Gramlich, E.M., Laren, D.S., 1984. Migration and income redistribution responsibilities. Journal of Human Resources 19 (4), 489–511.

363

Enchautegui, M., 1997. Welfare payments and other economic determinants of female migration. Journal of Labor Economics 15 (3), 529–554. Kvist, J., 2004. Does EU enlargement start a race to the bottom? Strategic interaction among EU member states in social policy. Journal of European Social Policy 14 (3), 301–318. Levine, P.B., Zimmerman, D.J., 1999. An empirical analysis of the welfare magnet debate using the NLSY. Journal of Population Economics 12 (3), 391–409. McKinnish, T., 2007. Welfare-induced migration at state borders: new evidence from micro-data. Journal of Public Economics 91, 437–450. McKinnish, T., 2005. qImporting the poor: welfare magnetism and cross-border migration.q. Journal of Human Resources 40 (1), 57–76. Meyer, B.D., 2000. Do the poor move to receive higher welfare benefits? Working Paper No. 58. Northwestern University, Joint Center for Poverty Research Working Paper. OECD, 2001. Trends in International Migration. Paris. Sinn, H., 2001. EU enlargement, migration and lessons from German unification. Discussion Paper no. 2174. CEPR. Sinn, H., 2004a. Europe Faces a Rise in Welfare Migration. Financial Times, p. 13 (July 13). Sinn, H., 2004b. EU enlargement, migration and the new constitution. Working Paper No. 1367. CESifo. Walker, J.R., 1994. Migration among low-income households: helping the witch doctors reach consensus. Discussion paper 94–1031, Institute for Research on Poverty. University of Wisconsin-Madison.