WPS6812 Policy Research Working Paper 6812 Remittances and Vulnerability in Developing Countries Giulia Bettin Andrea F. Presbitero Nikola Spatafora The World Bank East Asia and the Pacific Region Office of the Chief Economist March 2014 Policy Research Working Paper 6812 Abstract This paper examines how international remittances are shocks, such as natural disasters or large terms-of-trade affected by structural characteristics, macroeconomic declines. Financial development in the source economy, conditions, and adverse shocks in both source and which eases access to financial services for migrants and recipient economies. The paper exploits a novel, rich reduces transaction costs, is positively associated with panel data set, covering bilateral remittances from 103 remittances. Conversely, recipient-country financial Italian provinces to 87 developing countries over the development is negatively associated with remittances, period 2005–2011. Remittances are negatively correlated suggesting that remittances help alleviate credit with the business cycle in recipient countries and increase constraints. especially strongly in response to adverse exogenous This paper is a product of the Office of the Chief Economist, East Asia and the Pacific Region. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at nspatafora@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Remittances and Vulnerability in Developing Countries∗ Giulia Bettin† Andrea F. Presbitero‡ Nikola Spatafora§ March 20, 2014 JEL Codes: F33, F34, F35, O11 Keywords: Remittances, Shocks, Business Cycle, Vulnerability, Gravity Model ∗ We thank Giacomo Oddo, Roberto Tedeschi and Simonetta Zappa at the Bank of Italy for clarifications on the remittances data set. We also thank Caglar Ozden, Dilip Ratha, and participants at the 4th International Conference “Economics of Global Interactions: New Perspectives on Trade, Factor Mobility and Development” and at seminars at the International Monetary Fund and KNOMAD (The World Bank) for useful comments. The views expressed in this Working Paper are those of the author(s) and do not necessarily represent the views of the International Monetary Fund, the World Bank, or the policies of these institutions. † Giulia Bettin, Università Politecnica delle Marche (Italy) and MoFiR. E-mail: g.bettin@univpm.it. ‡ Andrea F. Presbitero, International Monetary Fund and MoFiR. E-mail: apresbitero@imf.org. § Nikola Spatafora, The World Bank. E-mail: nspatafora@worldbank.org. 1 1 Introduction Developing countries remain extremely vulnerable to adverse exogenous shocks. The global fi- nancial crisis and the world food price crisis of 2008 delivered a reminder of their macroeconomic, external, and fiscal vulnerabilities (Figure 1, left panel). In addition, over the past decades devel- oping countries have been increasingly subject to natural disasters, again with severe consequences in terms of output, trade, and fiscal balances (Raddatz, 2007; Noy, 2009). As a result, the policy debate is again focusing on developing countries’ vulnerability (Schindler et al., 2011). This paper examines how remittances are affected by structural characteristics, macroeconomic conditions, and adverse shocks in both source and recipient economies. It therefore sheds light on whether remittances should be viewed as a countercyclical shock absorber, helping smooth con- sumption during a downturn, in contrast to the typically pro-cyclical private capital flows. This issue is particularly salient for two reasons. First, spurred by increasing migration, remittances to developing countries have grown steadily relative to capital flows (Figure 1, right panel). Remit- tances to developing countries are projected to reach USD 414 billion in 2013, more than three times the size of official development assistance, and USD 540 billion by 2016 (The World Bank, 2013). Second, remittances have proved very resilient since the onset of the global financial crisis. The existing empirical evidence on the role of remittances as a shock absorber is inconclusive. Some studies suggest that remittances are countercyclical with respect to output in the recipient country, because they are driven by altruism (Agarwal and Horowitz, 2002; Osili, 2007), or because household members migrate as part of a risk-diversification strategy aiming to insure against income shocks (Yang and Choi, 2007). Other studies emphasize that remittances can be pro-cyclical, because migrants’ decision to remit is also driven by factors such as investment in physical and human capital (Yang, 2008; Adams Jr. and Cuecuecha, 2010; Cooray and Mallick, 2013). This paper re-examines the question using a novel, rich panel dataset, covering bilateral remit- tances from 103 Italian provinces to 87 developing countries over the period 2005-2011, to estimate a model for bilateral remittances. In this dataset, remittances display significant variability, both over time and across source provinces and recipient countries. Italy can be considered a representative case study given that remittance outflows increased substantially in the last years, notwithstanding the global slowdown, and they are destined to all world regions and not limited to neighboring regions such as Eastern Europe, Central Asia or Northern Africa. More specifically, the paper makes three main contributions to the literature. First, the availabil- ity of bilateral panel data for a large sample of recipients makes it possible to analyze systematically the correlation between remittances and the business cycle in both source and recipient economies, washing out the effect of time invariant factors at the level of country-province pairs. We consider separately the cyclical and trend components in GDP per capita. In addition, we control for specific factors of vulnerability in recipient countries, including in particular natural disasters, large declines in the terms of trade, and armed conflicts. In contrast, the existing literature focuses mainly either on bilateral remittances for a limited sample of receiving countries (Lueth and Ruiz-Arranz, 2008; Frankel, 2011; Docquier et al., 2012), or on country pairs, such as the US-Mexico or the Germany- Turkey corridors (Sayan, 2004; Vargas-Silva, 2008). As a result, existing works fail to settle the 2 Figure 1: Vulnerability and capital flows in developing countries Vulnerability index (Dabla Norris & Bal Gunduz 2012) .6 600 .5 current USD bln 400 .4 .3 200 .2 1990 1993 1996 1999 2002 2005 2008 2011 0 1990 1993 1996 1999 2002 2005 2008 2011 External sector Fiscal sector Overall economy and institutions Overall index FDI ODA Remittances Private debt & portfolio eq. (a) Vulnerability Index (b) Capital flows Source: World Development Indicators and International Debt Statistics (developing countries are defined as low and middle income countries), The World Bank, for the left panel. Data for the right panel are elaboration on the Vulnerability Index data set. See Dabla-Norris and Bal-Gunduz (2012) for details on how the index and its sub-components are constructed. empirical debate on the correlation between remittances to developing countries and the business cycle in the source and recipient economies. Related to this, our data on remittances cover the periods before and after the 2007-08 financial crisis, allowing an analysis of the response of remittances to economic conditions in source and recipient economies during the global financial crisis. This is particularly relevant because the global financial crisis affected jointly the migrants’ home and host countries, with an a priori ambiguous aggregate effect on remittances. On the one hand, the downturn in the home country might induce a positive change in remittances driven by altruism or insurance. On the other hand, the recession in the host country would reduce the income of migrants, including in particular temporary workers employed in the construction sector.1 Second, we deal with the possible endogeneity of the business cycle in the recipient country more satisfactorily than previous studies. Reverse causality from remittances to output may sig- nificantly bias estimates, since remittances often represent a large share of developing countries’ GDP, and they have been found to affect both output growth and financial development (Gupta et al., 2009; Giuliano and Ruiz-Arranz, 2009; Bettin and Zazzaro, 2012). In some cases, most of the transfers come from a limited set of origin countries, so that even bilateral remittances represent a significant share of GDP, potentially leading to reverse-causality issues in cross-country bilateral data. According to the data in Lueth and Ruiz-Arranz (2008), during 2002-2004 remittances from Russia to Tajikistan equaled on average almost 12 percent of Tajikistan’s GDP, while remittances from the USA to the Philippines represented almost 5 percent of the Philippines’ GDP.2 To the 1 According to the Italian National Institute of Statistics, the unemployment rate for foreign-born workers increased from 10.2 percent in 2005 to 12.1 percent in 2011. The figures for native workers were respectively 7.6 percent in 2005 and 8.0 percent in 2011. 2 We thank Marta Ruiz Arranz for sharing the data on bilateral remittances. 3 best of our knowledge, the existing studies either disregard this issue, or deal with it relying on lagged values or other internal instruments. However, such a strategy is not likely to fully solve the problem. Finding good instruments in a cross-country dataset is challenging, since one needs a variable which is related to economic conditions in the recipient country, but not to remittances (Arezki and Brückner, 2012). The structure of our dataset makes it possible to circumvent this problem: considering only remittances from Italian provinces, rather than aggregate remittance inflows, significantly attenuates the endogeneity of the recipient country’s business cycle. Third, we investigate the relationship between remittances and financial development in the remittance source economy. The literature has generally focused on financial development in the recipient country, finding that remittances promote financial development (Gupta et al., 2009) and that financial development enhances the impact of remittances on growth (Giuliano and Ruiz- Arranz, 2009; Bettin and Zazzaro, 2012). In contrast, we exploit the cross-sectional dimension of the bilateral dataset to test whether the degree of development and the functional proximity of source-province credit markets to local economies play a role in fostering remittances. We find that remittances from Italian provinces are negatively correlated with the business cycle in recipient countries, and increase especially strongly in response to adverse exogenous shocks, such as natural disasters or large declines in the terms of trade. In addition, remittances are positively correlated with potential GDP in recipient countries. These results are consistent with remittances being driven by both altruism and investment motives. Remittances are also positively correlated with economic conditions in the source province. Nevertheless, in the presence of similar negative shocks to both source and recipient economies, remittances remain countercyclical with respect to the recipient country. Financial development in the source province is positively associated with remittances, likely because it reduces transaction costs and eases access to financial services for migrants. Conversely, financial development in the recipient country is negatively associated with remittances, suggesting that remittances help alleviate credit constraints. The paper is structured as follows. Section 2 offers a detailed review of the existing literature on the macroeconomic determinants of remittances. Section 3 describes the data and the estimated model. Section 4 presents selected statistics about remittances outflows from Italian provinces to developing countries. Section 5 discusses the empirical results. 2 The determinants of migrants’ remittances 2.1 The cyclicality of remittances There exists a large literature on the determinants of migrants’ remittances.3 The empirical evidence so far, however, remains inconclusive as to how remittances react to business cycles in the migrants’ home country, and whether they help mitigate economic hardship. At the microeconomic level, some studies find that remittances increase to compensate relatives for negative shocks to 3 Rapoport and Docquier (2006) provide an exhaustive review of modern theoretical and empirical literature on remittances. 4 their income—the altruism motive (Agarwal and Horowitz, 2002). Others find a positive correlation between remittances and the economic conditions of families back home, suggesting that remittances are driven by self-interest motives such as investment or inheritance.4 In any case, positive shocks to migrants’ income in host countries are likely to translate into larger remittances (Bettin et al., 2012).5 Likewise, some macroeconomic studies find that remittances are negatively correlated with income levels in the recipient country (El-Sakka and McNabb, 1999; Frankel, 2011; Singh et al., 2011), mitigate the adverse effect of food-price shocks on the level and instability of household consumption in vulnerable countries (Combes et al., 2014), reduce output growth volatility in developing economies (Bugamelli and Paternò, 2011; Chami et al., 2012), and react positively to natural disasters (Yang, 2008; Mohapatra et al., 2012; Ebeke and Combes, 2013). In contrast, other studies find that remittances are procyclical with respect to the recipient countries, consistent with an investment motive (Sayan, 2004, 2006; Lueth and Ruiz-Arranz, 2008; Giuliano and Ruiz-Arranz, 2009; Durdu and Sayan, 2010; Cooray and Mallick, 2013). Finally, some works do not find any significant correlation either with the business cycle in migrants’ home countries (Akkoyunlu and Kholodilin, 2008) or with specific shocks such as armed conflicts (Naudé and Bezuidenhout, 2012). Ruiz and Vargas-Silva (2014) argue that results on the cyclicality of remittances with respect to the receiving economy are country- or corridor- specific and unlikely to be stable over time. In addition, how remittances’ react to negative economic shocks in recipient countries may depend on other country-level characteristics. Arezki and Brückner (2012), for example, show that the impact of rainfall-driven income shocks on remittance inflows decreases with the level of financial development in the country.6 2.2 Limitations of existing studies Cross-country analyses of the macroeconomic determinants of remittances typically use data on remittance inflows in developing countries, and disregard heterogeneity across source economies. To overcome this limitation, some studies use bilateral data on remittances to control for the host countries’ characteristics, such as output fluctuations. In most cases, however, the geographical coverage is limited to a single remittance corridor.7 4 Lucas and Stark (1985) and Osili (2007) both show that remittances are positively correlated with the income of recipient households. Analogously, de la Briere et al. (2002) and Hoddinott (1994) show that remittances are positively correlated with household wealth. 5 Macroeconomic studies have considered a wide range of potential determinants, including exchange rates (Faini, 1994), interest rate differentials (El-Sakka and McNabb, 1999), the size of the diaspora abroad and transaction costs (Freund and Spatafora, 2008), the skill and gender composition of migrant stocks (Faini, 2007; Adams Jr., 2009; Niimi et al., 2010), and the interaction with immigration policies (Docquier et al., 2012). Recent studies have also investigated whether remittances may represent an important channel in propagating global shocks (Chami et al., 2010; Barajas et al., 2012). 6 Sayan (2006) considers a sample of 12 countries, and highlights the acyclical behavior of remittances in some countries. The differences between the results obtained for the whole group and for single countries translate into a warning that cross-country results might conceal possibly significant differences across individual countries. 7 Sayan (2004) and Akkoyunlu and Kholodilin (2008) focus on the Germany-Turkey remittance corridor, while Vargas-Silva (2008) and Ruiz and Vargas-Silva (2014) look at U.S.-Mexico remittances. Durdu and Sayan (2010) consider both corridors. 5 A few studies use bilateral data but adopt a wider geographical perspective. Lueth and Ruiz- Arranz (2008) use a panel dataset on bilateral remittances for 11 European and Asian recipient countries during the period 1980-2004. Frankel (2011) merges their data with other bilateral data on remittances from the Inter-American Development Bank and the European Commission (Jimenez- Martin et al., 2007). Docquier et al. (2012) merges the sources used by Frankel (2011) with a database from the European Central Bank and a Romanian database; the resulting dataset include 89 sending countries but is still limited to 46 receiving countries, both developing and developed. As remittances represent a nonnegligible share of GDP in many recipient countries, the results of existing studies could be severely biased by reverse causality from remittances to GDP, with the exception of the few studies that focus on exogenous income shocks (see, for instance Yang, 2008). In cross-country aggregate level analyses this issue is seldom addressed by means of instrumental variables (Singh et al., 2011) or GMM techniques (Cooray and Mallick, 2013). In the context of bilateral data, Frankel (2011) addresses endogeneity issues concerning the size of migrant stocks, but disregards the possible bias related to the receiving country’s GDP. Lueth and Ruiz-Arranz (2008) acknowledge the endogeneity problem and maintain that GMM estimates that use lagged values of growth in the recipient countries confirm their results about the procyclical pattern of remittances. However, it remains unclear whether GMM estimates that use lagged values of growth address the issue satisfactorily, as concerns about the capacity of GMM to address causality are mounting, for instance because of weak instruments and the over-fitting of endogenous variables (Roodman, 2009; Bazzi and Clemens, 2013).8 3 Empirical Strategy and Data 3.1 The empirical model To identify the effect of business cycle fluctuations and financial development on remittances we estimate a model in which bilateral remittances are a function of a set of independent variables constructed by exploiting information on both migrants’ origin countries and their host Italian provinces. In the baseline specification, total bilateral remittances between the source province i and the recipient country j at time t (REMi,j,t ) are a function of the logarithm of actual GDP per capita over potential GDP per capita in the source province (CY CLEi,t ) and in the recipient country (CY CLEj,t ), the log of trend GDP per capita (T REN Di,t , T REN Dj,t ), the log of 1 + the bilateral stocks of migrants (M IGRAN T Si,j,t ), and the log of population levels (P OPi,t ,P OPj,t ): REMi,j,t = α1 CY CLEi,t + α2 CY CLEj,t + α3 T REN Di,t + α4 T REN Dj,t + +β1 M IGRAN T Si,j,t + β2 P OPi,t + β3 P OPj,t + µi,j + i,j,t (1) 8 Results on specific remittance corridors might also suffer from reverse causality. Sayan (2004) and Durdu and Sayan (2010) investigate the cross-correlations between remittances from Germany to Turkey and cyclical fluctuations in Turkish and German GDP without discussing the direction of causality. Akkoyunlu and Kholodilin (2008) estimate a VAR model and find no evidence in favor of Granger causality from remittances from Germany to Turkish GDP. 6 where i,j,t is the standard error term. To control for any time invariant bilateral unobservables, we include country-province pair fixed effects (µi,j ) in equation 1. The key coefficients of interest are the correlation between remittances and the business cycle in, respectively, the source province, α1 , and the recipient country, α2 . Remittances are countercyclical with respect to output fluctuations in the recipient country if α2 < 0; this case suggests an altruistic motivations behind transfers. A positive correlation between remittances and the long-run output trend in the recipient country, α4 > 0, instead offers evidence in favor of an investment motive for remittances: investment-driven remittances may be particularly sensitive to long-term prospects in migrants’ home country. The model in equation 1 can be augmented to include additional source-province and recipient- country controls: REMi,j,t = α1 CY CLEi,t + α2 CY CLEj,t + α3 T REN Di,t + α4 T REN Dj,t + β1 M IGRAN T Si,j,t + β2 P OPi,t + β3 P OPj,t + γ1 Xi,t + γ2 Zj,t + µi,j + i,j,t (2) where Xi,t and Zj,t refer respectively to province- and country-level characteristics. We deepen our analysis of the role of remittances as shock absorbers in recipient countries by including among the country-level characteristics, Zj,t , three specific factors of vulnerability for developing countries: an indicator equal to 1 if country j experienced natural disasters in year t (DISAST ERj,t ); an indicator equal to 1 if armed conflicts occurred in country j at time t (W ARj,t ); and an indicator equal to 1 if country j experienced a major negative shock to the terms of trade (T T SHOCKj,t ), defined as an observation falling in the lowest 5 percent of the distribution of the annual variation in the terms-of-trade index. Adverse shocks in these exogenous variables, controlling for output per capita, may be particularly likely to evoke a sympathetic (or, alternatively, insurance-type) response among migrants. We also examine the impact of financial development on remittances. First, we consider differ- ences in financial development across recipient countries, as proxied by the logarithm of the share of credit to the private sector over GDP (F IN DEVj,t ). Their effect is a priori ambiguous. On the one hand, countries with more developed credit markets should attract greater remittances, as a result of either lower transaction costs (Freund and Spatafora, 2008), or the capacity of an efficient banking system to channel profit-driven remittances towards growth-enhancing projects (Bettin and Zazzaro, 2012). On the other hand, remittances and financial development may be substitutes: migrants whose relatives have limited access to financial resources at home may transfer resources to relax liquidity constraints and fund either consumption or investments in physical and human capital (Giuliano and Ruiz-Arranz, 2009). Second, we consider differences in financial development across Italian provinces, the source of remittances. We expect more developed provincial financial markets to be correlated with greater remittance outflows, for two reasons. Greater provincial banking-sector penetration, as proxied by the number of local bank branches per inhabitant (BAN Kj,t ),9 will reduce the trans- 9 This is a widely used measure of local financial development; for an application to Italy, see Bonaccorsi di Patti and Gobbi (2001). 7 action costs associated with remittance transfers, and encourage greater remittances (particularly through the formal sector) (Freund and Spatafora, 2008).10 In addition, the propensity of mi- grants to remit (again, particularly through formal channels) may depend on the institutional, cul- tural, and informational gaps between migrants and the host province’s financial system (Albareto and Mistrulli, 2011). We proxy these gaps using a measure of the functional distance between banks and local economies, based on whether banks are headquartered in the relevant province (F U N CT ION AL DIST AN CEj,t ). Intuitively, when banks are headquartered in an area, they are better able to collect local information, and as a result are more likely to serve the economic needs of the area (Alessandrini et al., 2009), including the needs of resident migrant workers. Since the dependent variable REMi,j,t has a significant share of non-randomly distributed zeros (that is, many empty country-province cells), equation 1 and 2 are estimated using the Fixed Effects Poisson estimator. Despite deriving originally from the analysis of count data, the Poisson estimator can also be applied to non-negative continuous variables (Wooldridge, 2010). Poisson regression estimates are consistent in the presence of heteroskedasticity and reasonably efficient, especially when considering large samples. Thanks to its multiplicative form, the Poisson specification also provides a natural way to deal with zero observations in the dependent variable instead of either transforming or excluding them from the sample (Silva and Tenreyro, 2006; Burger et al., 2009). In the baseline estimates, we include country-province pair and time fixed effects and additionally control for the potential correlation of errors at the bilateral level by clustering standard errors by country-province pairs.11 3.2 Data sources The variables used in equations 1 and 2 are constructed using data collected from many sources. Here we provide an overview; a precise definition of each variable and of its sources is in Table 1. The main data source is a detailed panel dataset on bilateral outward remittances from 103 Italian provinces to 87 developing countries, providing annual data at constant prices for the period 2005-11, compiled by the Bank of Italy (see Table 2 for a list of recipient countries included in the sample).12 The dataset covers remittances sent through formal channels, and predominantly reflects transfers carried out through money-transfer operators and the postal system. The banking system has been included in the survey only since 2010, and accounts for 5 to 10 percent of total remittances. All formal transactions are reported, regardless of the amount. As a caveat, the dataset does not include remittances sent through informal channels. 10 Ideally, we should rely on a more precise measure of transaction costs, such as the service fees charged by banks and money transfer operators for international transfers, as done by Freund and Spatafora (2008). However, those data are not available at the provincial level and for the time span of our analysis. We employ bank branches penetration as a proxy for transaction costs and for the level of financial development at the provincial level. 11 In the robustness section, we control for unobservables by separately including country and province fixed effects instead of country-province pair fixed effects, so that we can also identify the effect of specific time invariant variables at bilateral level. 12 Data on remittance flows to 204 destination countries are collected as part of a monthly survey car- ried out by the Bank of Italy on a provincial basis since 2005. The dataset is publicly available at: www.bancaditalia.it/statistiche/rapp_estero. Our sample is limited to the 87 developing countries where data on the control variables for the extended model specification are available. 8 Bilateral data on migrant stocks for the period 2005-11, collected by the Italian National In- stitute of Statistics (ISTAT) from the population registers of Italian municipalities, represent the stock of the foreign population resident in each Italian province, by citizenship, at the beginning of each year. Data on the age structure of the foreign resident population in each province are un- available. Instead, in the robustness section we use the total growth rate of the number of migrants over 2005-11 in each province as a measure of how recently established a migrant community is. The data refer to the number of official foreign residents, and do not account for undocumented migrants residing in Italy. For each recipient country, GDP at constant prices for the period 1950-2012 is drawn from the IMF World Economic Outlook database. The cyclical and trend components are extracted using the Hodrick-Prescott filter. Data on total population for the period 2005-11, as well as the level of financial development, proxied by domestic credit to private sector as a share of GDP, are drawn from the World Development Indicators database. The annual frequency of natural disasters is drawn from the International Emergency Disasters database (EM-DAT) built by the Centre for Research on the Epidemiology of Disasters.13 Data on armed conflicts are drawn from the UCDP/PRIO Armed Conflict Dataset (Themnér and Wallen- steen, 2013).14 The terms of trade are drawn from the IMF World Economic Outlook database. For each province, real GDP for the period 1995-2010 is drawn from ISTAT and the Istituto Guglielmo Tagliacarne.15 The cyclical and trend components are again extracted using the Hodrick- Prescott filter. Data on total provincial population for the period 2005-2011 are provided by ISTAT. The number of bank branches is provided by the Bank of Italy.16 4 Descriptive analysis 4.1 Remittances from Italy to Developing Countries Total remittances from Italy to developing countries doubled between 2005 and 2011, reaching almost e 7 billion, in line with the growth in the stock of foreign residents in Italy (Figure 2). After 2007, however, the growth rate of remittances slowed down significantly, reflecting the impact of the global financial crisis and of the euro area crisis on Italian output and unemployment. Indeed, remittances declined in 2010, although 2011 saw a rapid recovery, consistent with the global pattern of international remittances (Figure 1, right panel). The geographic distribution of remittances from Italy largely mimics the global distribution, 13 The data are accessible at www.cred.be/emdat/. A disaster is defined as a “situation or event, which overwhelms local capacity, necessitating a request to national or international level for external assistance”. Formally, an event is classified as a disaster whenever it fulfills at least one out of four selection criteria: ten or more people killed; 100 or more people affected, injured or homeless following the disaster; declaration of a state of emergency; or calls for international assistance. See www.emdat.be/criteria-and-definition. 14 The most recent version (4-2013) is available at www.pcr.uu.se/research/ucdp/datasets/ucdp_prio_armed _conflict_dataset/. 15 Data from ISTAT cover the period 1995-2007 while those from Istituto Guglielmo Tagliacarne go from 2007 to 2010. The two series hence overlap in two years, highlighting minor differences. 16 Data on the provincial presence of money-transfer operators, which could represent a better measure of the access of migrants to remittance-transfer services, are not publicly available for the period of the analysis. 9 Figure 2: Remittances outflows to developing countries and foreign residents in Italy 7000 40 Annual change in remittance outflows (%) Stock of foreign residents (thousands) Remittance outflows (millions, Euro) 6000 30 5000 20 4000 10 3000 0 2000 05 06 07 08 09 10 11 20 20 20 20 20 20 20 Year Remittance outflows Foreign residents % change in remittances Source: Bank of Italy and ISTAT. Figure 3: Remittances by region of destination 100%   100%   90%   90%   80%   80%   70%   70%   60%   60%   50%   50%   40%   40%   30%   30%   20%   20%   10%   10%   0%   0%   2005   2006   2007   2008   2009   2010   2011   2005   2006   2007   2008   2009   2010   2011   East  Asia  and  Pacific   Europe  and  Central  Asia   La@n  America  and  Caribbean   East  Asia  and  Pacific   Europe  and  Central  Asia   La@n  America  and  Caribbean   Middle  East  and  Northern  Africa   South  Asia   Sub-­‐Saharan  Africa   Middle  East  and  Northern  Africa   South  Asia   Sub-­‐Saharan  Africa   (a) Italy (b) World Source: Bank of Italy and World Bank Migration & Remittances Factbook 2011. suggesting that Italy represents a relevant and representative case study (Figure 3). The East Asia and Pacific region is the main recipient of both Italian and global remittances to developing countries. The region’s share of remittances from Italy increased by 10 percentage points between 2005 and 2011. Europe and Central Asia’s share of remittances from Italy is twice as high as its share of global remittances, reflecting the relatively large number of migrants from Eastern Europe in Italy. South Asia accounts for a rising share of remittances from both Italy and the world. In contrast, Sub-Saharan Africa accounts for a limited share of remittances. 10 Focusing on individual countries, China, Romania, and the Philippines were the major recipients of remittances from Italy in both 2005 and 2011.17 Transfers to Bangladesh, Sri Lanka and Georgia increased dramatically between 2005 and 2011. Colombia is the only country listed that registered a decrease in remittances from Italy over this period. The stock of resident migrants by country of origin is positively correlated with remittances to the relevant recipient country in 2011.18 4.2 Provincial heterogeneity Italy is highly heterogeneous in terms of economic and financial development. Figure 4 (Panels a and b) shows the geographic dispersion, over the sample period, of GDP per capita and financial development, as proxied by the functional distance of local credit markets. There is a strong divide between provinces in the North and in the South of the country. In addition, even within the North and the South, there are significant differences across provinces. The geographical dispersion of the degree of functional distance is greater than that of GDP, and follows a somewhat different pattern, so that financial and economic development cannot be considered two sides of the same coin (Alessandrini et al., 2010). Since we are interested in the impact of business cycle fluctuations on remittance outflows, it is important to consider not just the level but also the growth rate of provincial GDP. Figure 4 (Panel c) shows that the economic slowdown during 2008-10 affected different Italian provinces differently, without any clear North/South divide. Real GDP contracted by 6.8% (its largest drop) in the southern province of Matera and by 5.4% in the northern province of Turin, while it increased by 3.7% in the southern province of Catanzaro and by 3.8% in the northern province of Pavia. Migrants’ remittances also show a great degree of variability across Italian provinces (Figure 4, panel d). Between 2005 and 2010, in the median province remittance outflows on average accounted for almost 0.2% of GDP, ranging from less than 0.1% of GDP in ten Italian provinces to 0.3% of GDP in Genoa, 0.4% in Milan, 0.6% in Florence, 0.8% in Rome and 3.9%, the maximum, in Prato. Larger remittance outflows reflect primarily the presence of larger communities of migrants, and do not mirror the North/South economic divide. In sum, the richness of the dataset in terms of destination countries and the heterogeneity across source provinces make it possible to identify how remittances react to economic shocks in not just destination countries but also source economies. In addition, we can assess how a contemporaneous negative shock in source and destination economies affects overall remittance flows. Finally, the provincial heterogeneity in credit-market structure enables us to assess the effect of source-economy financial development on remittances. 17 The Italy-China remittance corridor was the single most important at the EU level in 2010. The Italy-Romania and Italy-Philippines corridors were among the ten biggest corridors from Eu- rope. See http://epp.eurostat.ec.europa.eu/statistics_explained/index.php/Migrant_remittance_and_cross- border_or_seasonal_compensation_transfer_statistics. 18 There are some outliers, notably China, whose share of total remittances significantly exceeds its share of total migrants. This may reflect an incorrect classification of some trade payments to China as remittances. When estimating our baseline model, we therefore exclude China. However, when discussing the robustness of the results, we also present estimates including China. 11 Figure 4: Remittances, economic and financial development across Italian provinces Quintiles Quintiles [12718,15712] [.841,2.52] (15712,20732] (2.52,3.07] (20732,24357] (3.07,3.6] (24357,26732] (3.6,4.46] (26732,35061] (4.46,5.92] (a) Per capita GDP (aver. 2005-10) (b) Functional distance (aver. 2005-10) Quintiles [-.0678,-.0215] Quintiles (-.0215,-.0102] [.000742,.00127] (-.0102,-.0000343] (.00127,.00165] (-.0000343,.0116] (.00165,.00203] (.0116,.0833] (.00203,.00242] No data (.00242,.0394] (c) GDP growth 2008-2010 (d) Remittances as a share of GDP (aver. 2005-10) Source: Bank of Italy, ISTAT and Istituto Tagliacarne. See Table 1 for definitions. 5 Results 5.1 Remittances as a countercyclical financial flow Remittances increase in response to cyclical output declines in the recipient country, CY CLEj . The response is statistically significant across all specifications in Table 3. The elasticities range between 1.8 and 2.5. This suggests that remittances can indeed play a significant role in stabilizing output during downturns, smoothing consumption, and mitigating the effects of macroeconomic fluctuations in developing countries. Even after controlling for output per capita, remittances increase significantly in response to dif- ferent factors of vulnerability in recipient countries. Remittances are approximately 3 percent larger when recipient countries experienced natural disasters (Yang, 2008; Mohapatra et al., 2012; Ebeke and Combes, 2013), and 7.6 percent larger in the case of a significant terms-of-trade deterioration (see Table 3, column 6 and 7, respectively).19 The outbreak of armed conflicts is not associated with 19 Such effects are computed by means of the following formula: (exp(βi ) − 1) ∗ 100, where βi is the estimated coefficient. 12 a significant impact on remittances, in line with Naudé and Bezuidenhout (2012).20 These results are consistent with a particularly altruistic response to major and/or clearly exogenous shocks. Remittances are also positively and significantly correlated with trend GDP per capita in recip- ient countries, T REN Dj , across all specifications. This supports the hypothesis that remittances are at least partly driven by investment motives. Moreover, remittances seem to be influenced by economic conditions in the migrants’ host province, consistent with Barajas et al. (2012), although the significance of the coefficient on CY CLEi varies across specifications. As discussed, recessions may have a significant impact on relatively low-skilled migrant workers reducing their capacity to remit. A one percentage point reduction in provincial GDP relative to its long-term trend, CY CLEi , on average translates into a 1.6 percentage point reduction in transfers from that province. Nevertheless, a twin shock to both source province and recipient country (equal to one standard deviation of, respectively, CY CLEi and CY CLEj ) boosts overall remittances, although the positive effect is small (0.03 percentage points; calculations based on estimates in Table 3, column 4). 5.2 Remittances and financial development Remittances are negatively correlated with financial development in recipient countries (Table 3, column 9). On average, a 1 percent reduction in the ratio of domestic credit to the private sector over GDP translates into a 0.7 percent increase in migrants’ transfers. This suggests that remittances may help overcome the financing constraints of households living in countries with less efficient financial institutions, in line with Giuliano and Ruiz-Arranz (2009). By contrast, remittances are positively correlated with financial development in the source province. Specifically, remittances decrease with the functional distance of the provincial banking system from the host province (column 11). The impact of local financial development on remittance outflows is economically significant: a reduction of the functional distance of the banking system from the highest level observed (in Messina) to the lowest level (Bolzano) is associated with a 0.9 percent increase in remittances.21 The coefficient on the number of bank branches per inhabitant, although positive, is not statistically significant (column 10). These results continue to hold, and the magnitude of the coefficients remains relatively un- changed, when controlling jointly for all three measures of source-province and recipient-country financial development (column 12), as well as all sources of vulnerability in recipient countries (column 13). 5.3 Other results Bilateral remittances are, not surprisingly, strongly correlated with the size of the relevant migrant community in the relevant province. The elasticity is generally around 0.5, and does not vary significantly across alternative specifications. 20 These results continue to hold when disasters are expressed in terms of their annual frequency, and when the terms of trade are entered as the total terms-of-trade index. In addition, when armed conflicts are expressed in terms of their annual frequency, they have a significant, positive impact on remittances. 21 Calculations based on data for 2010 and on the estimation results in Table 3, column 12. 13 The populations of both the host Italian province, P OPi , and the home country, P OPj , are positively correlated with remittances, even after controlling for migrant stocks, indicating the presence of scale effects. Larger host-province populations may reflect better employment oppor- tunities for migrants; larger home-country populations may reflect better investment opportunities for remitters. 5.4 Robustness checks This section tests the robustness of the findings. We first investigate the impact of changes in the sample composition (Table 4). We then allow for additional covariates (Table 5). Finally, we employ a different estimation method (Table 6). Different samples The sample underlying our earlier results excludes remittances to China: these appear to be an outlier, possibly reflecting poor data quality (see footnote 18). However, we also estimate the baseline model including China, since this is the largest recipient of remittances from Italy. The earlier results are largely confirmed (Table 4, column 1). We also restrict our sample to low-income countries, to analyze whether the countercyclical behavior of remittances depends on the recipient country’s income level. However, the coefficient on CY CLEj remains negative, significant and similar in size to the baseline model in Table 3, column 4 despite the big drop in the number of observations. Finally, we drop observations which may add noise and lead to small-sample bias. In particu- lar, we are concerned about province-country pairs that are characterized by a limited number of resident migrants. Here, remittances may be driven by idiosyncratic factors, which could be largely unrelated to macroeconomic conditions in the recipient country as a whole. To avoid this possi- bility, we exclude all observations where the migrant community numbers less than 100 migrants (M IGRijt < 100). Although this threshold reduces the original sample by almost three-quarters, the results from our baseline model remain valid (column 4).22 A related concern is that, in large recipient countries, macroeconomic conditions could be highly heterogeneous within the country. Further, migrant remittances may be largely driven by conditions within some region of the coun- try, rather than in the recipient country as a whole. Hence, we drop from the sample the recipient countries with the largest population (Bangladesh, Brazil, China, India, Indonesia, Nigeria and Pakistan)23 . Again, our general findings are confirmed. Additional covariates We next augment our baseline model with a set of additional regressors (Table 5). In the first column we control for foreign aid (measured as aid per capita); aid and remit- tances are substitutes, in line with Amuedo-Dorantes et al. (2007). Columns 2 to 4 add, one by one, the different measures of macroeconomic stability. Remittances are larger in countries with better macroeconomic and institutional conditions. The coefficients on the fiscal balance (as a ratio of GDP) and on the ratio of external debt over GDP are, respectively, positive and negative, although the latter is not statistically significant. The elasticity of remittances to CY CLEj , however, is 22 Results are robust to alternative specifications of the threshold up to M IGRijt < 500. 23 We drop countries with a total population above the 95th percentile of the sample distribution. 14 almost unaffected. The positive coefficient on the variable measuring constraints on the executive indicates that remittances are larger for countries with stronger institutions (Singh et al., 2011). Finally, the negative effect on remittances of the cyclical component of output in recipient countries holds even when controlling jointly for these covariates (column 5). Different estimator The last robustness exercise relates to the estimation method. Here, we drop country-province pair fixed effects, enabling us to identify the effect of variables which vary only across province-country pairs and not over time, such as the log of distance between province i and country j (DIST AN CEi,j )24 and the percentage growth in the bilateral stocks of migrants (∆M IGRAN T Si,j ). We use the Poisson Pseudo-Maximum Likelihood estimator (Silva and Tenreyro, 2006), which provides unbiased estimates in presence of heteroskedasticity and a high proportion of zeros in the dependent variable. The specification separately includes country and province fixed effects, instead of province-country pair fixed effects; however, we control for the potential correlation of errors at the bilateral level by clustering standard errors by country-province pairs. Our main results from the baseline and the augmented specification, and in particular the significant negative coefficient on CY CLEj , are largely confirmed even when including separate province and country fixed effects (Tables 6). The magnitude of the elasticities is also quite similar, lending support to the validity of our main findings. In addition, remittances are also positively correlated with the fraction of recent migrants (as proxied by ∆M IGRAN T Si,j , the growth rate of the migrant stock over the period 2003-11), with an elasticity of around 0.8. This may reflect either altruism or investment motives. Recent migrants are more likely than older migrants to have strong emotional ties to their home country, including to relatives and friends left behind. They may also need to repay family loans used to defray migration costs. At the same time, recent migrants are more likely to be aware of solid investment opportunities in their home country. They may also be more likely to return, increasing their incentive to invest, for instance in real estate. The distance to migrants’ home country, DIST AN CEi,j , is positively correlated with remit- tances. A priori, we would instead expect distance to be positively correlated with remittance transfer costs, and therefore negatively correlated with remittances. The result may arise because remittance data only takes into account official transactions. Migrants from nearby regions, such as Eastern Europe or the Mediterranean, may send remittances informally, for instance bringing them in person when they travel back home. In contrast, migrants from distant countries are relatively more likely to use formal, if expensive, remittance channels. 6 Conclusions This paper examines the role of remittances as a source of external finance that may help mitigate the macroeconomic and external vulnerabilities of developing countries. The global financial cri- sis and the volatility of commodity prices have hit developing countries severely; increasing their 24 Bilateral distances (in kilometers) between Italian provinces and recipient countries are calculated using the geographical coordinates of the administrative capitals of provinces and nations. 15 resilience to external shocks is a key objective of international financial institutions and policy- makers. Concessional lending and foreign aid are traditional ways to address vulnerabilities, but their effectiveness is highly disputed. The use of contingent financing instruments has so far been quite limited (International Monetary Fund and World Bank, 2011). Many countries are increas- ingly relying on international reserves as a stabilization tool, but this imposes high social and economic costs (Rodrik, 2006). Removing barriers to remittances may be a useful complement to such measures. We analyze how remittances are affected by structural characteristics, macroeconomic condi- tions, and adverse shocks in both source and recipient economies, using a novel, rich panel dataset on bilateral remittances from 103 Italian provinces to 87 developing countries over the period 2005- 2011. Remittances are negatively correlated with the business cycle in recipient countries, and increase especially strongly in response to adverse exogenous shocks, such as natural disasters or large declines in the terms of trade. In addition, remittances are positively correlated with potential GDP in recipient countries. These results are consistent with remittances being driven by both altruism and investment motives. Remittances are also positively correlated with economic conditions in the source province. Nevertheless, in the presence of similar negative shocks to both source and recipient economies, remittances remain countercyclical with respect to the recipient country. All these results are robust to potential reverse causality from remittances to macroeconomic conditions in the recipient country. Financial development in the source province is positively associated with remittances, likely because it reduces transaction costs and eases access to financial services for migrants. Conversely, financial development in the recipient country is negatively associated with remittances, suggest- ing that remittances help alleviate credit constraints. All these results hold even controlling for unobserved country-province pairs fixed effects, which capture time-invariant institutional and ge- ographical factors which may drive remittance flows. We conclude that remittances may indeed contribute significantly to macroeconomic stability in recipient countries. This effect should be considered together with their positive impact on poverty alleviation and growth, emphasized in the existing literature. From a policy perspective, our results suggest that one way to increase developing countries’ resilience to exogenous shocks is to increase their reliance on remittances, rather than on more volatile capital inflows. One – direct – way to achieve this goal is to reduce barrier to international migration (Clemens, 2011). A second way is to promote migrants’ access to financial services in host countries. Overall, these findings corroborate the efforts being carried out by international financial institutions and the private sector to reduce the costs of migrant remittances and to foster freer remittance flows. References Adams Jr., R. H. (2009). The determinants of international remittances in developing countries. World Development, 37 (1), 93–103. 16 — and Cuecuecha, A. (2010). Remittances, household expenditure and investment in Guatemala. World Development, 38 (11), 1626–1641. Agarwal, R. and Horowitz, A. W. (2002). Are international remittances altruism or insurance? Evidence from Guyana using multiple-migrant households. World Development, 30 (11), 2033– 2044. Akkoyunlu, S. and Kholodilin, K. A. (2008). A link between workers’ remittances and business cycles in Germany and Turkey. Emerging Markets Finance and Trade, 44 (5), 23–40. Albareto, G. and Mistrulli, P. (2011). Bridging the gap between migrants and the banking system. Temi di discussione (Economic working papers) 794, Bank of Italy, Economic Research and International Relations Area. Alessandrini, P., Presbitero, A. F. and Zazzaro, A. (2009). Banks, distances and firms’ financing constraints. Review of Finance, 13 (2), 261–307. —, — and — (2010). Bank Size or Distance: What Hampers Innovation Adoption by SMEs? Journal of Economic Geography, 10 (6), 845–881. Amuedo-Dorantes, C., Pozo, S. and Vargas-Silva, C. (2007). Remittances and the macroe- conomy: the case of small island developing States. Working Paper Series UNU-WIDER Research Paper, World Institute for Development Economic Research (UNU-WIDER). Arezki, R. and Brückner, M. (2012). Rainfall, financial development, and remittances: Evi- dence from Sub-Saharan Africa. Journal of International Economics, 87 (2), 377–385. Barajas, A., Chami, R., Ebeke, C. and Tapsoba, S. J.-A. (2012). Workers’ remittances: an overlooked channel of international business cycle transmission? IMF Working Papers 12/251, International Monetary Fund. Bazzi, S. and Clemens, M. (2013). Blunt instruments: avoiding common pitfalls in identifying the causes of economic growth. American Economic Journal: Macroeconomics, 5 (2), 152–186. Bettin, G., Lucchetti, R. and Zazzaro, A. (2012). Endogeneity and sample selection in a model for remittances. Journal of Development Economics, 99 (2), 370–384. — and Zazzaro, A. (2012). Remittances and financial development: substitutes or complements in economic growth? Bulletin of Economic Research, 64 (4), 509–536. Bonaccorsi di Patti, E. and Gobbi, G. (2001). The changing structure of local credit markets: are small businesses special? Journal of Banking & Finance, 25 (12), 2209–2237. Bugamelli, M. and Paternò, F. (2011). Output growth volatility and remittances. Economica, 78 (311), 480–500. Burger, M., van Oort, F. and Linders, G.-J. (2009). On the Specification of the Gravity Model of Trade: Zeros, Excess Zeros and Zero–inflated Estimation. Spatial Economic Analysis, 4 (2), 167–190. Chami, R., Barajas, A., Garg, A. and Fullenkamp, C. (2010). The global financial crisis and workers’ remittances to Africa: what’s the damage? IMF Working Papers 10/24, International Monetary Fund. —, Hakura, D. S. and Montiel, P. J. (2012). Do worker remittances reduce output volatility in developing countries? Journal of Globalization and Development, 3 (1), 1–25. Clemens, M. A. (2011). Economics and Emigration: Trillion-Dollar Bills on the Sidewalk? Journal 17 of Economic Perspectives, 25 (3), 83–106. Combes, J.-L., Ebeke, C., Etoundi, M. N. and Yogo, T. (2014). Are foreign aid and remit- tance inflows a hedge against food price shocks in developing countries? World Development, 54 (1), 81–98. Cooray, A. and Mallick, D. (2013). International business cycles and remittance flows. The B.E. Journal of Macroeconomics, 13 (1), 1–33. Dabla-Norris, E. and Bal-Gunduz, Y. (2012). Exogenous shocks and growth crises in low- income countries: a vulnerability index. IMF Working Papers 12/264, International Monetary Fund. de la Briere, B., Sadoulet, E., de Janvry, A. and Lambert, S. (2002). The roles of destination, gender, and household composition in explaining remittances: an analysis for the Dominican Sierra. Journal of Development Economics, 68, 309–328. Docquier, F., Rapoport, H. and Salomone, S. (2012). Remittances, migrants’ education and immigration policy: Theory and evidence from bilateral data. Regional Science and Urban Economics, 42 (5), 817–828. Durdu, C. B. and Sayan, S. (2010). Emerging market business cycles with remittance fluctua- tions. IMF Staff Papers, 57 (2), 303–325. Ebeke, C. and Combes, J.-L. (2013). Do remittances dampen the effect of natural disasters on output growth volatility in developing countries? Applied Economics, 45, 2241–2254. El-Sakka, M. I. T. and McNabb, R. (1999). The macroeconomic determinants of emigrant remittances. World Development, 27 (8), 1493–1502. Faini, R. (1994). Workers remittances and the real exchange rate: a quantitative framework. Journal of Population Economics, 7 (2), 235–45. — (2007). Remittances and the brain drain: do more skilled migrants remit more? World Bank Economic Review, 21 (2), 177–191. Frankel, J. (2011). Are bilateral remittances countercyclical? Open Economies Review, 22 (1), 1–16. Freund, C. and Spatafora, N. (2008). Remittances, transaction costs, and informality. Journal of Development Economics, 86 (2), 356–366. Giuliano, P. and Ruiz-Arranz, M. (2009). Remittances, financial development, and growth. Journal of Development Economics, 90 (1), 144–152. Gupta, S., Pattillo, C. A. and Wagh, S. (2009). Effect of remittances on poverty and financial development in Sub-Saharan Africa. World Development, 37 (1), 104–115. Hoddinott, J. (1994). A model of migration and remittances applied to Western Kenya. Oxford Economic Papers, 46, 459–476. International Monetary Fund and World Bank (2011). Managing volatility in low-income countries: the role and potential for contingent financial instruments. Tech. rep., IMF and The World Bank, Washington, DC. Jimenez-Martin, S., Jorgensen, N. and Labeaga, J. (2007). The volume and geography of remittances from the EU. Tech. rep., Report to the European Commission. Lucas, R. E. and Stark, O. (1985). Motivations to remit: evidence from Botswana. Journal of 18 Political Economy, 93 (5), 901–918. Lueth, E. and Ruiz-Arranz, M. (2008). Determinants of bilateral remittance flows. The B.E. Journal of Macroeconomics, 8 (1 (Topics)), Article 26. Mohapatra, S., Joseph, G. and Ratha, D. (2012). Remittances and natural disasters: ex-post response and contribution to ex-ante preparedness. Environment, Development and Sustainabil- ity, 14 (3), 365–387. Naudé, W. and Bezuidenhout, H. (2012). Remittances provide resilience against disasters in Africa. UNU-MERIT Working Paper Series 026, United Nations University, Maastricht Economic and social Research and training centre on Innovation and Technology. Niimi, Y., Caglar, O. and Schiff, M. (2010). Remittances and the brain drain: skilled migrants do remit less. Annales d’Economie et de Statistique, (97-98), 123–141. Noy, I. (2009). The macroeconomic consequences of disasters. Journal of Development Economics, 88 (2), 221–231. Osili, U. O. (2007). Remittances and savings from international migration: Theory and evidence using a matched sample. Journal of Development Economics, 83 (2), 446–465. Raddatz, C. (2007). Are external shocks responsible for the instability of output in low-income countries? Journal of Development Economics, 84 (1), 155–187. Rapoport, H. and Docquier, F. (2006). The economics of migrants’ remittances. In S. Kolm and J. Mercier Ythier (eds.), Handbook on the economics of giving, altruism and reciprocity, vol. 2, Elsevier, pp. 1135–1198. Rodrik, D. (2006). The social cost of foreign exchange reserves. International Economic Journal, 20 (3), 253–266. Roodman, D. (2009). A note on the theme of too many instruments. Oxford Bulletin of Economics and Statistics, 71 (1), 135–158. Ruiz, I. and Vargas-Silva, C. (2014). Remittances and the business cycle: A reliable relationship? Journal of Ethnic and Migration Studies, 40 (3), 456–474. Sayan, S. (2004). Guest workers’ remittances and output fluctuations in host and home countries: the case of remittances from Turkish workers in Germany. Emerging Markets Finance and Trade, 40 (6), 68–81. — (2006). Business cycles and workers’ remittances: how do migrant workers respond to cyclical movements of GDP at home? IMF Working Papers 06/52, International Monetary Fund. Schindler, M., Papageorgiou, C., Weisfeld, H., Pattillo, C. A., Spatafora, N. and Berg, A. (2011). Global shocks and their impact on low-income countries: lessons from the global financial crisis. IMF Working Papers 11/27, International Monetary Fund. Silva, J. M. C. S. and Tenreyro, S. (2006). The log of gravity. The Review of Economics and Statistics, 88 (4), 641–658. Singh, R. J., Haacker, M., Lee, K.-w. and Le Goff, M. (2011). Determinants and macroe- conomic impact of remittances in Sub-Saharan Africa. Journal of African Economies, 20 (2), 312–340. The World Bank (2013). Migration and Development Brief. Tech. Rep. 21, The World Bank. Themnér, L. and Wallensteen, P. (2013). Armed conflict, 1946-2012. Journal of Peace Re- 19 search, 50 (4). Vargas-Silva, C. (2008). Are remittances manna from heaven? A look at the business cycle properties of remittances. The North American Journal of Economics and Finance, 19 (3), 290– 303. Wooldridge, J. M. (2010). Econometric Analysis of Cross Section and Panel Data. The MIT Press, 2nd edn. Yang, D. (2008). Coping with disaster: the impact of hurricanes on international financial flows, 1970-2002. The B.E. Journal of Economic Analysis & Policy, 8 (1 (Advances)), Article 13. — and Choi, H. (2007). Are remittances insurance? Evidence from rainfall shocks in the Philip- pines. The World Bank Economic Review, 21 (2), 219–248. 20 Tables 21 Table 1: Variables: definition, sources and summary statistics Variable Definition Source Mean St. Dev. REMi,j,t Total official remittances at constant prices from province i to country Bank of Italy 652969.4 1.14E+07 j in year t CY CLEi,t Logarithm of actual GDP over potential GDP in province i in year ISTAT and Istituto 0.000 0.029 t; potential GDP is calculated by applying the H-P filter to the GDP Tagliacarne series at constant prices CY CLEj,t Logarithm of actual GDP over potential GDP in country j in year t; WEO 0.001 0.031 potential GDP is calculated by applying the H-P filter to the GDP series at constant prices T REN Di,t Logarithm of potential GDP in province i in year t, calculated by ap- ISTAT and Istituto 9.994 0.251 plying the H-P filter to the GDP series at constant prices Tagliacarne T REN Dj,t Logarithm of potential GDP in country j in year t, calculated by ap- WEO 10.837 2.493 plying the H-P filter to the GDP series at constant prices M IGRAN T Si,j,t Logarithm of 1 + the stock of migrants living in province i and coming ISTAT 3.220 2.096 from country j in year t W ARi,t Indicator = 1 if country j experienced armed conflicts in year t; both UCDP/PRIO Armed 0.170 0.376 22 interstate and intrastate conflicts are considered, in which the govern- Conflict Dataset ment of country j represents one of the warring parties DISAST ERi,t Indicator = 1 if country j experienced natural disasters in year t EM-DAT, CRED 0.742 0.437 T T SHOCKi, t Indicator = 1 if country j experienced a large negative shock to the WEO 0.048 0.214 terms of trade, defined as an observation falling in the lowest 5 percent of the distribution of the annual variation in the terms-of-trade index P OPi,t Logarithm of population in province i in year t ISTAT 13.072 0.734 P OPj,t Logarithm of population in country j in year t WDI 16.554 1.592 F IN DEVi,t Logarithm of the ratio of domestic credit to the private sector over WDI 3.257 0.798 GDP in country j in year t BAN Ki,t Logarithm of the number of bank branches per 10,000 inhabitant in Bank of Italy and IS- 1.744 0.370 province i in year t TAT F U N CT ION AL Logarithm of the ratio of the number of branches in province i weighted Bank of Italy 1.176 0.341 DIST AN CEi,t by the logarithm of 1 plus the kilometric distance between the province of the branch and the province where the parent bank is headquartered, over total branches in the province i in year t. (Continued) Table 1: Continued Variable Definition Source Mean St. Dev. AIDi,t Logarithm of official aid per capita received in country j in year t WDI 3.296 1.386 F ISCAL Fiscal balance (+ surplus/ - deficit) as a share of GDP in country WDI -0.015 0.043 BALAN CEi,t j in year t EXT ERN AL External debt stocks as a share of GDP in country j in year t WDI 0.465 0.591 23 DEBTi,t EXEC CON STi,t Constraint on the executives’ index in country j in year t (1 = Polity IV - Center for 4.733 1.904 unlimited authority; 7 = Executive parity or subordination) Systemic Peace ∆M IGRAN T Si,j Growth rate of the migrant stock M IGij over 2003-2011 ISTAT 0.037 0.107 DIST AN CEi,j Logarithm of the kilometric distance between province i and country Built-in STATA rou- 8.415 0.764 j tine Notes : WDI: World Development Indicators (The World Bank); WEO: World Economic Outlook (IMF). Table 2: List of countries Albania Gabon Morocco Algeria Gambia Mozambique Argentina Georgia Nicaragua Armenia Ghana Niger Azerbaijan Guatemala Nigeria Bangladesh Guinea Pakistan Belarus Guinea-Bissau Panama Benin Haiti Paraguay Bolivia Honduras Peru Bosnia and Herzegovina India Philippines Brazil Indonesia Russia Bulgaria Iran Senegal Burkina Faso Jamaica Seychelles Burundi Jordan Sierra Leone Cambodia Kazakhstan South Africa Cameroon Kenya Sri Lanka Cape Verde Kyrgyzstan Sudan Chad Laos Tanzania Chile Lebanon Thailand Colombia Libya Togo Congo Lithuania Tunisia Costa Rica Macedonia Turkey Cote d’Ivoire Madagascar Uganda Dominica Malaysia Ukraine Dominican Rep. Mali Uruguay Ecuador Mauritania Venezuela Egypt Mauritius Vietnam El Salvador Mexico Yemen Ethiopia Moldova Zambia 24 Table 3: Baseline and extended specification (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) MIGRANTSi,j,t 0.537*** 0.509*** 0.509*** 0.499*** 0.493*** 0.503*** 0.498*** 0.497*** 0.473*** 0.501*** 0.498*** 0.475*** 0.475*** [0.138] [0.125] [0.125] [0.123] [0.122] [0.123] [0.123] [0.122] [0.111] [0.125] [0.122] [0.113] [0.113] POPi,t 0.381 1.106* 1.106* 1.483** 1.462** 1.478** 1.486** 1.455* 1.760** 1.859* 0.856 1.531** 1.543** [0.300] [0.652] [0.652] [0.736] [0.741] [0.737] [0.735] [0.743] [0.690] [1.016] [0.570] [0.768] [0.766] POPj,t 2.913** 3.152** 3.152** 4.360*** 4.354*** 4.290*** 4.349*** 4.256*** 1.649* 4.328*** 4.380*** 1.637** 1.502* [1.275] [1.245] [1.245] [1.213] [1.213] [1.226] [1.218] [1.233] [0.847] [1.187] [1.137] [0.827] [0.808] CYCLEi,t 1.431* 1.431* 1.674* 1.650* 1.681* 1.675* 1.658* 1.941** 1.726* 0.824 1.131* 1.147* [0.858] [0.858] [0.917] [0.920] [0.922] [0.920] [0.930] [0.807] [0.953] [0.701] [0.633] [0.634] CYCLEj,t -1.787** -1.787** -2.470*** -2.395*** -2.571*** -2.492*** -2.534*** -1.780** -2.472*** -2.485*** -1.798*** -1.857*** [0.731] [0.731] [0.694] [0.702] [0.691] [0.694] [0.697] [0.691] [0.695] [0.693] [0.690] [0.702] TRENDi,t 0.587 0.575 0.567 0.590 0.554 0.915* 0.549 0.412 0.702 0.712 [0.542] [0.543] [0.542] [0.543] [0.545] [0.498] [0.548] [0.526] [0.482] [0.475] TRENDj,t 1.906*** 1.946*** 1.921*** 1.909*** 1.977*** 1.917*** 1.904*** 1.954*** 1.966*** 1.978*** 25 [0.519] [0.524] [0.518] [0.518] [0.524] [0.510] [0.519] [0.516] [0.511] [0.497] WARj,t -0.067 -0.078 0.003 [0.050] [0.050] [0.052] DISASTERj,t 0.031* 0.038** 0.011 [0.018] [0.017] [0.018] TT SHOCKj,t 0.073** 0.101*** 0.177*** [0.032] [0.034] [0.039] FINDEVj,t -0.680*** -0.681*** -0.702*** [0.148] [0.136] [0.138] BANKi,t 0.400 0.431 0.430 [0.463] [0.447] [0.449] FUNCTIONAL DISTANCEi,t -0.454** -0.466*** -0.467*** [0.210] [0.177] [0.177] Observations 34,494 34,494 34,494 34,494 34,494 34,494 34,494 34,494 34,494 34,494 34,494 34,494 34,494 Number of pair 5,881 5,881 5,881 5,881 5,881 5,881 5,881 5,881 5,881 5,881 5,881 5,881 5,881 Notes : The table reports regression coefficients and (in brackets) the associated robust standard errors clustered by country-province pairs. * significant at 10%; ** significant at 5%; *** significant at 1%. Estimations are carried out by using the Poisson Fixed Effects estimator. The dependent variable is the value of total official remittances at constant prices from province i to country j in year t (REMi,j,t ). A constant and a set of province-country pairs (i, j ) and year (t) dummies are included. Table 4: Robustness: sample definition (1) (2) (3) (4) Including Low Income Large migrant No large China Countries communities recipients CYCLEi,t 1.526 0.290 1.712* 1.642 [1.005] [1.215] [0.971] [1.016] CYCLEj,t -2.505*** -2.033* -2.914*** -2.469*** [0.924] [1.146] [0.711] [0.745] TRENDi,t 0.234 3.361** 0.724 0.643 [0.671] [1.696] [0.570] [0.571] TRENDj,t 1.732** -1.345* 1.931*** 2.563*** [0.766] [0.698] [0.607] [0.612] MIGRANTSi,j,t 0.120** 0.445*** 0.729*** 0.523*** [0.055] [0.095] [0.193] [0.140] POPi,t 0.329 1.816** 1.554** 1.524* [0.871] [0.854] [0.766] [0.805] POPj,t 4.484*** -17.825*** 4.888*** 4.583*** [1.474] [1.652] [1.405] [1.446] Observations 38,851 6,686 10,127 31,939 Number of pairs 6,563 1,175 1,845 5,501 Notes : The table reports regression coefficients and (in brackets) the associated robust standard errors clustered by country-province pairs. * significant at 10% level; ** significant at 5% level; *** significant at 1% level. Estimations are carried out by using the Poisson Fixed Effects estimator. The dependent variable is the value of total official remittances at constant prices from province i to country j in year t (REMi,j,t ). A constant and a set of province- country pairs (i, j ) and year (t) indicators are included. 26 Table 5: Robustness: additional covariates (1) (2) (3) (4) (5) CYCLEi,t 1.266* 1.562 1.699* 1.813* 1.878 [0.705] [1.051] [0.916] [1.088] [1.471] CYCLEj,t -2.609*** -3.668*** -2.883*** -2.922*** -3.371*** [0.399] [0.922] [0.761] [0.483] [1.243] TRENDi,t 0.588 0.508 0.663 1.364** 1.318 [0.564] [0.647] [0.527] [0.682] [0.855] TRENDj,t 1.379*** 1.563* 2.038*** 0.417 1.820*** [0.469] [0.864] [0.570] [0.549] [0.598] MIGRANTSi,j,t 0.661*** 0.574*** 0.519*** 0.337*** 0.362*** [0.108] [0.153] [0.131] [0.071] [0.078] POPi,t 1.285** 1.528* 1.533** 1.306 1.729 [0.616] [0.917] [0.757] [0.941] [1.262] POPj,t 2.638*** 5.061*** 3.774*** 1.277 -6.691** [0.947] [1.832] [0.971] [0.846] [2.663] AIDj,t -0.108*** -0.218*** [0.013] [0.037] FISCAL BALANCEj,t 2.521*** 1.344** [0.753] [0.641] EXTERNAL DEBTj,t -0.212 -0.664 [0.313] [0.525] EXEC CONSTj,t 0.117*** 0.282*** [0.029] [0.044] Observations 32,055 24,152 34,127 15,996 10,706 Number of pairs 5,496 4,402 5,800 2,701 1,903 Notes : The table reports regression coefficients and (in brackets) the associated robust standard errors clustered by country-province pairs. * significant at 10% level; ** significant at 5% level; *** significant at 1% level. Estimations are carried out by using the Poisson Fixed Effects estimator. The dependent variable is the value of total official remittances at constant prices from province i to country j in year t (REMi,j,t ). A constant and a set of province- country pairs (i, j ) and year (t) indicators are included. 27 Table 6: Robustness: PPML estimator – baseline and extended specification (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) CYCLEi,t 1.191 1.196 1.363 1.345 1.378 1.363 1.356 1.675** 1.426 0.495 0.825 0.845 [0.885] [0.889] [0.934] [0.936] [0.940] [0.937] [0.946] [0.832] [0.974] [0.720] [0.658] [0.658] CYCLEj,t -1.666** -1.601** -2.245*** -2.184*** -2.373*** -2.269*** -2.349*** -1.539** -2.240*** -2.264*** -1.544** -1.640** [0.738] [0.737] [0.694] [0.701] [0.688] [0.693] [0.693] [0.685] [0.692] [0.693] [0.683] [0.694] TRENDi,t 0.525 0.511 0.510 0.532 0.501 0.816 0.482 0.344 0.584 0.609 [0.575] [0.578] [0.577] [0.577] [0.583] [0.526] [0.582] [0.551] [0.502] [0.498] TRENDj,t 1.781*** 1.814*** 1.802*** 1.785*** 1.850*** 1.787*** 1.779*** 1.796*** 1.799*** 1.807*** [0.541] [0.546] [0.541] [0.541] [0.547] [0.535] [0.543] [0.544] [0.541] [0.526] MIGRANTSi,j,t 0.920*** 0.919*** 0.911*** 0.911*** 0.911*** 0.911*** 0.911*** 0.911*** 0.911*** 0.911*** 0.911*** 0.911*** 0.911*** [0.038] [0.037] [0.037] [0.037] [0.037] [0.037] [0.037] [0.037] [0.036] [0.037] [0.037] [0.036] [0.036] ∆ MIGRANTSi,j 0.850*** 0.852*** 0.852*** 0.854*** 0.852*** 0.854*** 0.859*** 0.852*** 0.853*** 0.860*** 0.861*** [0.132] [0.133] [0.133] [0.133] [0.133] [0.133] [0.133] [0.133] [0.133] [0.134] [0.134] POPi,t 0.280 0.877 0.869 1.152 1.134 1.152 1.155 1.133 1.466** 1.570 0.511 1.281 1.297 [0.311] [0.671] [0.674] [0.737] [0.740] [0.737] [0.736] [0.742] [0.703] [1.019] [0.574] [0.798] [0.796] POPj,t 3.570*** 3.905*** 4.118*** 5.251*** 5.259*** 5.153*** 5.240*** 5.133*** 2.558*** 5.208*** 5.355*** 2.582*** 2.413*** [1.108] [1.160] [1.188] [1.160] [1.161] [1.179] [1.165] [1.187] [0.863] [1.130] [1.131] [0.858] [0.838] 28 DISTANCEi,j 0.276* 0.276* 0.319** 0.316** 0.316** 0.316** 0.316** 0.316** 0.319** 0.316** 0.314** 0.315** 0.315** [0.157] [0.157] [0.153] [0.153] [0.153] [0.153] [0.153] [0.153] [0.153] [0.153] [0.152] [0.152] [0.152] WARj,t -0.055 -0.067 0.018 [0.051] [0.051] [0.053] DISASTERj,t 0.039** 0.045** 0.018 [0.019] [0.018] [0.019] TT SHOCKj,t 0.078** 0.104*** 0.183*** [0.032] [0.035] [0.040] FINDEVj,t -0.693*** -0.703*** -0.728*** [0.146] [0.137] [0.141] BANKi,t 0.442 0.518 0.512 [0.467] [0.457] [0.459] FUNCTIONAL DISTANCEi,t -0.467** -0.499*** -0.501*** [0.206] [0.178] [0.178] Observations 34,494 34,494 34,494 34,494 34,494 34,494 34,494 34,494 34,494 34,494 34,494 34,494 34,494 Number of pair 5,881 5,881 5,881 5,881 5,881 5,881 5,881 5,881 5,881 5,881 5,881 5,881 5,881 Notes : The table reports regression coefficients and (in brackets) the associated robust standard errors clustered by country-province pairs. * significant at 10%; ** significant at 5%; *** significant at 1%. Estimations are carried out by using the Poisson Pseudo Maximum Likelihood (PPML) estimator (Silva and Tenreyro, 2006). The dependent variable is the value of total official remittances at constant prices from province i to country j in year t (REMi,j,t ). A constant and a set of province (i), country (j ) and year (t) dummies are included.