96200  i The role of transnational family distribution in shaping remittance flows greenback 2.0 Working paper N. 2 Laura Bartolini1 APRIL 2015 Suggested citation: Bartolini, Laura (2015), “The role of transnational family distribution in shaping remittance flows,” The World Bank, Greenback Working Paper n. 2. This paper is part of the Greenback 2.0—Remittances champion cities project, funded by the World Bank, and was conceived by the author in strict collaboration with Ferruccio Pastore (FIERI), Eleonora Castagnone (FIERI) and Marco Nicolì (Payment Systems Development Group, Finance and Markets, World Bank). The views expressed in this working paper are those of the author and do not necessarily represent those of the World Bank Group. 1 Global Governance Programme, European University Institute (Florence, Italy) and FIERI (Turin, Italy)—email laura.bartolini@eui.eu  iii ABSTRACT M igration experiences are often associated with some sort of transnational economic activity which connects the past and the present of migrants abroad, and specifi- cally with remittances. Motivations to send money at origin have been deeply inves- tigated at the micro as well as at the macro level, as remittances can affect recipient households’ well-being, investment and consumption levels in the receiving countries and play an insurance role against external shocks. This paper contributes to the literature on migrants’ remittances providing evidence for migrants from Morocco, Peru and Romania, three traditional labor-exporting coun- tries with a medium level of economic development, from three different geographical areas and with different migration patterns to Italy. Exploiting a relatively rich, albeit small-scale, dataset we analyze the spatial distribution of migrants’ nuclear families and households and we build three different migratory profiles—Loners, Pioneers and Followers—characterized by the timing and sequence of the migration event with respect to the rest of the nuclear family. Then we test a negative binomial model to describe the variation in the variable “remittances amount”. Beyond cross-country varia- tions and after controlling for the most commonly used individual demographic and economic characteristics, our analysis consistently clusters migrants according to their family and household structure in Italy and abroad to explain the total amount of remit- tances sent to the origin country.  v Contents Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Analyzing determinants to remit.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 The Greenback 2.0 data and some descriptive statistics. . . . . . . . . . . . . . . . 6 Data strengths and limitations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Three different migratory profiles. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 The empirical model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 The dependent count variable: annual volume per migrant.. . . . . . . . . . . . . . . . . . . . . . . . . . 11 A model for count, overdispersed data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 Regression results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 List of tables and figures vii List of Tables and Figures Table 1: The Greenback 2.0 sample. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Table 2: Average amount sent per transaction and per year. . . . . . . . . . . . . . . . . . . . . . 11 Table 3: Negative binomial model (NB2) regression results.. . . . . . . . . . . . . . . . . . . . . . 13 Table 4: Chi2 test on migration profile (base: Loner).. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 Figure 1: Spatial distribution of remittance determinants.. . . . . . . . . . . . . . . . . . . . . . . . . 4 Figure 2: Three migration profiles by country of origin. . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Figure 3: Spatial distribution of all relatives who live in Italy, in the country of origin and in third countries, by migratory profile.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 Figure 4: Spatial distribution of children only, who live in Italy, in the country of origin and in third countries, by migratory profile.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 Figure 5: Main sample characteristics*, by migratory profiles and total. . . . . . . . . 10 Figure 6: Annual remittances in euro, total and by country of origin.. . . . . . . . . . . . 12 Figure 7: Predictive amount of remittances (calculated on model 3). . . . . . . . . . . . 15 Introduction M igration experiences are often associated with some sort of transnational economic activity which connects the past and the pres- Azam and Gubert 2006; Erdal 2012). Moreover, for migrants’ destination countries the analysis of remittance outflows helps understanding the level ent of migrants abroad. Migrants’ long-distance and depth of migrants’ labor market integration at economic relations with their homelands are the destination and of their connection with the origin subject of an extensive, multidisciplinary inquiry households, which directly influence the amount (Guarnizo 2003), often focused on monetary and regularity of flows. remittances as the most visible sign of transna- tional engagement and virtually the only one This paper contributes to the literature on which can be traced both at the micro and macro- migrants’ remittances in many respects. While a level. Quantitative information about monetary consistent part of existing studies concentrates on remittances comes in aggregate records from migrants’ transfers to developing countries (Sin- financial institutions and from sample surveys ning 2007, 3), we provide evidence for migrants either on the sending-side, the receiving-side or from three traditional labor-exporting countries both (matched-sample surveys). While official with a medium level of economic and human figures allow for cross-country and historical com- development (UNDP 2014), from three different parisons, sample surveys offer deeper information geographical areas and with different migration and accuracy and are in principle able to grasp patterns to Italy. Exploiting a relatively rich, albeit also informal flows not recorded at the aggregate small-scale, dataset on migrants from Morocco, level (Brown et al. 2014). Romania and Peru in the City of Turin—one of the biggest cities in Northern Italy with a long tradi- Many important issues are related to the study tion of internal and international immigration— of remittance flows and of migrants’ behavior in we provide an empirical analysis for the under- sending money to their households and countries standing of the key determinants of migrants’ of origin. International development institutions, remittances. Although the data come from a academics and policy makers have progressively sender-side survey which took as observational integrated migration and remittances into the unit first-generation migrants, their geographi- development discourses, policies, and programs cal perspective allows for the analysis of the (Ratha 2007; Ratha and Mohapatra 2007; Ratha spatial distribution of migrants’ nuclear families et al. 2014). Also because of their magnitude and and households, with information on the poten- importance at the aggregate level, motivations tial and actual remittance recipients and their behind remittances have been deeply investigated characteristics. at the micro-level, as well as the extent to which remittances affect recipient households’ well- In particular our research question concerns the being, investment and consumption levels in possibility of describing a model for the variable the receiving countries and play an insurance “remittances amount” which takes into account role against external shocks. The study of remit- the migration history of migrants and their nuclear tances helps shed some light on intra-household families. How do the household and family struc- resource allocation, disentangling preferences and ture and its spatial distribution in the country of behaviours of migrants and individual household origin, at destination and eventually in third coun- members that receive the money (Posel 2001; tries, influence the remitting behavior of observed  1 2 GREENBACK 2.0 survey 2015 first generation migrants in Italy? Indeed, beyond The paper proceeds as follows. The next para- cross-country variations and after controlling for graph briefly reviews the existing literature on the most commonly used individual demographic remittance determinants. Section 2 presents the and economic characteristics, our analysis intends Greenback 2.0 data and provides descriptive to explicitly test the specific role of family struc- evidence of three different migratory profiles tures and networks across borders in determining associated with history of migration and family variation of the annual remittances amount sent structure. Section 3 presents the empirical model to the origin country. and discusses the results. Section 4 concludes. 1 Analyzing Determinants TO REMIT T here is an extensive literature on the motives behind migrants’ remittances and on the determinants of such international economic instead rarely focus primarily on remittances as an autonomous study object, but offer deeper insights on the complex relationships between transfers. The first and most cited article on the the migrant and the origin household (Erdal 2012; topic is the one by Lucas and Stark (1985) which Carling 2014). set the framework for the development for the so- called ‘new economics of labor migration’ (NELM). As testified by the increasing literature on the Starting from the premise that decisions about topic, there is a great variation in the nature and remittances are connected with those on migra- logic of these economic transnational transac- tion and that this decision-making process also tions. In testing the responsiveness of remitting involves the household of origin, they designed behavior to changes in the migrants’ and/or a taxonomy of motives to remit which goes from recipients’ conditions in terms of income, wealth true altruism to a set of pure self-interest motives. or well-being, variations due to different contexts The combination of such motives within each as well as due to different conditions within a single household is also dependent upon its own single setting should be taken into consideration. structure, and is what makes the arrangement As pointed out by Carling, “neither economics nor among household members self-reinforcing ethnography has engaged fully with the combi- (Carling 2008). nation of complexity and variation in remittance transactions” (Carling 2014, 219). Moreover, since A huge number of empirical works have ana- it is always difficult to control for all the different, lyzed the determinants of remittances starting intertwined components of remittances which from the framework outlined by Lucas and Stark. find their actual balance in empirically determined Some authors also reviewed the existing literature conditions and contexts, a general explanation of on remittances in order to systematize theories what causes remittance flows is hardly achievable and empirical evidence collected so far by social (Carling 2008). Hence, from time to time, altruist scientists and to get to a general explanation and self-interest motives are modelled in differ- of what causes remittance flows (Rapoport and ent ways, making sense of the incredible variation Docquier 2006; Hagen-Zanker and Siegel 2007; of micro and macro motivations and behaviors Carling 2008). A recent paper from Carling, which derive from personal attitudes and ability, inscribes remittance transfers into more complex families’ and households’ structures and needs, and composite transactions which include at migration context and historically determined the same time material, emotional and relational factors. elements (Carling 2014), also discussing the dif- ferent approaches in the existing economic and In the discussion on the relative importance of dif- ethnographic literature. Economists tend to test ferent motives and determinants of remittances, empirical models for disentangling the determi- two important aspects influencing the variation of nants of remittances, which can all be attributed flows are often disregarded (Carling 2008). Firstly, to one of the following main motives: altruism, migration itself has to be taken into consideration, insurance, investment, and repayments (Lucas and as migration patterns from and to specific regions Stark 1985; Rapoport and Docquier 2006; Cox, and localities define multiple demographic Eser, and Jimenez 1998). Ethnographic studies dynamics which are not often explicitly integrated  3 4 GREENBACK 2.0 survey 2015 in empirical analysis. Secondly, some key variables moments in their migration history. Although have to be identified as preconditions for remit- migrants are generally primarily senders, these tances, necessary in order to distinguish between transfers are often associated with various forms the capacity and the desire to remit (Carling and of reciprocity (Åkesson 2011; Mazzucato 2011; Thai Hoelscher 2013). 2014). If we draw a stylized picture with at least three individuals—the migrant and two relatives While most of the studies only focus on one of or household members—placed in three differ- the (at least) two ends of the remittance corridor, ent locations—migrant’s country of origin and of treating separately what happens in the migrant’s destination and a third country—remittance flows context of origin and of destination depending can be observed from and to the three localities, on the availability and depth of the data, explana- in multiple, bilateral relationships which change tions of the variation of remittance flows at the across time and due to a variety of individual micro level have to be found in many different and household characteristics and of local and locations at the same time. Migrants are often national contexts. Figure 1 presents our adapta- portrayed by default as remittance senders to tion to the scheme proposed by Carling (2008) to their origin households. Indeed, the spatial distri- properly locate the determinants of remittances. bution of migrants and their relatives and house- hold members has to be analyzed in the country Beyond the theoretical mechanisms presented of origin, in the country of destination, and even- in the picture, a further, non-trivial aspect to tually in third countries where other relatives are be considered for a sound empirical analysis of settled. Moreover, the remittance relationship migrant’s remittance capacity is the ‘technical’ between a migrant and the origin household and choice of the model to be adopted. The exist- family members—wherever they live—is often ing empirical literature has proposed different presented as mono-directional: migrants send econometric models and methodologies. Earlier money back to their origin country and eventu- papers used more often Ordinary Least Squares ally to other family members elsewhere. The (OLS) regression not only for the size of remit- reality indeed is far more complex and migrants tances but also for the decision to remit, lead- can behave as senders or receivers of mon- ing to biased estimations. More recent papers etary resources in correspondence of different instead have adopted a greater variety of models Figure 1: Spatial distribution of remittance determinants Country-to-country remittance CoD environment bilateral corridor CoO environment (national and local context) (national and local context) Potential RECEIVERS Potential SENDER and/or Senders: and/or Receivers: characteristics of the HH migrant individual and of origin and of single HH characteristics individual receivers/senders Senders—Receivers bilateral relationship (including migration timing and history) Other family members abroad (potential remittance receivers and/or senders) TC environment Source: Author’s adaptation from (Carling 2008, 586). ANALYZING DETERMINANTS TO REMIT 5 and specifications in order to better identify the our data do not provide information on the fun- explanatory factors for remittances. In particular, damental determinants of the decision to remit, some use one-stage decision models where the as the Greenback 2.0 sample only includes remit- decisions on whether to remit and on the amount ting migrants (i.e. migrants having decided not to send are taken together. In other cases, a two- to remit were not included in the survey sample), stage (hurdle) approach portraits the decision in our empirical section we will still have to take to remit and the decision on the amount in two into consideration the peculiar features of our distinct, chronologically consequent models (see dependent variable (remittance amount) in order Hagen-Zanker and Siegel 2007; Carling 2008 for a to choose the most appropriate model for our review of most recent applied methods). Although empirical test. 2 The Greenback 2.0 data and some descriptive statistics O ur empirical data on migrant remittances are drawn from the Greenback 2.0 Survey (The World Bank 2014). The survey collected in-depth, adopted to capture also irregular migrants and to design a balanced sample.2 quantitative data on migrants residing in Turin, in The average profile of the final sample is reported the North of Italy, during the summer of 2013. The in Table 1 which presents the main general statis- aim was exploring migrants’ financial needs and tics on sex, arrival in Italy, education level, marital behaviors, with a focus on their remittance prac- status and type of occupation of the 480 inter- tices. The overall sample is composed by three viewed individuals. The overall sample is relatively equally large subsamples according to the citizen- gender-balanced (43 percent of the interviewed ship at birth of the interviewees: short-range EU are women), but gender differences become more migration (Romania), short-range non-EU migra- evident within each subsample: women represent tion (Morocco) and long-range migration (Peru). 61 percent of Romanian, 54 percent of Peruvian, These are the first three countries of origin per and only 14 percent of Moroccan interviewed number of residents in the city (almost 60 per- migrants, the latter less frequently complying cent of total migrant population) and per total with the survey criteria because of their low activ- amount of remittances outflows from the Province ity rate. With regard to formal qualifications and of Turin (Fondazione Moressa 2013; Banca d’Italia skills, Peruvians show the highest level of educa- 2014). At the same time, these three countries tion attained (24 percent of highly educated), differ for their geographical position, patterns while three quarters of all Romanians declared of socio-economic integration in Italy (in terms a medium level of education (high school) and of participation to the labor market by sex and almost half of the Moroccan subsample only has distribution in different economic sectors) and a low education level. Almost half of the overall migratory systems (in relation to the organization interviewed migrants are married. Moroccans have of the migratory chain within families). the higher level of single individuals (35 percent), while among Romanians and Peruvians there is The sample is composed by foreign-born individu- a higher incidence of separated or divorced indi- als residing in the City of Turin during the sum- viduals (respectively 18 and 16 percent). mer of 2013, including naturalized immigrants, between 18 and 64 years of age. To be included in Information on the type of job and sector of occu- the sample, the interviewed migrants had to com- pation3 has been re-codified to present the most ply with four criteria: 1) to have resided in Italy for significant, frequent occupation among those at least one year (with or without a regular resi- listed by migrants: around 41% of the interviewees dence status); 2) to live in the metropolitan area are workers in the construction sector or in manu- of Turin; 3) to have an income (broadly speaking, facturing, while jobs related to the domestic and from any type of occupation, including informal 2 No weights have been used, but a full coverage of aggregation centres. activities); and 4) to have sent remittances to For a detailed analysis of the sampling strategy, see A Methodological Note, his/her country of origin at least once since the in World Bank 2014, 40. 3 The original dataset provides information on the type of occupation and beginning of 2013. A ‘centre sampling technique’ sector of activity in accordance with the European NACE (Nomenclature of (Baio, Blangiardo, and Blangiardo 2011) was Economic Activities) classification. 6 THE GREENBACK 2.0 DATA AND SOME DESCRIPTIVE STATISTICS 7 Table 1: The Greenback 2.0 sample Morocco Peru Romania Total Freq. Col % Freq. Col % Freq. Col % Freq. Col % Sex Male 136 85.5 74 46.0 62 38.8 272 56.7 Female 23 14.5 87 54.0 98 61.3 208 43.3 Education Low 76 47.8 26 16.1 23 14.4 125 26.0   Medium 58 36.5 96 59.6 121 75.6 275 57.3 High 25 15.7 39 24.2 16 10.0 80 16.7 Marital status Partnership 88 55.3 99 61.5 101 63.1 288 60.0 Separated/Divorced 15 9.4 27 16.8 25 15.6 67 14.0 Single 56 35.2 35 21.7 34 21.3 125 26.0 Job type Worker (industry/ 76 47.8 37 23.0 57 35.6 170 41.7 construction) Domestic worker 15 9.4 86 53.4 49 30.6 150 36.8 Seller (street vendor, 29 18.2 3 1.9 6 3.8 38 9.3 salesperson, cashier) Nurse/care giver (OSS) 3 1.9 22 13.7 9 5.6 34 8.3 Cook & barman 12 7.6 3 1.9 16 10.0 31 7.6 Shop owner 3 1.9 2 1.2 8 5.0 13 3.2 Others 21 13.2 8 5.0 15 9.4 44 10.8 Italian citizenship   15 9.4 10 6.2 3 1.9 28 5.8 Total 159 100 161 100 160 100 480 100 health care sectors employ respectively 37% and Also, data provide us with the transnational family 8% of the total sample. structure and the household structure at destina- tion of respondents, which tells us who are the Data strengths and limitations potential senders and recipients of transnational monetary flows. The Greenback 2.0 survey only Measuring remittances through a small scale sur- recorded monetary transfers and does not allow vey presents many conceptual and methodologi- comparing them with in-kind transfer. Neverthe- cal issues that we had to take into account for the less, recording the presence of reverse remit- purposes of our analysis (Brown et al. 2014). The tances gives a sense of the existing reciprocity Greenback 2.0 Survey was designed to provide mechanisms between Italy and abroad. information about individuals and about transac- tions, asking interviewees to describe each flow of The survey is also extremely rich in terms of remittances they were sending to different recipi- contextual information about individuals, their ents. Each recorded flow is characterized in terms households at destination and their nuclear of amount, frequency, channels, and cost, while families, which helps inscribing the remittance recipients were defined as the individuals who behavior in a broader picture. Information on materially receive the money. As such, recipients transnational family formation, on the existence do not necessarily coincide with beneficiaries, but and timing of family reunification processes, on they represent those who control the money and the composition of the household at destination, who may use it to benefit a third person. This is helps understanding the number and type of often the case, for example, of children of minor relationships which might generate money flows age left in the country of origin, who are not (hence the number of potential senders and directly receiving the money, but can benefit from recipients). remittances sent to other relatives. 8 GREENBACK 2.0 survey 2015 Although collected data do not provide informa- We define as Loners those who do not have any tion on possible return intentions, they do fully member of the nuclear family living in Italy at the cover the remitter’s migration history with timing time of the interview. Hence, for the moment at and length of migration, family formation pro- least, they are the only one migrated in Italy. The cesses before and after migration, etc. Also, with second group is that of Pioneers: they have been regard to individual and household income and the first to move to Italy among their first grade wealth, we have full information on the situation relatives and they have been then followed by at at destination in terms of income, type of job and least one of them: like the Loners, Pioneers have specialization, presence of second income earners initiated the migration history of their family in in the household, economic trends over time, bank Italy, but they have been reached by other family and investment decisions (bank services, loans members. The last profile is that of Followers, a and investments in mortgages or other activities). category where we include those who migrated only after at least one of their family members Hence, although our analysis may suffer from data was already in Italy: they can be the last to be limitations and from the fact that few longitudi- arrived or there can be others to follow, but they nal information is available, the Greenback 2.0 do not initiate the migratory experience of their dataset provides enough information for drawing family to Italy. Figure 2 illustrates how these three a precise and in-depth picture over a relatively different types are present within each coun- numerous sample of people (480 interviews). try subsample. Consistently with the history of migration in Italy and in Turin in particular and Three different migratory with some specificities of these three nationali- ties (Cingolani and Ricucci 2013; Pastore, Salis, profiles and Villosio 2013), in comparison with the rest of To start with describing the collected evidence the sample, Moroccans are characterized by both beyond the possible cross-country comparison a wider presence of migrants alone in Italy—who among the three subsamples (Moroccans, Peru- have not yet formed a family or have decided not vians, and Romanians), we depict three different to undertake family reunification processes—and profiles of migrants according to their history of of Pioneers, who had the time to pursue family migration. Our data provide information on the reunification after settlement thanks to their long composition and demographic characteristics of presence in Italy. the nuclear family: age, sex, marital status, place of residence and year of migration of parents, The radar graphs below combine the cross- siblings, partner, and children of the interviewees. country disaggregation with the distinction of Combining this information, we can establish the the three migratory profiles in order to illustrate presence of family re-unification processes and the demographic and family characteristics as we can situate the interviewed migrant in a chron- ological sequence of migratory events. Hence we build three different migratory profiles character- Figure 2: Three migration profiles by country ized by the timing and sequence of the migration of origin event for the interviewed migrants with respect to the rest of his/her own nuclear family. The Romania 17.50 42.50 40.00 three profiles are by no means necessary steps towards a unique, unavoidable end, nor is there a chronological order among them: although we do Peru 21.12 38.51 40.37 not have information on intentions to return, each single migration history varies from the beginning and keeps being different at each phase of the Morocco 28.30 43.40 28.30 migrant life for decisions about where to settle and for how long, if and how to form a family, if 0% 20% 40% 60% 80% 100% and how to keep a connection with the household of origin. Loners Pioneers Followers THE GREENBACK 2.0 DATA AND SOME DESCRIPTIVE STATISTICS 9 well as the income and remittance patterns of the through migration or not), makes the remittance sampled migrants. burden for the migrant progressively weaker. Fol- lowers then show a higher number of relatives in The total number of relatives and their spatial dis- Italy than Pioneers and, of course, Loners (Fig- tribution between Italy, the country of origin, and ure 3). Also, differences among the three national eventually other third countries can help us asso- communities increase if we look more specifi- ciating different observed integration patterns cally at the number of children and their location and transnational behaviors in terms of remit- rather than at nuclear family members in general. tances to different stages in migrants’ life and in Figure 4 shows that pioneers are clearly those their migration history. It is reasonable to expect with more children on average and more chil- that migrants with more relatives still in the coun- dren in Italy, while loners generally have very few try of origin, and especially those with children children at home (or not at all). Hence, with the left behind, are those who sent more money, exception of Romanians who are more likely to be more frequently. As time passes, the distribution parents even when they are alone in Italy, being of family members between the country of origin a loner seems to be associated with not having and of destination may change, determining an yet started to form a family. From this descrip- alleviation of the family burden for the migrant tive evidence, one can expect that remittances abroad. Indeed, family reunification processes, as are particularly driven by the presence of children well as the decreasing number of dependent rela- and by their place of residence, rather than by the tives at origin (because parents might pass away location of family members in general. and siblings might become more independent, Figure 3: Spatial distribution of all relatives who live in Italy, in the country of origin and in third countries, by migratory profile Morocco rel_OR Peru Romania rel_OR rel_OR 6 6 6 5 5 5 4 4 4 3 3 3 2 2 2 1 1 1 0 0 0 rel_TC rel_ITA rel_TC rel_ITA rel_TC rel_ITA Loners Pioneers Followers Loners Pioneers Followers Loners Pioneers Followers Figure 4: Spatial distribution of children only, who live in Italy, in the country of origin and in third countries, by migratory profile Morocco Peru Romania rel_OR rel_OR rel_OR 1.5 1.5 1.5 1.2 1.2 1.2 0.9 0.9 0.9 0.6 0.6 0.6 0.3 0.3 0.3 0.0 0.0 0.0 rel_TC rel_ITA rel_TC rel_ITA rel_TC rel_ITA Loners Pioneers Followers Loners Pioneers Followers Loners Pioneers Followers 10 GREENBACK 2.0 survey 2015 Figure 5: Main sample characteristics,* by migratory profiles and total Female % Morocco Female % Peru 0.8 0.8 0.7 0.7 0.6 0.6 Rem/Inc 0.5 Age Rem/Inc 0.5 Age ratio 0.4 ratio 0.4 0.3 0.3 0.2 0.2 0.1 0.1 0.0 0.0 Remittances Age at Remittances Age at migration migration Loners Loners Pioneers Pioneers Income Length of Followers Income Length of Followers stay stay Female % Romania Female % Total 0.8 0.8 0.7 0.7 Rem/Inc 0.6 Rem/Inc 0.6 0.5 Age 0.5 Age ratio 0.4 ratio 0.4 0.3 0.3 0.2 0.2 0.1 0.1 0.0 0.0 Remittances Age at Remittances Age at migration migration Loners Loners Pioneers Pioneers Income Length of Followers Income Length of Followers stay stay *Continuous variables have been standardized between 0 and 1 in order to appreciate differences across profiles and countries rather than real measures. Beyond family composition and distribution, the has a short migration history in Italy and earns three typologies combined with the three coun- less than pioneers and followers. On the other tries of origin appear to be different also for what hand, being alone allows the migrant to remit concerns their demographic characteristics (sex, more both in absolute and relative terms (to the age, age at migration and length of stay) and their income) than those migrants who live in Italy with economic profile. Figure 5 illustrates that females other relatives. are more frequently followers in Morocco, while representing the 70 percent of Pioneers among To test the statistical strength of these observed Peruvians. Pioneers on average have the longer characteristics and to see to what extent the length of stay, especially within the Romanian family composition and distribution counts in subgroup where they also have higher earnings determining the total amount of remittances sent, and fewer remittances than followers and loners. we proceed with an empirical model in the next In general, the average lone migrant is younger, section. 3 The empirical model The dependent count variable: (Figure 6). Although in principle there is no upper limit for this variable, in practice values are empiri- annual volume per migrant cally linked to the availability of resources (mea- Differing from other types of international finan- sured by income) of remitters. cial flows to developing countries, remittances are usually sent at relatively high frequencies and A model for count, in small magnitudes (Yang 2011). Data from our sample show that the average amount of money overdispersed data sent per transaction is of €236. For the purpose Hence, we need to model a regression which of our analysis, we combined the magnitude and could explain our count variable y through a vec- frequency of each flow to estimate the average tor of explanatory variables x. The most straight- annual amount sent to the same recipient and forward approach is a linear model estimated by then the average amount sent by each migrant OLS of the form E( y | x) 5 x. But the  OLS esti- overall (see the Greenback 2.0 Survey for the mators will allow for the predicted values of y to original figures: The World Bank 2014). The Peru- be negative, while we would like to have E( y | x) vian subsample, with the highest share of monthly non-negative for all because y  0 by defini- transactions, shows the biggest total amount per tion. For strictly positive variables, the natural year (€2113), followed by Romanians (€1732), and log transformation is often used in order to test a Moroccans (€1594) (see Table 2). Average annual linear model of the type ln( y) 5 x 1 . Another values are in line with the most recent data pro- possibility is instead to fit a Poisson model of vided by the Bank of Italy at the national level the form y 5 exp(x 1 ). The Poisson regres- (Banca d’Italia 2014; Fondazione Moressa 2013). sion model has some interesting features and its assumptions may fit well with our data: The total amount of remittances sent during a year is a count variable, with no natural a priori • The distribution is discrete upper bound and the possibility that the outcome • The response values are non-negative integers is zero for at least some observations. Indeed, dependent variable is defined as to take non- • Observations are independent from one negative, non-zero values and it is characterized another by a distribution strongly skewed to the right • As the value of the mean increases, the prob- ability of zero counts is reduced • Conditional variance and mean are identical Table 2: Average amount sent per transaction or nearly the same: Var( y | x) 5 E( y | x). This and per year means that Poisson distributions with higher   Per Transaction Per Year mean values have correspondently greater variability. Mean Standard Mean Standard (€) Deviation (€) Deviation • The Pearson Chi2 dispersion statistic has a Morocco 212.9 223.03 1594.4 1518.8 value of approximately 1.0, which results when Peru 230.1 166.57 2113.0 1783.8 the observed and predicted variances of the Romania 253.9 278.98 1732.3 2066.2 response are the same.  11 12 GREENBACK 2.0 survey 2015 Figure 6: Annual remittances in euro, total and by country of origin Morocco Peru 30 20 10 Percent 0 Romania Total 30 20 10 0 0 5000 10000 0 5000 10000 Remittance per year ( ) The y axis shows the share (%) of observations per each amount of remittances. The lines plot a normal density distribution on the histograms. When a Poisson model is overdispersed, the Pois- son dispersion statistic, Pearson Chi2/(n 2 r), is Regression results greater than one and the negative binomial value Table 3 presents the regression results for three of  is greater than zero. A true Poisson model different specifications of the negative binomial has a dispersion statistic of one and a negative model. Model (1) presents a full list of demo- binomial dispersion parameter of zero (Hilbe graphic variables for migrants and their families, 2014). Our empirical data seem to be overdis- while Model (2) adds information on the level of persed even after checking for bigger outliers labor market integration and income at destina- at the top of the distribution and using a robust tion. Model (3) substitutes the variables referring variance estimator to get robust standard errors. to family composition across borders (number Moreover, we do not have zero values. Hence we of relatives and their place of residence) with the finally adopt a negative binomial model, which summary variable forged on the three migration is still based on the Poisson one but controls for profiles presented above. The estimated signs of some overdispersion and allows reducing the bias the relationship between each independent vari- of our estimated coefficients and standard errors. able and our outcome (remittance amount) are The traditional negative binomial model (NB2) those expected. Table 3 presents the regression has the same distributional assumptions as the results in the form of incidence rate ratios, the Poisson distribution, with the exception that it has estimated rate ratio for one unit increase in each a second parameter (the dispersion parameter) variable, holding constant the other regressors.4 which provides for a wider shape to the distribu- tion of counts than that allowed in the Poisson Interestingly, gender has no significant effect distribution. In NB2 the variance is affected by a on the total amount remitted, once controlled dispersion parameter (a) and the square of the for other demographic and economic variables. mean (2):  1 a2. In our case, the NB2 model Hence, females would remit the same amount specification proved to fit data better than a Pois- son model, adjusting for the overdispersion of 4 If the IRR is less than 1, the effect on the dependent variable (remittances) the data. is negative. If the IRR is higher than 1, the effect is positive. THE EMPIRICAL MODEL 13 Table 3: Negative binomial model (NB2) regression results (1) (2) (3) VARIABLES IRR IRR IRR Sex (1=female, 0=male) 0.987 1.090 1.106 (0.0694) (0.0831) (0.0886) Age (years) 0.952* 0.943** 0.915*** (0.0244) (0.0227) (0.0243) Age squared (years2) 1.001* 1.001* 1.001** (0.000297) (0.000279) (0.000307) Age at arrival in Italy (years) 1.013** 1.024*** 1.046*** (0.00642) (0.00668) (0.00839) Education level (base: low) – medium 1.042 0.979 0.956 (0.0783) (0.0710) (0.0763) – high 0.885 0.770** 0.660*** (0.0979) (0.0820) (0.0792) Mixed couple (1=partner with another citizenship at birth) 0.874* 0.859** 0.843** (0.0626) (0.0626) (0.0650) Children in Italy (1=has at least 1 child in Italy) 0.740*** 0.716*** (0.0787) (0.0819) Children in the Country of Origin (1=has at least 1 child 1.497*** 1.535*** in the CoO) (0.128) (0.130) N. of relatives living in Italy, same HH 0.897*** 0.888*** (0.0336) (0.0366) N. of relatives for whose expenditure he feels to 1.086*** 1.074*** contribute to (burden) (0.0178) (0.0171) Receiving remittances (1=receives money from outside Italy) 1.001 0.978 (0.102) (0.0914) Migratory profile (base: Loners) – Pioneers 0.710*** (0.0691) – Followers 0.644*** (0.0654) Regular at arrival (0=no visa or tourist visa) 0.836*** 0.796*** 0.791*** (0.0522) (0.0504) (0.0567) Country of Origin (base: Morocco) – Peru 1.242** 1.124 1.140 (0.105) (0.0916) (0.0993) – Romania 0.943 0.779*** 0.767*** (0.0906) (0.0743) (0.0772) Annual income (€) 1.000*** 1.000*** (8.32e-06) (7.38e-06) Mono income HH (1=individual and HH incomes coincide) 1.068 1.129 (0.0674) (0.0863) Type of job (base: Low qualified) – Medium (qualified workers in trade & services) 1.123* 1.190** (0.0749) (0.0897) – High (officers, professionals, technicians and managers) 1.284* 1.316* (0.179) (0.213) Savings in the last year (1=was able to save some money over 1.133* 1.174* the last year) (0.0843) (0.103) Constant 3,378*** 2,252*** 4,692*** (1,820) (1,134) (2,499) lndelta 1,122*** 978.0*** 1,212*** (97.49) (80.70) (87.45) Observations 476 474 474 Robust standard errors in parentheses ***p , 0.01, **p , 0.05, *p , 0.1 14 GREENBACK 2.0 survey 2015 as males if they earned the same, had the same regular at arrival’ in determining a lower amount education level and the same age and family of remittance now. While migrants in an undocu- structure. Age has a convex, decreasing relation- mented position at the time of the interview were ship with remittances: recalling that interviewed very few, their legal status at the beginning of migrants were aged between 18 and 64, younger their presence in Italy was much more diversified migrants remit more but the decrease in remit- with many entering without the required docu- tance with age reaches a maximum and then ments or overstaying a tourist visa. These results stops. Consistently, age at arrival which testifies seem to be in line with other studies which found for the seniority of migrants in Italy, is positively that undocumented migrants keep stronger con- associated with remittances: migrants arriving at nections with their origin families because of their minor ages have probably less connections with greater uncertainty and the need of putting their their origin countries than those who migrated money safely out of their destination country as adults. As for the education level, the highly- (Markova and Reilly 2007). Also, this effect could educated remit less in comparison with those with be partially due to the fact that those regular at a low level of education, maybe because they the beginning often entered through family reuni- belong to wealthier families at origin. Being part fication processes, hence are less economically of a mixed couple—the spouse or the partner is active on average than those migrated for the of a different nationality than the interviewee—is primary purpose of working in Italy. associated with a significantly lower amount of remittances sent: this might be the effect of a As for the cross-country comparisons, Peruvians stronger and more permanent integration in Italy are associated with a bigger amount of annual with an Italian partner or spouse or alternatively remittances, followed by Moroccans and then the result of a bargaining process within the cou- Romanians. Since this pattern holds significant ple about who receives the money among rela- even after controlling for all the demographic and tives in two different countries of origin. family formation patterns, this might be the sign of some underlying differences between migrants All the variables on the composition and distribu- of our three origin countries which are not cov- tion of the family members between Italy and the ered by our data. country of origin are significantly associated with the amount remitted. Having at least one child in The second model specification also includes Italy has the opposite effect on remittances than variables accounting for the level of labor mar- having one or more children left behind at origin. ket integration in Italy and for the individual and household economic condition. Indeed, individual The number of relatives living in the same house- income can be seen as a proxy of the capacity to hold in Italy is associated with less money sent remit as an economic possibility and the ability abroad. On the contrary, the higher the number of control over earned money (Carling and Hoel- of relatives for which the respondent feels to con- scher 2013), coupled by the degree of job quali- tribute to, the higher the overall amount devoted fication (often associated with higher incomes). to remittances each year. Interestingly, declaring Although the Italian labor market does not often of having received money from relatives at origin offer migrants good chances of matching employ- or in third countries over the past year does not ment with acquired qualifications and tends to impact on the amount of remittance sent: this employ migrants in low qualified, service sectors’ means that money flows in more than one direc- occupation (Castagnone et al. 2014), migrants’ tion and that, over the same year, migrants can better labor market trajectories testified by better receive money from some relatives and at the jobs and higher incomes are associated with more same time send money to others within a com- remittances sent abroad, other things being equal. plex, multi-directional reciprocal network. As for This might be in contradiction with the effect the legal immigration status, we tried to test both found for the education level, but it is plausible the current status and the one held by migrants; since there is a strong degree of over-qualification interestingly, we found no effect for the variable (many of the medium and highly-skilled work in ‘being regular at the time of interview’, while there low qualified occupations) and the two variables is a strong, significant estimated effect of ‘being are not strongly correlated. In the same direction, THE EMPIRICAL MODEL 15 migrants who declared to have been able to save Table 4: Chi2 test on migration profile some money since the beginning of the year— (base: Loner) independently from the use of these savings—are (1) [remittances]2.pioneer 5 0 those who engage in higher monetary transfers to their origin households. (2) [remittances]3.follower 5 0 chi2(2) 5 18.4 The third and last column of Table 4 tries to test a Prob > chi2 5 0.0001 model with a reduced number of variables with- out loss of explanatory power and goodness of fit. Substituting all family and household related variables with a categorical variable which sum- Indeed, migrants who are living and working in marizes them proves that our three migration pro- Italy alone, without any members of the nuclear files are a meaningful way of clustering migrants family in the country of destination, are remitting around their family and household structure in significantly more than both Pioneers and Fol- Italy and abroad. After running a two degree-of- lowers, as they both have family members in Italy freedom chi-square test on the migratory profiles’ who supposedly require migrants to address more variable, we know that this variable is a statisti- expenses in Italy and to leave remittances for a cally significant predictor of the amount annual of residual part of their savings. Loners are those remittances sent annually. who have not yet formed a family in Italy and who engage in more transnational behaviors to keep The significant difference in the amount of remit- alive their family network abroad: they might be tances associated with being either Loner, Pio- either repaying their families for the efforts made neer or Follower (Figure 7) tells us that it is not to start the migration process, or to be preparing only each single variable, while everything else is a family reunification process or, on the contrary, kept constant, to be associated with the amount their return at origin after having earned a pre- of remittances, but that also complex migration fixed amount of money. As we showed, Pioneers profiles help us in distinguishing the remittance and Followers have a similar family spatial distribu- behavior of a sample of selected working and tion between Italy and the country of origin and remitting migrants. they also look similar in the amount of remittances Figure 7: Predictive amount of remittances (calculated on model 3) Predicted amount of remittances per migratory profile, by country or origin. 3000 2500 Remittances (euro) 1500 2000 1000 Morocco Peru Romania Loners Pioneers Followers 16 GREENBACK 2.0 survey 2015 sent as they have few relatives at home and, in due to unobserved variables which characterize particular, virtually no children left behind. the three sub-samples. Indeed, we are not able to test the effect of different return intentions, which Moreover, also in the last model specification, are likely to impact the level of economic engage- differences across the three national communi- ment at origin, and we do not have measures ties remain significant. As visually summarized by of physical mobility. Those who can travel more Figure 7, Peruvians remit more on average than often and easily to the origin countries because Moroccans and Romanians taking other variables of shorter geographical distance (Romanians and at their means. The difference is big enough that Moroccans) or thanks to legal provision on free Peruvian Followers not only surpass their Moroc- mobility within the EU (Romanians) are plausi- can and Romanian counterparts but also have bly more likely to bring money and consumption level of remittances comparable to Moroccan goods during their journeys, decreasing the need Pioneers and Romanian Pioneers and Loners. As for remittance transfer in comparison with trans- said, these cross-country differences might be oceanic migrants (Peruvians). 4 Conclusions T his paper discussed the role of the household and family structure in determining the amount of remittances sent by migrants, taking Italy, the origin country and eventually third coun- tries, and the composition of their household in Italy help explain the variation in the amount of into account their migration history and the trans- remittances, in accordance with other empirical national distribution of their nuclear families. studies (Ulku 2012; Nziramasanga and Yoder 2013; Marchetti and Venturini 2014). Family reunifica- Making use of a new and quite rich dataset on tion processes, especially of children left behind, migrants residing in Turin, the analysis tried to alter the transnational distribution of relatives and outline the key determinants of remittances. the related framework of migrants’ obligations In particular, it provided evidence for working (Ambrosini 2013). migrants from three traditional labor-exporting countries—Morocco, Peru, and Romania—which Cross-country differences appear to be non- represent three of the main origin countries for negligible: other things being held constant, migrants in Italy both in terms of population and Peruvians remit more than Moroccans and Roma- remittances. Notwithstanding the data limita- nians on average, and we know from the original tions which derive from the cross-sectional struc- research that they are also more constant, as they ture of the sample, its geographical scope and remit small amounts at high frequencies (The the restrictive selection criteria, the depth and World Bank 2014). The combination of results breadth of the original questionnaire allowed a by country of origin and by migratory profiles detailed analysis of migrants’ demographic and resonates with the type and history of migration socio-economic characteristics, their level of labor in Italy and specifically in Turin, with regards to market integration, their savings and remittances the three observed nationalities (Pastore, Salis, behaviors. and Villosio 2013). Indeed, in comparison with the rest of the sample, Moroccans are character- Our research question revolved around the role ized by both a wider presence of male migrants of migration history, the family spatial distribu- alone—who have not yet formed a family—and tion in determining the amount of remittances of Pioneers, who had the time to pursue fam- sent each year by working migrants. In order to ily reunification after settlement thanks for their test the specific role of family structures and net- long presence in Italy. Although this paper does works across borders, we identified three distinct not engage with the impact of the economic cri- migratory profiles—the Loners, the Pioneers, and sis on remittances (see Bartolini and Castagnone the Followers—characterized by the timing and 2015 for a detailed analysis on this), we know that sequence of the migration event with respect to Peruvians are more concentrated in occupations the rest of the nuclear family. Our profiles proved (qualified jobs in the domestic and health care to behave differently with regards to remittances: services) which have been less hit by the eco- migrants alone at destination, with no members nomic crisis than the construction and industry of the nuclear family in Italy, remit significantly sectors where more Romanian and Moroccan, more than both pioneers and followers, who on especially male, workers are employed. the contrary have to address higher expenses in Italy to support their relatives. Hence, the spatial Reasons for these significant cross-country dif- distribution of migrants’ nuclear families between ferences are also likely to be found in variables  17 18 GREENBACK 2.0 survey 2015 which we do not observe directly from our sur- distance and cheaper means of transportation vey data. Remittance decisions, consistently with (Romanians and Moroccans) or thanks to the free what we describe in our analysis, are connected mobility provisions within the EU (Romanians) can to the life cycle of migrants and to their plans in easily bring with them money and durable goods, terms of settlement or return: those with tem- decreasing the need for money transfers which is porary migration plans, who are not engaging in higher for transoceanic migrants (Peruvians). family reunification processes and who plan to return home after a definite period of time, are The empirical evidence on micro motives to remit more likely to invest more in keeping economic should then be combined with considerations and social relationships with their origin house- on more long-term, institutional drivers of remit- hold and to send more remittances (Dustmann tances with regard to the economic and policy and Mestres 2010; Delpierre and Verheyden 2014). environment in the origin and destination coun- Also, the literature suggests that return plans are tries which might hinder or foster remittances adjusted when professional prospects change and at the macroeconomic level (World Bank Group family formation choices are done by migrants. 2015). Our results resonate with the overall Ital- A further element which could help in explaining ian context of migrant integration and economic the national differences in size of remittances is transnationalism which is (still) a ‘basic’ one the feasibility of physical mobility for migrants of (Ambrosini 2013, 632) where remittance flows are different origins. The geographical distance, the consistent with a relatively new migration history existence and accessibility of various means of and a strong although decreasing transnational transportation (not only airplanes, but also cars, family distribution. Further research could con- buses, ferry-boats, and trains) make the difference nect this type of economic transfers with other in determining how often migrants can visit their (‘advanced’) levels of transnationalism, which per- origin countries. Hence, those who can more often tain more to the sphere of entrepreneurship and/ travel home because of shorter geographical or symbolic belonging to the origin country.  19 References Åkesson, Lisa. 2011. “Remittances and Relation- African-European Migrants. Evidence from the ships: Exchange in Cape Verdean Transnational MAFE Survey.” International Migration Review Families.” Ethnos 76 (3): 326–47. 48 (4). Ambrosini, Maurizio. 2013. “Migration and Transna- Cingolani, Pietro, and Roberta Ricucci. 2013. tional Commitment: Some Evidence from the Le Collettività Di Origine Nordafricana in Italian Case.” Journal of Ethnic and Migration Piemonte, Tra Continuità E Cambiamento. Studies 40 (4): 619–37. Turin: FIERI. http://www.integrazionemigranti Azam, Jean-Paul, and Flore Gubert. 2006. .gov.it/ricerche/Documents/IRES%202013/ “Migrants’ Remittances and the Household in Focus%205_integrazione%202013.pdf. Africa: A Review of Evidence.” Journal of Afri- Cox, Donald, Zekeriya Eser, and Emmanuel can Economies 15 (suppl 2): 426–62. Jimenez. 1998. “Motives for Private Transfers Baio, Gianluca, Gian Carlo Blangiardo, and Marta over the Life Cycle: An Analytical Framework Blangiardo. 2011. “Center Sampling Technique and Evidence for Peru.” Journal of Develop- in Foreign Migration Surveys: A Methodologi- ment Economics 55 (1): 57–80. cal Note.” Journal of Official Statistics 27 (3): Delpierre, Matthieu, and Bertrand Verheyden. 2014. 451–65. “Remittances, Savings and Return Migration Banca d’Italia. 2014. “Le Rimesse Verso L’estero under Uncertainty.” IZA Journal of Migration Degli Immigrati in Italia.” July 7. http://www 3 (1): 22. .bancaditalia.it/statistiche/rapp_estero/rimesse. Dustmann, Christian, and Josep Mestres. 2010. Bartolini, Laura, and Eleonora Castagnone. 2015. “Remittances and Temporary Migration.” Jour- Remittances and the Economic Crisis: Evidence nal of Development Economics 92 (1): 62–70. from the Greenback 2.0 Survey in Italy. Green- Erdal, Marta Bivand. 2012. “Who is the money for? back Working Paper 1. Washington, D.C.: The Remittances within and beyond the household World Bank. in Pakistan.” Asian and Pacific Migration Journal Brown, Richard P. C., Jørgen Carling, Sonja Fran- 21 (4): 437–58. sen, and Melissa Siegel. 2014. “Measuring Fondazione Moressa. 2013. Rapporto Annuale Remittances through Surveys: Methodological Sull’economia Dell’immigrazione—Edizione and Conceptual Issues for Survey Designers 2013, Tra Percorsi Migratori E Comportamento and Data Analysis.” Demographic Research 31 Economico. Bologna: Il Mulino. (41): 1243–74. Guarnizo, Luis Eduardo. 2003. “The Economics of Carling, Jørgen. 2008. “The Determinants of Transnational Living.” International Migration Migrant Remittances.” Oxford Review of Eco- Review 37 (3): 666–99. nomic Policy 24 (3): 581–98. Hagen-Zanker, Jessica, and Melissa Siegel. 2007. ———. 2014. “Scripting Remittances: Making Sense The Determinants of Remittances: A Review of of Money Transfers in Transnational Relation- the Literature. Working Paper MSGoG/2007/ ships.” International Migration Review 48 WP003. Maastricht University. (September): S218–62. Hilbe, Joseph M. 2014. Modeling Count Data. Cam- Carling, Jørgen, and Kristian Hoelscher. 2013. “The bridge: Cambridge University Press. Capacity and Desire to Remit: Comparing Local Lucas, Robert E. B., and Oded Stark. 1985. “Moti- and Transnational Influences.” Journal of Ethnic vations to Remit: Evidence from Botswana.” and Migration Studies 39 (6): 939–58. Journal of Political Economy 93 (5): 901–18. Castagnone, Eleonora, Tiziana Nazio, Laura Barto- Marchetti, Sabrina, and Alessandra Venturini. 2014. lini, and Bruno Schoumaker. 2014. “Understand- “Mothers and Grandmothers on the Move: ing Transnational Labour Market Trajectories of Labour Mobility and the Household Strategies 20 GREENBACK 2.0 survey 2015 of Moldovan and Ukrainian Migrant Women in Washington, D.C: The World Bank. http:// Italy.” International Migration 52 (5): 111–26. siteresources.worldbank.org/INTPROSPECTS/ Markova, Eugenia, and Barry Reilly. 2007. “Bulgar- Resources/334934-1288990760745/ ian Migrant Remittances and Legal Status: MigrationandDevelopmentBrief23.pdf. Some Micro-Level Evidence from Madrid.” Ratha, Dilip, and Sanket Mohapatra. 2007. “Increas- South-Eastern Europe Journal of Economics 5 ing the Macroeconomic Impact of Remittances (1): 55–69. on Development.” Development Prospects Mazzucato, Valentina. 2011. “Reverse Remittances Group Note, The World Bank, November. in the Migration–development Nexus: Two-Way http://dilipratha.com/index_files/G8Berlin.pdf. Flows between Ghana and the Netherlands.” Sinning, Mathias. 2007. Determinants of Savings Population, Space and Place 17 (5): 454–68. and Remittances: Empirical Evidence from Nziramasanga, Mudziviri, and Jonathan Yoder. Immigrants to Germany. Discussion Paper 2013. “The Check in the Mail: Household Char- No. 2966. Bonn: IZA. acteristics and Migrant Remittance from the US Thai, Hung Cam. 2014. Insufficient Funds: The to Mexico.” Applied Economics 45 (8): 1055–73. Culture of Money in Low-Wage Transnational Pastore, Ferrucio, Ester Salis, and Claudia Villosio. Families. Stanford University Press. 2013. “L’Italia e l’immigrazione ‘low cost’: fine di The World Bank. 2014. Migrants’ Remittances from un ciclo?.” Mondi Migranti 1 (19). Italy—International Remittances and Access Posel, Dorrit. 2001. “How Do Households Work? to Financial Services for Migrants in Turin, Migration, the Household and Remittance Italy. Greenback 2.0 Report. Washington, D.C. Behaviour in South Africa.” Social Dynamics 27 https://remittanceprices.worldbank.org/sites/ (1): 165–89. default/files/migrants_remittances_italy.pdf. Rapoport, Hillel, and Frédéric Docquier. 2006. “The Ulku, Hulya. 2012. “Remitting Behaviour of Turk- Economics of Migrants’ Remittances.” In Hand- ish Migrants: Evidence from Household Data in book of the Economics of Giving, Altruism and Germany.” Urban Studies 49 (14): 3139–58. Reciprocity, edited by Serge-Christophe Kolm UNDP. 2014. Human Development Report 2014— and Jean Mercier Ythier, 2:1135–98. Applica- Sustaining Human Progress: Reducing Vulner- tions. Elsevier. abilities and Building Resilience. United Nations Ratha, Dilip. 2007. Leveraging Remittances for Development Programme. Development. Policy Brief. Migration Policy World Bank Group. 2015. Global Economic Pros- Institute. http://dilipratha.com/index_files/ pects, January 2015: Having Fiscal Space and Ratha-Remittances-Oslo-February2007.pdf. Using It. Washington, D.C.: The World Bank. Ratha, Dilip, Supriyo De, Ervin Dervisevic, http://www.worldbank.org/content/dam/ Christian Eigen-Zucchi, Sonia Plaza, and Worldbank/GEP/GEP2015a/pdfs/GEP15a_ Kirsten Schiettler. 2014. Migration and web_full.pdf. Remittances: Recent Developments and Yang, Dean. 2011. “Migrant Remittances.” The Jour- Outlook—Special Topic: Forced Migration. nal of Economic Perspectives 25 (3): 129–51. 23. Migration and Development Brief.