Michael Barrios Batu. A Thesis presented to The University of Guelph

Transcription

1 Three Essays on Remittance Effectiveness by Michael Barrios Batu A Thesis presented to The University of Guelph In partial fulfilment of requirements for the degree of Doctor of Philosophy in Economics Guelph, Ontario, Canada c Michael Barrios Batu, July, 2015

2 ABSTRACT THREE ESSAYS ON REMITTANCE EFFECTIVENESS Michael Barrios Batu University of Guelph, 2015 Advisors: Laurent Cellarier, Stephen Kosempel Latest estimates from the World Bank has pegged remittance flows around $404 billion in 2013 and is expected to reach $516 billion by 2016 (The World Bank, 2014). Remittance has surpassed foreign direct investment, portfolio flows from financial markets, and official development assistance in many developing countries. The common wisdom is that if remittances are invested, they should contribute to output growth. Surprisingly, most scholars find the absence of any robust positive relationship between economic growth and remittances. In this dissertation I propose three reasons as to why it is difficult to find this relationship. In the first essay, I highlight the role of migration as an important channel through which remittances can influence economic growth. The effect remittance have on migration decision is ambiguous and it depends on two competing forces: the benefit of giving and the benefit of receiving remittances. The basic idea is that increased remittances affects the composition of the labor force of the recipient country and over time it will lead to an increase in savings, capital, and consequently, output. In the second essay, a model of an artificial economy which has the structure of the canonical small open economy real business cycle model augmented with stochastic remit-

3 tance shocks was developed. The model was calibrated using data for remittance recipient countries and the calibration exercise reveals that the response of output to remittance shocks is small in comparison to other shocks. The model predicts that temporary inflows of worker remittances positively affect GDP per capita while a permanent increase of remittances does not. In the third essay I built and calibrated a two-sector real business cycle model to analyze the effects of remittances on human capital accumulation in developing countries. The model has two key results: First, I find that an unanticipated temporary increase in remittances leads to an initial decline in labor supply and an increase in consumption demand. Second, the increase in remittances leads to an increase in education spending which raises human capital and output in the short run. Increasing disposable income through remittances may increase investment in schooling and relaxes the budget constraint.

4 ACKNOWLEDGEMENTS First and foremost, I would like to thank God Almighty, whose many blessings have made me who I am today. Only due to His blessings I could finish my dissertation. I would like to thank my mother, siblings, nephews and nieces for their love and encouragement. I wish to express my sincere thanks to Dr. Laurent Cellarier, who has been a great supervisor. His guidance and encouragement helped me grow personally and professionally. I will always be indebted to him. My sincere thanks also go to Dr. Stephen Kosempel, Dr. Thanasis Stengos, Dr. Kurt Annen, Dr. Alex Maynard, and Dr. Asha Sadanand for their advice, insightful comments and discussions over the years. Thanks also to Dr. Jeff Nugent for his valuable comments for the improvement of this thesis. I take this opportunity to express gratitude to Dr. James Atsu Amegashie who has been a great mentor and friend. I also thank Dr. Mike Hoy, Dr. Chris McKenna, Dr. Patrick Martin, Nancy Bower, Lois Lamble, Sandra Campbell, and Greg Radovan at the Department of Economics and Finance for their friendship and support. My thanks also go to my friends Esmond Lun, Katherine Dare, Diana Alessandrini, Jennifer Teng, Fraser Summerfield, Haigang Wang, Joniada Milla, Robert Bright, and Marcel Oestreich who made this experience as enjoyable as possible. iv

5 Table of Contents List of Tables List of Figures viii ix 1 Manna from Heaven? International Worker Remittances and Output Gaps Across Countries Introduction Brief literature review The model Household preferences Technology Endogenizing migration Equilibrium Autarky equilibrium (ε = 0 and γ = 0) Open economy equilibrium with remittances (ε > 0 and γ > 0) Dynamics and numerical experiments Comparative statics on γ Comparative statics on σ Empirical analysis Testable hypotheses Identification Bilateral remittances Data Empirical results Conclusion Appendix 1: Proofs Appendix 2: Countries in the Sample v

6 2 International Worker Remittances and Economic Growth in a Real Business Cycle Framework Introduction Brief Background Literature Canonical Small Open Economy RBC Model with Remittances Remittances in the Long Run: A Special Case with Inelastic Labor Supply Calibration and Simulations Data and Business Cycles Calibration Permanent changes in remittance flows Temporary changes in remittance flows Sensitivity Analysis Cross-Country Empirical Evidence Econometric Specification Empirical Estimates Conclusion International Worker Remittances and Human Capital Formation Across Countries Introduction Brief literature review The model Data and calibration The Philippines at a glance Parameters Impulse response analysis Sensitivity analysis Econometric evidence Regression models Data and sources Instruments Regression results Conclusion Conclusion 103 References 105 vi

7 List of Tables 1.1 Data description and sources Impact of bilateral remittances on migrant stocks Impact of bilateral remittances on output gaps across countries Business cycle statistics from data for remittance recipients for the period Model parameter values Permanent changes in share of remittance to output per capita Model economy business cycle statistics Remittances and Economic Growth: Panel Estimation ( ) a Remittances and Economic Growth for Recipients with ρ d 0.67: Panel Estimation ( ) a Appendix: List of recipient countries and descriptive statistics Appendix: List of recipient countries and descriptive statistics, cont Selected remittance recipients: business cycle statistics from Data and the benchmark model Descriptive statistics Remittances and education attainment cross section regressions Remittances and education attainment panel data regressions vii

8 List of Figures 1.1 Remittances, foreign direct investment and official development assistance Determination of steady state equilibrium physical capital and output per worker differentials A situation where an increase in γ widens the gap in capital between h and f A situation where an increase in γ helps close the gap in capital between h and f Net Official Development Assistance, Remittances and Foreign Direct Investments, averages in current US dollars (from World Development Indicators) Effects of a permanent increase in international worker remittance flows Effects of a temporary increase in international worker remittance flows Impulse response to a one-percent remittance shock Composition of various inflows to selected countries, Impulse response to a positive one-percent remittance shock Sensitivity analysis for remittance parameter d viii

9 Chapter 1 Manna from Heaven? International Worker Remittances and Output Gaps Across Countries 1.1 Introduction International migration coupled with the advent of convenient money transfer services have created significant impacts on developing countries, which account for many of these migrants and receive a substantial portion of remittances sent worldwide. Remittances, generally defined as transfers of money from migrants to the family members they leave behind, often sent a few hundred dollars at a time, nonetheless add up to billions of dollars annually. Latest estimates from the World Bank has pegged remittance flows around $404 billion in 2013 and is expected reach $516 billion by 2016 (The World Bank, 2014). Remittance has surpassed foreign direct investment, portfolio flows from financial markets, and official development assistance in many developing countries. As shown in Figure 1

10 1, remittances has become an important source of funds for many developing countries. Moreover, some countries total remittance receipts amount to a substantial portion of their imports and a nontrivial fraction of GDP (Barajas et al., 2008). Given their volume and importance, can remittances be used to help poor countries catch up with their wealthy counterparts? Recent studies were able to address this question mostly through reduced-form cross-country growth regressions. The common wisdom is that if remittances are invested, they contribute to output growth thereby allowing the recipient country to achieve convergence in output per capita with rich countries. Surprisingly, most scholars find the absence of any robust positive relationship between economic growth and remittances: 1 Decades of private income transfers - remittances - have contributed little to economic growth in remittance-receiving economies... the most persuasive evidence in support of this finding is the lack of a single example of a remittances success story: a country in which remittances-led growth contributed significantly to its development... no nation can credibly claim that remittances have funded or catalyzed significant economic development. (Barajas et al., 2009) Part of the reason for this finding may lie in the difficulty of disentangling the complicated links between remittances and economic growth. 2 For example, identifying the direction of causal links between remittances and economic growth may not be fully solvable by using instrumental variables to control for endogeneity and reverse causation. Another reason, Clemens and McKenzie (2014) suggests, is that misspecification of reduced-form models makes it hard to detect any relationship between remittances and economic growth. Important channels that influence economic growth through remittances may have been 1 For instance see Barajas et al. (2009), Chami et al. (2003), Chami et al. (2006), and Barajas et al. (2008). 2 Clemens and McKenzie (2014) provides three possible answers: First, growth in remittances may actually be illusory because of measurement errors and not changes in real financial flows. Second, data on remittances is very limited such that cross-country regressions would have too little power to detect their effects on growth. And third, the effect of remittances on GDP growth therefore depends upon how the money is spent by the recipients. 2

11 Figure 1.1: Remittances, foreign direct investment and official development assistance overlooked as well. For instance, migratory flows influence both output and the amount of remittance flows to recipient countries. In this paper I try to address this question by modeling the motivations to migrate and remit via a two-country overlapping generations model with endogenous growth and spillovers. In this framework I was able to show that migration along with generosity of migrants through remittances may actually help reduce the gap in per capita incomes between a rich and a poor country in the steady state. Even if the differences in productivity between the two countries are permanent I find that, through remittances, a poor country can build enough capital internally to allow it to catch up with the rich country. This result could be considered the core of my contribution. Nevertheless, I have also tried to make a contribution in the literature through other ways: 3

12 1. Micro-foundations. The main feature that distinguishes my work with the current literature on remittance effectiveness is solid micro-foundations. In particular, a key feature of my approach is the use of a preference parameter to capture the psychological cost of a joint decision to migrate and remit goods to individuals left behind. The competing effects of benefits and costs of staying and becoming a recipient or migrating and becoming a sender of remittances turns out to be a crucial driver of my results. 2. Endogenous growth. I exploit a specific form of production function which exhibits positive spillovers of investments in physical capital on labor productivity. In this set up, not only is growth explained by the model, one can also investigate the effect of migration and remittances have on labor productivity and, consequently, output. 3. Econometric evidence. My theoretical model highlights the complex relationship between output differentials across countries, migratory flows and remittances received. In contrast to recent studies that focus on recipient countries, my econometric analysis makes use of bilateral or country-pair data with OECD countries as the remittance sources and migrant hosts. This way I can fully exploit the heterogeneity in output, remittance inflows and migratory flows that exists across countries. Another advantage of using bilateral data is the big number of observations which help in the estimations. Before presenting the model, some of its main results will be previewed. First, I argue that in transition to the steady state, the generosity of households should generate a positive effect to the migrant-sending country in terms of higher growth rates in physical capital. Remittances from migrant to non-migrant households enter as non-labor income which contributes to higher savings, and consequently investment. Under certain condi- 4

13 tions, lower migration rates are observed when remittances are high. Second, I theoretically analyze the influence of migrant generosity on growth and, consequently, on convergence. I find that increased migrant generosity, which results in higher remittances, may propel a country to higher levels of per capita incomes or potentially lead it to lower levels of output. Catching up is likely to occur if the benefits of migrating and sending remittances outweigh the benefits of staying and becoming a recipient. My principal conclusion is that, at least theoretically, contrary to most open economy neoclassical models, migration does not always lead to an improvement in standard of living across countries. This paper also empirically tests the predictions of the model using bilateral migratory flow data to and remittances from OECD countries. 3 My approach differs from extant literature in the sense that I exploit the two-way nature of migration and remittances. I investigate the impact of bilateral remittances to output differentials in a bilateral setting (two-way) instead of focusing on recipient countries (one-way). Motivated by predictions from my theoretical model, the empirical results show that the relationship of remittances and migration to output differentials is very significant. In particular, I find that higher remittances leads to higher migration, which in turn lead to a narrowing of the output per capita differentials. The extent of this narrowing, however, is rather modest. The rest of this paper is organized as follows: Section 2 presents a brief literature review. Section 3 the core of the paper, presents the model of migration with generous households. In this section I analyze the growth rates in physical capital and determination of steady state. Comparative statics analysis of the effects of increased migrant generosity follows. Section 4 provides the numerical simulations as well as an exploration of transitional dynamics of the model. Section 5 presents the empirical results. Section 6 concludes. 3 The use of bilateral migratory flow data is relatively new in the migration literature in light of recent advances in data collection for inter-country migratory stocks. 5

14 1.2 Brief literature review The literature on remittance effectiveness can be mainly divided into two: the effects on poverty and inequality; and the influences upon investment, macroeconomic expansion and stability. In this brief literature review I focus on the latter. It is increasingly accepted that remittances can generate output growth either by increasing consumption or by increasing investment. Given the growing importance of remittances, they could potentially become an important tool for economic development, especially if they can be channeled into productive investment (Ratha, 2007). The World Bank s Global Development Finance Report cites the case of high savings out of remittances in Pakistan (The World Bank, 2003). There are also signs that remittances may have accelerated investment in Morocco and perhaps in India (Lucas, 2005). Using data for seven Mediterranean countries, Glytsos (2002), finds that investment rises with remittances in six out of seven countries, and in four of these investment rises by more the initial amount remitted. In the same vein, Leon-Ledesma and Piracha (2001) find similar evidence of positive association between remittances and investment for eleven countries in Eastern Europe. Remittances was found to have a positive effect on cattle accumulation and crop productivity for families of workers in South African mines (Lucas, 1985). There are few studies which found a positive link between remittances and economic growth. Using dynamic panel regressions, Catrinescu et al. (2009) found that remittances exert a positive impact on macroeconomic growth when controlling for endogeneity. They conclude that remittances will be more likely to contribute to longer-term growth in countries with higher quality political and economic policies and institutions. Faini (2002) suggests that in order for the full impact of remittances to be realized a sound policy environment is required. He also found a positive impact of remittances on output growth. In another work, Faini (2006), using instrumental variables estimation, found a positive rela- 6

15 tionship between total remittance and economic growth. However, the estimated coefficient in his model is not statistically significant. Despite the positive contribution of remittances to household welfare, most scholars have found that international worker remittances do not have a positive effect on economic growth. Chami et al. (2003), in a study covering up to 113 countries over the period 1970 to 1998, found that international remittances actually have a negative and significant effect on economic growth. Using a variety of fixed effects models, the authors find a negative and significant relationship between international remittances and economic growth for different groups of countries over various sets of years. On the basis of this finding, they conclude that remittances do not serve as capital for economic development, but rather as a type of compensation for countries with poor economic outcomes. The International Monetary Fund (2005) performed cross-country growth regressions with specifications similar to those in Chami et al. (2003) on a set of 101 countries measured over the 1970 to 2003 period. The IMF study used two instruments for remittances: distance between the migrants home and main destination country, and a dummy measuring whether the home and main destination country shared a common language. They found no statistically significant effect of total remittances on economic growth. Giuliano and Ruiz-Arranz (2005), using fixed effects and system GMM models, did not find total remittances to be significantly related to growth. 4 In a similar study covering up to 101 countries for the period 1970 to 2003, Spatafora (2005) comes to slightly different conclusions. Specifically, the author finds no statistically significant link between international remittances and per capita output growth. The author also finds no significant link between remittances and investment (investment/gdp), or between remittances and education. 4 Giuliano and Ruiz-Arranz (2005) also tried other specifications and found that total remittances appeared to have positive effects on growth only in countries with small financial sectors where presumably credit constraints would be more pervasive. 7

16 It is often difficult to disentangle labor mobility from remittances in order to fully understand its effectiveness to the economy. A prerequisite in understanding remittance effectiveness is to initially look at labor mobility and its effects to economic growth, development and theory on convergence. In fact, there is already a large number of studies found in the economic development literature which tries to associate economic development with factor mobility. For instance, neoclassical growth models predict that migration from low to high income countries should make the former grow more rapidly in transition to the long run. Migration will enhance the convergence process until factor payments and income levels between countries converge, and migration should dissipate over time (Reichlin and Rustichini, 1998). However, this conjecture is controversial because several high income countries (i.e. United States and Canada) have been receiving migrants for over a century now and migration persists in these countries, as well as remittances. In recent decades, attention has shifted to endogenous growth models with a view to understand the connections between economic growth and migration, as well as to address contradictory findings in the theoretical and empirical literature. For example, Faini (1996) provided a microeconomic foundation for the decision of people to permanently migrate. He constructed a two-sector model with increasing returns to scale and finds that with migration, convergence may occur for some parameter values. Larramona and Sanso (2006) studied temporary migration in the context of endogenous growth model with capital accumulation and spillovers. They find that while migration positively affects the sending country in the steady state, the differences existing between two countries, and consequently migration, persist. A more recent contribution by Ukueva (2011) explored the effect of permanent migration and remittances on a small, open, migrant-sending country in an endogenous growth model with technology transfers. She found that remittances allow people to share the benefits of technological advances developed elsewhere and dampen 8

17 the negative impact of migration. 1.3 The model Household preferences I consider a simple, two-country overlapping generations framework. For each country, households are assumed to live in two periods where they are workers in the first and retirees in the second. In the first period, they decide whether to work in their own country or to temporarily migrate. In the second, migrants return to their home country to retire. In what follows, we focus our attention on households born in their home country (h) making the decision to remain or to migrate to a foreign country (f). Following Faini (1996), let migration incur some welfare cost represented by θ, which for analytical convenience is a random variable with a Pareto distribution. For households born in country h it will have the following properties: θ > 1 if j = h = 1 if j = f, g(θ) = ( ε/θ ε+1), θ [1, ), ε > 0, j = h, f, where j is the country, ε determines the shape of the distribution, and g(θ) is the density function. 5 Introducing migration costs allows one to represent heterogeneous households (migrant and non-migrant) in each country according to the intensity of their preference. 6 Migrant households consume the quantity c f t, send transfers d f t to households they leave behind, and save w f c f t d f t in period 1. In period 2, they retire and return to their country of birth, give d h t+1 to non-migrant households and consume all their assets leaving 5 The parameter θ can also capture fixed costs of migrating to another country. 6 The assumption on θ implies a strong home preference for migrant households. Consequently, giving at home is more preferred than giving abroad. 9

18 no bequests. 7 In the spirit of Andreoni (1990) and others, think of the transfers d f t and d h t+1 as remittances migrant households give away out of generosity. 8 It should be emphasized here that, in a strict sense, only the first period transfer can be considered as international remittance as the transfer was made while the migrant is in the foreign country. Formally, a household in h planning to work in country f solves the problem: max (1 γ)[ln(c f c f t, ch t+1, df t, t ) + β ln(θ h c h t+1)] + γ[ln(d f t ) + β ln(θ h d h t+1)] dh t+1 subject to c f t + 1 c h Rt+1 t+1 + d f t + 1 d h Rt+1 t+1 = w f t where β is the discount rate, γ is the parameter which captures generosity or joy of giving, R is the interest rate factor and w f t is the wage rate in country f. Migrant households can allocate a fraction of their saving in the home country and this fraction is given by σ [0, 1]. Hence, the interest rate R for a given time t is defined as: R t σr h t + (1 σ)r f t For now the analysis will be for the case of σ = 1. Households who decide not to migrate receive income in the first period which consists of the wage they receive from working and the present value of transfers received from migrant households. I assume that the present value of aggregate transfers are equally divided among non-migrant households. Non-migrants consume the quantity c h t and save 7 A model with migrant (temporary) workers is more attractive than a model where migrants permanently stay in the host country in a sense that the former is empirically supported. For instance in 2010, according to OECD s international migration database there were only 66,989 permanent migrants in the US out of the total of 1,130,818 migrants admitted that year. While illegal migration exists that may confound the data, only around 5% are permanent migrants in the US and this is the trend for most of other OECD host countries. 8 This set up, first introduced by Ukueva (2011), assumes that all migrant households possess the same degree of motivation to give or joy of giving motives. 10

19 s h t in period 1. In period 2, they also consume all their assets and leave no bequests. A non-migrant household in h solves: max ln(θ h c h c h t, t ) + β ln(θ h c h t+1) ch t+1 subject to c h t + 1 c h Rt+1 h t+1 = wt h + m t 1 m t ( d f t + 1 ) d h Rt+1 h t+1 where m t is the proportion of migrants. At the optimum, I can express the optimal savings decision for each household as: s h t = ( ) [ β wt h β m t 1 m t ( d f t + 1 )] d h Rt+1 h t+1 and s f t = ( ) β w f t (1.1) 1 + β Technology Our discussion on firms and technology follows closely Kemnitz (2001). Consider a large number of infinitely-lived firms in each country which produce an identical good according to the constant returns to scale production function: ( Q j t = F (A j tl j t, K j K j ) t t ) = F 1, A j tl j A j tl j t F 1, F 2 > 0, F 11, F 22 < 0, F 12 > 0 t where A j is a technological labor-augmenting factor exogenous to each firm, K j t is physical capital and L j t is labor. At the aggregate level A j is endogenously determined, and reflects the positive spillover that investment in physical capital has on labour productivity: A j t = 1 α j K j t L j t = 1 α j kj t, α j > 0, k j t = Kj t L j t where α j is a productivity parameter and k j t is the capital to labor ratio. With the spillover 11

20 effect, the aggregate production function can be written as: q j t = Qj t L j t ( K j ) t = F 1, A j tl j A j t = q(α j ) 1 t α j kj t, where q(α j ) = F (1, α j ). According to the assumptions on F, q(α j )/α j > q (α j ) > 0 and q (α j ) < 0. Perfect competition among firms drives factor prices to correspond to their marginal productivities: ˆR j t = q (α j ) (1.2) ŵ j t = (q(α j )/α j q (α j ))k j t (1.3) Returns to capital is constant and independent of the capital to labor ratio. On the other hand, wage income is influenced positively by k j t. The two countries do not necessarily have the same technology. Let q q be the schedule which describes the gap in output per worker: q h t q f t = ( q(α h )/α h ) k h t q(α f )/α f k f t = Λ(α h, α f ) kh t k f, Λ q(αh )/α h t q(α f )/α, Λ < 1, f where Λ is the exogenous productivity gap. As a result of differences in labor productivity, it follows that country h workers are paid a lower wage than those in foreign country, as 12

21 shown by the condition 9 : q(α h )/α h q (α h ) q(α f )/α f q (α f ) ζ < 1, (1.4) where the parameter ζ reflects the exogenous technological gap between the two countries. We restrict our attention to differences in technology that are permanent and specific (i.e. institutional differences). The country with the superior technology can exploit the externality more effectively than the other country Endogenizing migration In the first period, households must choose where to live. I assume that they have a preference for living in their country of birth. To be precise, a household living in country h earns w h plus remittances and the welfare cost from migrating is represented by a θ > 1. In country f, he will earn wage w f but θ = 1. Thus, a household born in country h will migrate to f if his intertemporal utility in the foreign country, V f is higher than home, V h. Once the inequality in V f and V h is calculated, and knowing that θ follows a Pareto distribution, we can obtain the proportion of non-migrant households: 10 1 m t = [ (ŵh 1 t z(γ) ŵ f t + γ m )] ε(1+β) [ ( t 1 = ζ kh t 1 m t z(γ) k f t + γ m )] ε(1+β) t (1.5) 1 m t where w h t /w f t is the wage gap, k h t /k f t is the gap in physical capital, ε is a parameter which reflects labor mobility and z(γ) = γ γ (1 γ) (1 γ) is an artifact derived from V f. 11 The 9 The proof here is straightforward. Assuming that both countries have the same capital to labour ratio, as long as ζ < 1 then w f > w h. 10 See Appendix A for details on how equation (1.5) was derived. 11 The artifact z(γ) has the following properties: z(0) = z(1) = 1, z(0.5) = 0.5, decreasing for γ (0, 0.5), increasing for γ (0.5, 1) and reaches a minimum at γ =

22 following properties of m t can be verified, ceteris paribus: An increase (decrease) in returns to labor in country f, raises (lowers) the migration rate. On the other hand, an increase (decrease) in returns to labor in country h, lowers (raises) the migration rate. Thus, an increase (decrease) in the relative wage wt h /w f t lowers (raises) the migration rate. An increase (decrease) in labor mobility or in the discount factor raises (lowers) the migration rate. If ε = 0 then labor is immobile and hence, there is no migration. Assuming that z (γ) < 0, an increase in migrant generosity decreases (increases) the migration rate. Some assumptions are made to guarantee that the direction of migration is from h to f and, consequently, the migration rate is within the limits of zero and 1. First, in the case of γ = 0, let us assume that k h t /k f t < 1/ζ, which means that the returns to labour of the foreign country is better than that of the home country. Second, for γ > 0, the differential in capital must satisfy k h t /k f t < z(γ)(1/ζ), because households will only migrate when V f > V h. These conditions also imply that when countries have identical technologies and for m t > 0, then k h t /k f t < z(γ) < 1. When these conditions do not hold, migration does not occur or will be in the opposite direction, from country f to h Equilibrium Let L be the number of workers which, for simplicity, is assumed to be constant and the same for each country. 12 Since f is the high wage country, labor supply in the current period will then be equal to (1 m t )L t in country h and (1 + m t )L t in country f. The demands for physical capital and labor are determined by equations (1.2) and (1.3). I consider that 12 The model has been setup such that there are no scale effects. 14

23 migrants contribute to capital accumulation in the home country with the savings they bring back when they return. I further assume that migrant workers can invest a part of their savings in the home country (σ) and the rest in the foreign country (1 σ). Similar to Larramona and Sanso (2006), and assuming full depreciation of capital, the equilibrium conditions for the physical capital rental market for each country are: K h t+1 = s h t (1 m t )L + σs f t m t L K f t+1 = s f t L + (1 σ)s f t m t L. In terms of capital to labor ratio, the resulting expressions are: kt+1 h 1 m t+1 = s h t + σs f m t t (1.6) 1 m t 1 m t k f 1 + m t+1 t+1 = s f t 1 + m t m t + (1 σ)s f t m t 1 + m t (1.7) Autarky equilibrium (ε = 0 and γ = 0) In a regime where there is no factor mobility, households only decide on how much income to save. Thus, the equilibrium physical capital in each period is equal to the savings of households in the previous periods: k j t+1 = s j t. In autarky regime, each country experience sustained growth as shown by: k j t+1 k j t k j t = β [ q(α j ] ) q (α j ) 1 (1.8) 1 + β α j There are no transitional dynamics in growth and, because of the externality in labor 15

24 k h ss k h ss k f ss k k k f ss q q z(γ) ζ k m 1 ( k h k f ) E ( k h k f ) E 45 o 0 m 1 m ss 0 ( q h q f ) 1 qss h qss f Panel A: Steady state capital per worker and migration Panel B: Steady state output and capital per worker Figure 1.2: Determination of steady state equilibrium physical capital and output per worker differentials. productivity, the foreign country grows at a higher rate than the home country which leads to divergence. Hence, convergence in growth in physical capital per worker can only be achieved when countries have identical technologies, that is Λ = 1. One can draw similar conclusions for output per worker for these two countries. Steady state output per worker differential in this economy is determined jointly by the gaps in productivity and capital per worker levels of each country. Open economy equilibrium with remittances (ε > 0 and γ > 0) Suppose we open up the economy. Introducing generosity can induce households to send remittances while they are in the foreign country resulting in limited mobility in physical capital. Labor is fully mobile. For now, I assume that migrants can only save in their home country, σ = From equations (1.1), (1.6), and (1.7), the growth rate for the home country can be expressed as: 13 Comparative statics analysis for the case of σ [0, 1) will be conducted later. 16

25 k h t+1 k h t k h t = [ 1 m t β q(α h ) 1 m t β m t β + (1 + γ) 1 m t β α h q (α h ) ] [ q(α f ) α f q (α f ) ] k f t kt h 1, (1.9) In a similar fashion, I can express the growth rate for the foreign country: k f t+1 k f t k f t = [ 1 β q(α f ] ) q (α f ) m t β α f (1.10) Savings is influenced positively by higher remittances and negatively by lower remittances, thereby affecting economic growth. On the other hand, growth in the foreign country with labor mobility is small relative to autarky. From (1.9) and (1.10), I can express the fundamental law of motion of capital per worker for each country as a function of the migration rate, α, β and γ: kt+1 h = β ( [ 1 mt q(α h ] ) q (α h ) k h m t 1 + β 1 m t+1 α h t + (1 + γ) 1 m t+1 [ q(α f ] ) ) q (α f ) k f α f t, (1.11) k f t+1 = [ 1 β q(α f ] ) q (α f ) k f 1 + m t β α f t. (1.12) It can be verified that, although these countries will continue to accumulate physical capital at different levels, their growth rates will converge in their steady state. Lemma Suppose that γ > 0, k h t /k f t < z(γ)/ζ and ζ < 1. Then there exists differentials in capital per worker for given migration rates in which both countries capital per worker grows at the same rate in the steady state. 17

26 Proof See appendix. I can construct the capital per worker differentials by taking the ratio of k h t+1 to k f t+1 and then impose the steady state condition. Let k h ss/k f ss and m ss be the steady state values, the differential in capital per worker is: k h ss k f ss = (1 + γ) m ss/(1 m ss ) 1/(1 + m ss ) ζ k k(m ss, γ, ζ). (1.13) The above expression coincides if we set equation (1.9) equal to equation (1.10) and imposing the steady state condition. 14 Thus, the previous result of Larramona and Sanso (2006) on convergence in growth rates in capital per worker, for the case of γ = 0, carries through in an economy populated with generous households, γ > 0. Equation (1.13), or the upward-sloping k k schedule depicted in Figure 1.2, shows that while growing at the same rate in the steady state, a high migration rate would result to the capital differential to be high in favor of the home country. From the specification of the production function, convergence in steady state growth of capital per worker implies: Corollary Growth in output per capita for each country is equal in the steady state. Lemma Suppose that γ > 0. The steady state migration rate is bounded between zero and less than one. Proof See appendix. 14 The proof is explained in detail in the Appendix A. 18

27 Imposing the steady state condition to equation (1.5) and after some rearranging the expression which I call the k m schedule can be obtained: k h ss k f ss = 1 ζ [ ] 1 m z(γ)(1 m ss ) ε(1+β) ss γ k m (m ss, γ, ζ). (1.14) 1 m ss Depicted in Figure 1.2, the downward-sloping k m schedule can be interpreted as the number of migrants who are willing to migrate for a given differential in capital per worker. 15 The higher the differential in capital per worker, the lower the wage gap between countries, and consequently, migration will be lower. For the case of γ > 0 and kss/k h ss f = 0 then the steady state migration rate satisfies: 16 z(γ)(1 m ss ) ε(1+β) +1 = γm ss 1 It can be easily shown that the steady state migration rates which satisfy equation (1.14) is between 0 and less than Finally, steady state migration rates is equal to zero if the differential in capital is k h t /k f t = z(γ)(1/ζ). A simple extension of the results provided by Larramona and Sanso (2006) by introducing generosity among households gives: Proposition Suppose that γ > 0, k h t /k f t < z(γ)/ζ, ε(1 + β) = 1 and ζ < 1. Then there exists a unique steady state equilibrium differential in capital per worker and migration rate. The equilibrium migration rate, m, is bounded between zero and less than 15 The k m schedules shown in Figure and 1.4 depicts the situation where labor mobility is high enough, that is ε(1 + β) = 1. This assumption, mainly for illustrative purposes, guarantees that the k m schedule is concave in the capital differential and migration rate space. In spite of this assumption, we will always consider the numerical results that would be derived if ε(1 + β) It is straightforward that if γ = 0 and k h ss/k f ss = 0 then the steady state migration rate is bounded between 0 and 1. See Larramona and Sanso (2006). 17 There are actually two steady state migration rates which satisfy equation (1.14): 0 < m ss < 1 and m ss < 0. We only consider the former case because migration occurs only from country h to f. If migration reversals take place then it is plausible to have m ss < 0. 19

28 one. The equilibrium differential in capital per worker, (k h /k f ), can be found anywhere between zero and z(γ)(1/ζ). Proof The proof follows directly from Lemmas (1.3.1) and (1.3.3). Equations (1.13) and (1.14) form a system defined in the capital per worker differential and migration rate space. The steady state equilibrium differential in capital per worker and migration rate exist at the intersection of the k m and k k schedules. As shown in Figure 1.2, the k k schedule is increasing and will intersect the k m schedule in the range of migration rates between 0 and 1 only once. Although a closed-form solution for m and (k h /k f ) are not available, it can be easily verified that there is only one steady state equilibrium pair of differentials in capital to labor ratio and migration rates which satisfy the system. In this steady state equilibrium, there is a constant flow of workers from country h to f in which the growth rates in capital to labor ratio and output per capita for both countries are equalized. 1.4 Dynamics and numerical experiments This section gives numerical examples to illustrate the qualitative features of the model. To begin, let us assume that aggregate output Q j t is described by a Cobb-Douglas production function with constant returns to scale and φ as the share of capital to production. The resulting expressions for aggregate output and wage will be: ( ) 1 1 φ Q j t = (A j tl j t) 1 φ (K j t ) φ, ŵ j t = (1 φ) k j α j t I can write the first order difference equation which describes the evolution of migration 20

29 rate: (1 m t+1 ) ω+1 γm t+1 } ζ(1 + m t+1 ) {{ } g(m t+1 ) = (1 m t ) ω+1 + ϕ z(γ) m t }{{} g(m t) (1.15) where ω = 1/ε(1+β) and ϕ = ζ(1 γ) γ. From equation (1.15), it can be deduced that if in the initial period m 0 = 0, in the following period, thanks to the differentials in wage, the value of the migration rate will be positive. Note the left hand side, g(m t+1 ), is downward sloping while the right hand side, g(m t ) is parabolic with a unique minimum m min. 18. The location of the minimum is crucial in predicting the trajectory of the migration rate. If m min is located to the right of the bisector dividing the g(m t ) and m t space then convergence in migration rates will be achieved through oscillations. 19 The sending country will experience a cycle of migration flows and capital differentials until they converge to a steady state. If m min is to the left of this bisector then monotone convergence in migration rates as well as capital differentials will occur. In both cases, and under certain constellation of parameters, the steady state equilibrium is stable. Knowing all combinations of parameters which will generate a stable equilibrium however is a non-trivial exercise considering the structure of equation (1.15) as well as the number of parameters involved. Parameter values have to be chosen carefully to ensure that convergence occurs in a reasonable number of periods such that the results of different cases can be well compared. The parameter values used in the numerical experiments are as follows: the initial value for the population L is set to 100; initial capital to labor ratio is set to 100 for both countries; the time preference parameter, β, is set 1/2; the capital share, φ, is set to 1/3; finally, I will consider two possibilities for the labor mobility parameter, ε. The first case is where [ 18 The minimum of g(m t ) is m t = 1 ] 1/ω ϕ z(γ)(1+ω) 19 In the case of Larramona and Sanso (2006), they have shown convergence in capital differentials, the results of which can be easily carried through migration rates. 21

30 labor is highly mobile and another where labor mobility is low. Although different values are possible, I chose for illustration purposes ε = 0.20 and ε = 0.05 for high and low labor mobility, respectively. For each case of the labor mobility parameter I investigate the properties of the model for different levels of the generosity parameter γ. 20 Reversals in migratory flow may occur for values of γ from ranging 0.0 to 0.5 for the high labor mobility case. A substantially high migration rate can occur if the initial output differential is very large. The high rate of initial migration, along with high remittances, can lead to an increase in output in country h, which in turn can trigger reverse migration in subsequent periods. Convergent oscillations happen thereafter until variables for each country reach their stable steady states. For these values of γ, the benefit of generosity with migration is decreasing, that is z (γ) 0. Steady state migration rates are lower the higher the generosity parameter is (as γ reaches 0.5) as a result of workers staying which in turn, happens because of higher remittance received. Moreover, steady state wages for the poor (rich) country are lower (higher) the higher the generosity parameter is. The reverse happens when γ is from 0.6 to 0.9 where the benefit of generosity with migration is increasing. No reversals in migration occur and oscillatory convergence in variables is observed as they proceed to their steady states. The steady state wages for the poor (rich) country are higher (lower) the higher the generosity parameter is. For the low labor mobility case, ε = 0.05, I observe a lower steady state value of the migration rate compared to when labor mobility parameter is high. Similar to the case where mobility is high, for values of γ from 0 to 0.50, the gap in steady state capital to labor ratio and output per capita becomes larger. Variables proceed to their steady states in damped monotone convergence. For γ high enough, from 0.60 to 0.90, the the gap in steady state capital to labor ratio and output per capita becomes small. Damped monotone 20 Details on the results of the numerical experiments including plots as well as the simulation code can be availed upon request. 22

31 kss h kss f k k kss h kss f q q z(γ) ζ k m 1 E 1 E 1 E 2 E m ss 45 o 0 1 Figure 1.3: A situation where an increase in γ widens the gap in capital between h and f. qss h qss f convergence in migration rates and differentials in physical capital to their stable steady states can be observed as well. 1.5 Comparative statics on γ Having established the equilibrium steady state physical capital and migration rate, I can now investigate particularly the comparative statics effects of an increase in generosity of migrants. To simplify matters, I assume that ε(1 + β) = 1 and σ = A preference shock through γ, say from 0 to γ > 0, produces two effects. First, it alters the benefit of remaining in the home country and migrating, thereby affecting the supply of migrants. Second, thanks to generosity of migrants, the demand for migrants (to keep the equal growth rates in capital) is reduced, and the k k schedule shifts inward. The total effect on equilibrium steady state physical capital differential after the change in γ is ambiguous, depending on how the generosity parameter affect the supply of migrants. Depending on how households respond to the preference shock, an increase in generosity may actually bring more harm than good for a migrant-sending country. 21 All the term ε(1 + β) does is to influence the curvature of the k m schedule. 23

32 Proposition Suppose that γ > 0, k h t /k f t < z(γ)/ζ, ε(1 + β) = 1 and ζ < 1. Then a rise in the migrant generosity parameter γ lowers the equilibrium migration rate and ambiguously affects the equilibrium differential in physical capital per worker if, and only if: z (γ)(1 m 1 m ) ε(1+β) < 1 m. (1.16) Proof See appendix. The comparative statics effect of an increase in γ in the equilibrium steady state physical capital differential is: d(k h /k f ) dγ = + {}}{ k k γ {}}{ k m m + k m m }{{} {}}{ k m γ + k k m }{{} + + {}}{ k k m (1.17) and migration rate: dm dγ = {}}{ k m γ k m m }{{} + {}}{ k k γ + k k m }{{} + (1.18) The sign of equation (1.16) implies that k m γ to the left towards the origin. Clearly m γ equilibrium migration rate. However, the sign of d(kh /k f ) is negative and the k m schedule shifts is negative and the higher γ should lower the dγ is ambiguous, depending on the number of households who decide to migrate despite the greater benefits of staying at home because of higher γ. If k m m k k m k k γ k m m then d(kh /k f ) dγ 0; this occurs when fewer households decide to migrate despite the wage differential. As shown in Figure 1.3, 24

33 kss h kss f kss h kss f q q z(γ) ζ k k 1 E 2 E 2 k m E 1 E 1 45 o 0 1 m ss Figure 1.4: A situation where an increase in γ helps close the gap in capital between h and f. 0 1 qss h qss f primarily driven by the extra income earned out of generosity of migrants, more households stay behind leading to diffusion of capital in country h (despite a higher savings thanks to generosity of migrants) and a larger steady state differential in physical capital per worker. 22 It is plausible, however, that the migrant sending country benefits from the generosity of its migrants. Under certain conditions, migrant generosity can actually be beneficial and help propel a technologically inferior country to higher per capita incomes. Consider again the previous proposition and suppose that z (γ)(1 m ) 1 ε(1+β) m 1 m In this situation, the k m schedule shifts to the right and the effects of migrant generosity is high enough to generate higher savings leading to a higher capital for the migrant-sending country. However, the net effect on the migration rate after a change in γ is ambiguous. More formally: 22 As I have shown in the numerical experiments section, the results are sensitive not just on γ but also on the values of other parameters in the model. As an example, for the case of high labour mobility (ε = 0.20), the plausible values for γ where there will be outward migration is 0.6 to 0.9. In this case the gap output remains but it is smaller meaning the poor country was able to catch up slightly to the rich country. 23 The term z (γ)(1 m 1 ) ε(1+β) represent the marginal benefit of being a generous migrant and the term m 1 m represent the marginal benefit of becoming a remittance recipient. 25

34 Corollary Consider k m γ > 0. Then a rise in the migrant generosity parameter γ raises both the equilibrium migration rate and differential in physical capital per worker if, and only if k k γ < k m γ. Thus, one can see a decline in physical capital differential in favor of country h as shown in Figure 1.4. Despite the exogenously determined productivity differentials, the rise in country h s capital per worker is sufficient to narrow down the gap in output. 1.6 Comparative statics on σ Let us now assume that migrants can invest a part of their savings in country h and the rest in country f, that is σ (0, 1). This hypothesis implies the following laws of motion for capital for each country: k h t+1 = β ( [ 1 mt q(α h ] ) q (α h ) k h 1 + β 1 m t+1 α h t m t +(σ +γ) 1 m t+1 [ q(α f ] ) ) q (α f ) k f α f t, (1.19) k f t+1 = 1 + (1 σ)m [ t β q(α f ] ) q (α f ) k f 1 + m t β α f t. (1.20) Following the same procedure described in the previous sections, the relationship between the differential in capital and the migration rate can be described by the modified k k schedule: k h ss k f ss = (σ + γ) m ss 1 m ss 1+(1 σ)m 1+m ss ζ k k(m ss, γ, ζ). (1.21) The relationship between k h ss/k f ss is increasing in m ss, similar to equation (1.13). It can be easily shown that an increase in σ makes the k k schedule shift inward. In the case where 26

35 σ < 1, the gap in capital between the countries would be greater than the case where σ = 1. Moreover, migratory flows in the case where σ < 1 is more intense given the bigger gap in wages between the two countries. The k m schedule can likewise be modified as follows: 1 m t = [ (ŵh g(σ) t z(γ) ŵ f t + γ m )] ε(1+β) [ ( t g(σ) = ζ kh t 1 m t z(γ) k f t + γ m )] ε(1+β) t (1.22) 1 m t with g(σ) = R h /[σr h + (1 σ)r f ]. It can easily be verified that for as long as R f > R h then g (σ) > 0. As long as g (σ) > 0, a decrease in σ leads to higher migration. The effect of a reduction in σ is a widening of the wage gap between the poor and rich country. 1.7 Empirical analysis My analysis so far has identified migration as an important channel by which remittance influences capital accumulation, and consequently output. In this section, I use the theoretical model as a tool to guide a regression analysis as it is typically done in the migration-remittances literature. I will use the insights generated in the previous section to motivate and analyze the relationship of migration and remittances to output Testable hypotheses Let us recall some of the highlights from the theoretical model. First, migration is primarily driven by the differences in income across countries. Typically, an individual from country h (the source) will decide to move to country f (the destination) if income in the latter is greater than in the former. Migration decisions also depend on how easy it is to migrate as some countries are more restrictive than others. In the presence of remittances, 27

36 the decision to migrate is influenced not only by the differences in income or migration restrictions but also by feelings toward recipients of remittances who may be left behind. 24 The decision to migrate is further complicated when one is a recipient of remittances. As I have shown in the theoretical model, migration decisions ultimately depend on the benefits of staying (not migrating) and receiving remittances against the benefits of obtaining a higher income abroad and giving remittances. Thus, there is a need to test the hypothesis whether remittances exert influence on migration decisions and likewise determine the direction of this influence. For a country pair (h, f), I estimate a modified gravity equation for migration of the form: 25 log(mig h,f,t ) = β 0 + β 1 log(remit h,f,t ) + β 2 log(y h,t 1 /y f,t 1 ) + β 3 pol h,f,t 1 + µ h + µ f + u h,f (1.23) where log(mig h,f ) is the log of bilateral migrant stock, log(remit h,f ) is the log of remittances remittances received by residents in country h from migrants in country f, log( y h yf ) is the log of income differentials, pol h,f is a variable representing political stability and µ denote fixed effects. The theoretical model suggests that generosity and remittances should produce different effects on migration decisions depending on the benefits of migration so we leave the sign of β 1 undetermined for now. From the theoretical model, we can also find that improvements in home country incomes should lower the incentive to migrate thus the sign of β 2 should be negative. Improvements in institutional quality in the home country can also hinder migration thus the sign of β 3 is negative as well. It is possible that the relationship between bilateral migration and remittances is non-linear so I included the square 24 Motivations to immigrate and remit such as contractual obligations, joy of giving, and altruism has been mentioned in the literature (Lucas and Stark, 1985). 25 The model is a modified version of the gravity model in way that it does not include population of countries h and f as regressors. See Lewer and Van den Berg (2008) for variants of gravity models of immigration. 28

37 of log of bilateral remittances. I also included the lag of the bilateral migrant stock to test for dynamic structure. The theoretical model also suggests that remittances will eventually affect the economy of the remittance-receiving country. Remittances sent to family members in the home country may end up spent on current consumption goods or saved for future consumption which in turn influence economic conditions the local economy. However, depending on the effects of remittances and generosity to migration decisions, the impact of remittances to output, home and abroad, can vary. Thus, for a country pair (h, f), we estimate a modified regression model akin to Grossmann and Stadelmann (2013): log(y h,t /y f,t ) = γ 0 + γ 1 log(remit h,f,t ) + γ 2 log(remit h,f,t 1 ) + x h,f,t 1γ x + µ h + µ f + u h,f (1.24) where log(y h /y f ) and log(remit h,f ) are variables defined in equation (1.23), x h,f is a vector of controls that potentially affect log income differences between countries, such as differentials in primary school enrollment rates, differentials in tertiary school enrollment rates, differentials in investment rates and differentials in urban population shares. The signs of the coefficients for the main of variable of interest, log(remit h,f ), depends on how remittances influence migration decisions. According to the theoretical model, if the impact of remittances to migration decision is negative, that is fewer households decide to move and remit, then remittances may lead to households staying and thus worsen the differentials in incomes between countries. On the other hand, if the impact of remittances to migration decision is positive, then remittances may lead to more households moving to the higher-wage country thereby improving the differentials in incomes between countries. Thus, the model predicts that, if the sign of β 1 is positive (negative) then we expect the sign of γ 1 to be positive (negative) as well. 29

38 1.7.2 Identification Estimating equations (1.23) and (1.24) may pose some challenges. First and foremost of these problems is endogeneity. There is an obvious reverse causal relationship between remittance and migration in equation (19) and for remittance and output differentials in equation (20). Second, given the cross section of countries, regression models of this type are usually plagued with heteroscedasticity preventing valid statistical inference. Third, variables in the foregoing equations are a mix of unilateral and bilateral variables. A variable is bilateral in the sense that it applies to countries h and f while unilateral variables apply only to country h or f. It has been shown that regression estimates for models with such a mix of variables would likely be biased by standard error clustering. Finally, credible national data on bilateral remittances are not currently available. To address the foregoing estimation problems I employ several approaches. To control for endogeneity bias, as it is usually done in the literature, we focus our attention for a fixed year (2010) for the dependent variables and measure the controls other than remittances in lagged form (year 2000). To further account for endogeneity I use instrumental variables method. For equation (1.23), I take log(remit h,f ) as the endogenous variable and use the following variables as instruments: visa h,f, colony h,f, and mig h,f,1980. The variable visa hf measures the number of countries residents in country h can visit without the need of a visa. This variable should capture the ease of migration and we expect that it has a positive influence in remittances. The variable colony h,f denotes whether countries h and f have a colonial link. It is expected that one can observe higher migrant stocks for countries with a past colonial relationship. The variable mig h,f,1980 measures the stock of migrants from country h in country f for the year The intuition is that a larger percentage of past migrants from country h in country f act as a signal to current potential migrants regarding the destination country s openness and its migration policy. Past migration also takes into 30

39 account other factors such as trust, cultural proximity, and social openness to migrants as perceived by residents in country h. 26 Similarly for equation (1.24), we take log(remit h,f ) as the endogenous and use the following as instruments: mig h,f,1980, contig h,f, dist h,f and colony h,f. The variable dist h,f captures the distance between the country pair (h, f). contig h,f captures whether the country pair are geographically contiguous. The expectation is that the larger the distance between countries is, the lower the number of migrants and remittances. To account for heteroscedasticity, I estimate the model using heteroscedasticityconsistent clustered standard errors. I use remittance-source and -destination fixed effects to address potential bias due to the mix of unilateral and bilateral variables Bilateral remittances The lack of credible national bilateral remittances data has been often cited as a problem in the remittances literature. Lueth and Ruiz-Arranz (2008) were among the first to produce such data coming mostly from the International Transaction Recording System and surveys of banks. It has been shown, however, that data reported from these sources may not be accurate because funds coursed through banks may be attributed to a country other than the actual source country. 28 To remedy this problem, Ratha and Shaw (2007) proposed several approaches to allocate various remittance received by countries one of which is used in our paper. Their approach involve calculating weights w h,f in order to transform unilateral remittances data into bilateral. The simplest method assumes that aggregate remittances received by country h from country f is proportional to the number of m migrants in f: The use of past migration as an instrument is not new in the literature, for instance see Grossmann and Stadelmann (2013). In the same vein, Beine et al. (2011) found that existing diasporas (the stock of people born in a country and living in another one) affect the size and human-capital structure of current bilateral migration flows. 27 Redding and Venables (2004) has shown that adding fixed effects to model eliminates this bias. 28 For details see Ratha and Shaw (2007). 29 Ratha and Shaw (2007) notes that the large variation of incomes across remittance source countries may limit the usefulness of this method. 31

40 w h,f,t = m h,f,t f m h,f,t One should note, however, that for a given time t, the foregoing method assume that migrants sends a constant share of remittance regardless of income in the destination country Data To empirically implement equations (1.23) and (1.24) I assembled data from various sources summarized in Table 1.1. In light of recent advances in collection of bilateral migration stocks across countries, The World Bank was able to produce several bilateral migration matrices every ten years, 1960 to 2000, of which the years 1980 and 2000 were used in this paper. The bilateral migration data for 2010 were taken from the Migration and Remittances Factbook 2011 (The World Bank, 2011). Data for real GDP per capita and investment shares were sourced from the Penn World Tables 7.1. Aggregate remittances, primary school enrollment rates, tertiary school enrollment rates and urban population shares were obtained from the World Development Indicators (WDI). 30 Data for other bilateral variables such as contiguity, distance and colony were from Mayer and Zignago (2011). Institutional quality (political stability) were sourced from Kaufmann and Kraay (2002). Finally, the number of countries residents can visit without a visa were obtained from Henley and Partners (from various years). The countries included in our study are listed in Appendix B. My analysis covers 175 h countries (including 34 OECD member-countries) and 34 f OECD countries bringing our country (h, f) pairs to 5,950 all in all, although the number used in some of the estimations is lower due to data limitations on some of the regressors. I restrict the f countries to OECD given that they account for 70%-90% of 30 In the WDI database remittances are referred to as personal remittances received. Personal remittances comprise personal transfers and compensation of employees. 32

41 remittances received by developing countries Empirical results Table 1.2 presents the estimation results for the modified gravity equation for migration. Columns 1 and 2 report results using OLS with contemporaneous remittance and together with its lag, respectively. The results show that there is a positive and significant relationship between remittance and bilateral migrant stocks. A one percentage increase in remittance raises bilateral migration by about 0.69 percent, ceteris paribus. I included in the regression the square of remittances to check for non-linearity. With the square of remittances, the coefficient for contemporaneous remittance hardly changed, including its sign and significance. The sign of lagged migration and lagged remittances are positive and significant as well, suggesting dynamic structures for these variables. 31 Across the OLS estimations, the signs of the income differentials, log(y h,t 1 /y f,t 1 ) and institutional quality variable, pol hf,t 1 were negative and consistent with expectations. A rise in the income differential could mean an improvement (deterioration) of home (destination) country incomes which translates to a reduction in number of migrants. Similarly, a rise in the ranking for political stability could mean a lowering in number of migrant stocks. The modified gravity model explains around 87% of variations in bilateral migrant stocks. Columns 3, 4 and 5 present the results using instrumental variables. The first stage regressions indicate that the instruments (bilateral migrant stocks in 1980, colonial link and visa restrictions) are not weak and that the overidentification restrictions are valid, giving us the confidence that the set of instruments is appropriate. 32 The signs of the instruments 31 The lagged remittance variable was taken out of the regressions as it is positively correlated with lagged output. 32 The Cragg-Donald Wald F statistics exceed all of the Stock-Yogo critical values suggesting the absence 33

42 Table 1.1: Data description and sources. Variables N Mean Std Dev Description and Source log(y h,t /y f,t ) 6, Log of GDP per capita of country h minus the log of GDP per capita of country f in year GDP data from PWT 7.1 log(remit hf,t ) 4, Log of personal remittances received by residents of country h weighted by the stock of migrants of country h in country f in year Remittance data from WDI. log(remit hf,t 1 ) 3, Log of personal remittances received by residents of country h weighted by the stock of migrants of country h in country f in year Remittance data from WDI. prim hf,t 1 5, Primary school enrollment in country h divided by primary school enrollment in country f in year Enrollment data from WDI. tert hf,t 1 3, Tertiary school enrollment in country h divided by tertiary school enrollment in country f in year Enrollment data from WDI. inv hf,t 1 6, Investment share in country h divided by investment share in country f in year Data from PWT 7.1. urban hf,t 1 6, Urban population share in country h divided by the urban share in country f in year Urban population shares sourced from WDI. mig hf,1980 5, Stock of migrants from country h in country f in Migrant stock data from World Bank Bilateral Migration Matrices. contig hf 5, Bilateral dummy variable denoting a common border between countries h and f. Data sourced from Mayer and Zignago (2011) dist hf 5, Log of the geodesic distance between important cities between countries h and f. Data sourced from Mayer and Zignago (2011) colony hf 5, Bilateral dummy variable denoting colonial link between countries h and f. Data sourced from Mayer and Zignago (2011) pol hf,t 1 5, Political stability ranking of country h divided by political stability ranking of country f in year Political stability data sourced from Kaufmann and Kraay (2002) visa hf 5, Number of countries residents in country h can visit without the need of a visitor s visa. Sourced from Henley and Partners (2010) 34

43 Table 1.2: Impact of bilateral remittances on migrant stocks. OLS (1) OLS (2) OLS (3) OLS (4) 2SLS (4) 2SLS (5) 2SLS (6) log(remit h,f,t ) *** *** *** *** *** *** *** (0.0080) (0.0301) (0.0071) (0.0171) (0.0335) (0.0403) (0.0929) log(remit 2 h,f,t ) *** (0.0011) log(mig h,f,t 1 ) *** (0.0084) log(remit h,f,t 1 ) *** (0.0142) log(y h,t 1 /y f,t 1 ) ** ** ** *** *** *** *** (0.0266) (0.0244) (0.0171) (0.0238) (0.0347) (0.0369) (0.0654) pol h,f,t *** *** ** *** (0.0476) (0.0472) (0.0324) (0.0489) (0.0694) (0.0730) (0.1153) Intercept *** *** *** *** *** (0.2384) (0.1313) (0.1873) (0.6272) (0.7500) (1.7158) Remittance source FE Yes Yes Yes Yes Yes Yes Yes Remittance destination FE Yes Yes Yes Yes Yes Yes Yes Adjusted R N First stage regressions log(mig h,f,1980 ) *** (0.0011) colony h,f *** (0.2267) visa h,f *** *** (0.0019) (0.0019) Hansen s J Hansen s J p value (0.9534) Cragg-Donald Wald F F statistic (first stage) Notes: Dependent variable is log of bilateral stock of migrants from country h in country f in year Time subscripts denote time t as 2010 and t 1 as log(remit hf,t ), log(y h,t 1 /y f,t 1 ) and pol hf,t 1 denote bilateral remittance received by country h from country f, the log of GDP per capita differentials between countries h and f in year 2000 and differentials in political stability rankings between country h and f, respectively. mig hf,1980, colony hf and visa hf denote the stock of migrants from country h in country f in 1980, whether h and f have a colonial link and the number of countries residents in country h can visit without the need of a visa, respectively. Figures in parentheses are robust clustered standard errors. ***, ** and * denote significance at the 1%, 5% and 10% levels, respectively. 35

44 Table 1.3: Impact of bilateral remittances on output gaps across countries. OLS 1 OLS 2 2SLS 1 2SLS 2 2SLS 3 2SLS 4 log(remit hf,t ) *** *** *** *** *** *** (0.0005) (0.0005) (0.0017) (0.0011) (0.0018) (0.0011) log(remit h,f,t 1 ) *** *** *** (0.0008) (0.0009) (0.0008) Controls Yes Yes Yes Yes Yes Yes Remittance source FE Yes Yes Yes Yes Yes Yes Remittance destination FE Yes Yes Yes Yes Yes Yes Adjusted R N 2,680 2,280 2,680 2,680 2,280 2,280 First stage regressions mig h,f, *** *** *** *** (0.0015) (0.0012) (0.0015) (0.0012) contig h,f *** *** (0.2404) (0.2374) dist h,f *** *** (0.0000) (0.0000) colony h,f *** *** (0.3889) (0.4193) Hansen s J Hansen s J p value (0.9128) (0.7132) Cragg-Donald Wald F F statistic (first stage) Notes: Dependent variable is log of GDP per capita differentials between countries h and f in year Time subscripts denote time t as 2010 and t 1 as Control variables include differentials in primary school enrollment, tertiary school enrollment, investment shares and urbanization shares, all lagged (year 2000 values). mig hf,1980, contig hf, dist hf and colony hf denote the stock of migrants from country h in country f in 1980, whether h and f have a common border, the distance between h and f and whether h and f have a colonial link, respectively. Figures in parentheses are robust clustered standard errors. ***, ** and * denote significance at the 1%, 5% and 10% levels, respectively. 36

45 in the first stage also conform to expectations. The coefficients for the main variables of interest in the second stage regressions were a bit higher relative to OLS estimations, reflecting the endogeneity bias. The coefficients for contemporaneous remittances remained positive and significant. The positive effect of remittances to migration has an important implication in testing the predictions of the theoretical model. It predicts that the effect of remittances to output differentials should be positive. The sign for the income differentials remained positive and significant as well. The coefficient for institutional quality was negative and significant for the most part. Its significance was lost when visa restrictions alone was used as in instrument. Now we turn our attention to the estimation results for equation (20). The first two columns of Table 1.3 presents the results using OLS. Across different specifications as shown in columns (1) and (2), the signs of the coefficient for remittance is positive and very significant. The result suggests that as remittances increase by 1%, the differential in output increases by around %, ceteris paribus. Although the magnitude is quite small, the results are robust across specifications and is consistent with our theoretical predictions. An increase in remittances raises the number of migrants, which in turn boosts remittance received by residents in the home country thereby improving the differentials in output between countries. Results are quite robust, as shown in column (2), even if I replace contemporaneous remittance with is lagged value (2010) to reduce endogeneity bias. Finally, the results of the instrumental variables regressions for output differentials and remittances are presented in columns (3)-(6). Similar to the results for the modified gravity equation for migration, and shown by the tests for weak instruments and overidentifying restrictions, it is not worrisome that the remittances are weak or are not appropriate for this of weak instrument problem. The F statistics for all first stage regressions are also greater than 10 which also indicates that the instruments were appropriate. 37

46 setting. The signs of the instruments in the first stage regressions likewise conform to expectations. As for the second stage, the results are robust even accounting for endogeneity as the signs of the coefficients or even their magnitudes remain unchanged, relative to the OLS estimations. Moreover, even if we include the lagged remittance, the signs and relative sizes of the coefficients remained the similar. The model was able to explain around 80% of variation in real output per capita differentials across countries in our sample. My main conclusions remain qualitatively unchanged and overall robust. 1.8 Conclusion This paper presents a different approach to the analysis of remittances and its effects to economic growth and convergence. Using a two-country overlapping generations framework, I build a model of two artificial economies characterized by heterogeneous households. I allow for labor mobility through temporary migration and a limited form of capital mobility through remittances. Thus, we have one group of households who, out of generosity to others, migrate and transfer goods to non-migrants; and another group who stays behind and receive these transfers. Migration is a consequence of utility maximization which incorporates the costs and benefits of moving to another country or staying behind. The decision to migrate is influenced by the wage differential, degree of labor mobility and the benefits derived from generosity - the act of giving remittances or manna from heaven to other households. Each country possess a different level of productivity which creates the condition for migration to occur. Productivity in these countries is influenced by the positive spillovers in the form of investments in physical capital, which is in turn influenced by migration, thereby generating endogenous growth. In this framework, the growth rates in output for each country will converge to its steady state value while maintaining their respective levels of output and capital per worker, and 38

47 wages. Differentials in wages leads to the persistence of migration in the steady state. In transition to the steady state reversals in migratory flow may occur until the migration rate converges to be a stable steady state. Augmenting the model with generosity and remittances does not completely eliminate migration in the steady state. I have shown that generosity of migrants may help lower migration rates and widen the gap in per capita income across countries. However, if high enough, generosity can actually help propel a technologically backward country to high per capita incomes resulting in the poor country to catch up. I conclude that, in theory, migration with generosity does not always lead to convergence in levels of output and capital per worker across countries. In this paper, I also tested the predictions of the theoretical model using bilateral data. I found empirical evidence that remittances exert a positive influence on bilateral migration. This result also suggests that remittances are affected by variables usually associated with migration such as past migrant stocks, past colonial relationships, distances between countries and travel restrictions. A key result of the theoretical model is that if the effects of remittances to migration is positive then remittances should have a positive effect on output differentials - a feature supported by the data. Thus, the foregoing empirical evidence lends support to our theoretical model. My principal conclusion is that, although the impact of remittances to output differentials is small, it nonetheless can be used as a policy tool for developing countries to catch up with the rich ones. Appendix Derivation of Equation 1.5 The solution to migrant households optimization problem are: ĉ f 1t = 1 γ 1+β wf t, ˆd f 1t = γ 1+β wf t, ĉ h 2t+1 = 1 γ 1+β βr t+1w f t, and ˆd h 2t+1 = γ 1+β βr t+1w f t. Substituting these solutions to the mi- 39

48 grant households utility function gives us their indirect utility: V f t = ln (θβr t+1 ) β ( z(γ)w f t 1 + β ) 1+β (1.25) The solution to non-migrant households optimization problem are: ĉ h 1t = 1 1+β Ih t and ĉ h 1t = β 1+β R t+1i h t where I h t = w h t + γ ( m t 1 m t ) w f t. Substituting these solutions to the non-migrants utility function gives us their indirect utility: V h t = ln (βr t+1 ) β ( θi h t 1 + β ) 1+β (1.26) Let t V f t θ: V h t. It can be shown that the difference in indirect utilities is decreasing in t θ = 1 θ < 0, and t = 0 at θ = ˆθ. Comparing the indirect utilities such that V f > V h, we find: Thus, V f t V h t θ < > 0 if θ < ˆθ and V f t [ z(γ) wf t I h t V h t Since θ follows a Pareto distribution then: m t = ˆθ 1 f(θ)dθ = ] 1+β = ˆθ < 0 if θ > ˆθ. [ I h t z(γ)w f t ] ε(1+β) With some rearranging, we can obtain from the above expression equation (1.5). Derivation of Lemma In this proof we will show that in the steady state (where a stable, constant continuous flow 40

49 of migrant is achieved) the growth rates in physical capital and output per worker in both countries are equalized. Perpetual growth in output is a result of the way the production function is specified. Setting (1.9) equal to (1.10) we get: 1 m t β 1 m t β 1 = [ q(α h ] ) q(α h m t β ) + (1 + γ) α h 1 m t β [ q(α f ] ) q(α f ) 1 α f 1 β 1 + m t β [ q(α f ] ) k f q(α f t ) α f kt h (1.27) We then impose the steady state conditions m t+1 = m t = m ss and k t = k ss : [ q(α h ] j) q(α h ) α j m ss + (1 + γ) 1 m ss [ q(α f ] ) k q(α f f ) ss α f ksst h = m ss [ q(α f ] ) q(α f ) α f Dividing both sides by [ q(α f ) α f q(α f ) ] and using the wage gap definition in equation (1.4): m ss kss f ζ + (1 + γ) 1 m ss k h sst = m ss With some algebra we can obtain equation (1.13) from the above expression. We can conclude that along the k k schedule the growth rates for the two countries are equalized. Proof of Lemma Recall the expression for the k m schedule: k h ss k f ss = 1 ζ [ ] 1 m z(γ)(1 m ss ) ε(1+β) ss γ 1 m ss To exclude cases of migratory reversals let the following hold: γ > 0, k h t /k f t < z(γ)/ζ and ζ < 1. If migration rate is zero then the terminal value for the differentials in capital to 41

50 labor ratio is kh ss k f ss = z(γ) ζ in capital to labor ratio is [ 0, z(γ) ζ. Thus, the possible range of values for the steady state differentials ]. It is quite clear that the km schedule has a vertical asymptote at negative infinity as m ss approaches 1. Now suppose kh ss k f ss schedule must satisfy: = 0. Then the k m z(γ)(1 m ss ) 1 ε(1+β) = γ m ss 1 m ss Note that the left hand side is downward sloping and has an intercept of m ss = 1. On the other hand the right hand side is increasing and has a vertical asymptote approaching m ss = 1. Since the right hand side intersects the left hand side only once in the interval of 0 and 1 then the migration rate has to be less than one. Thus, the possible steady state migration rates which satisfy the k m schedule must be in somewhere in the interval [0, 1). Proof of Proposition Recall the generalized expressions for the k k and k m schedules and substituting the equilibrium steady state capital to labor ratio differential and migration rate: k = k k(m, ζ, γ) (1.28) k = k m(m, ζ, γ, ε, β) (1.29) where k = (k h /k f ). Totally differentiating equations (A.4) and (A.5) with respect to the parameters, k and m and setting dζ = dβ = dε = 0: dk = k k m m γ dγ + k k γ dγ dk = k m m m γ dγ + k m γ dγ 42

51 Dividing through dγ and arranging the results in matrix form: 1 k k m 1 k m m dk dγ dm dγ = k k γ k m γ From our assumptions: k m m < 0, positive: k k m > 0, and k k. The determinant is unambiguously γ Det = k m m + k k m Using Cramer s rule we obtain the derivatives: and dk dγ = k k γ k m γ k k m k m m k m m + k k m = k k γ km + k m m kk γ m k m m + k k m dm dγ = 1 k k γ 1 k m γ k m m + k k m = k m γ k k γ k m m + k k m which are the comparative static effects of generosity on the capital to labor ratio differential and migration rate. Appendix Albania, Algeria, American Samoa, Andorra, Angola, Antigua and Barbuda, Argentina, Armenia, Aruba, Australia, Austria, Azerbaijan, Bahamas, Bahrain, Bangladesh, Barbados, Belarus, Belgium, Belize, Benin, Bermuda, Bhutan, Bolivia, Bosnia and Herzegov- 43

52 ina, Botswana, Brazil, Bulgaria, Burkina Faso, Burundi, Cambodia, Cameroon, Canada, Cape Verde, Chile, China, Colombia, Congo, Rep., Costa Rica, Cote d Ivoire, Croatia, Cyprus, Czech Republic, Denmark, Djibouti, Dominica, Dominican Republic, Ecuador, Egypt, Arab Rep., El Salvador, Estonia, Ethiopia, Fiji, Finland, France, Gambia, Georgia, Germany, Ghana, Greece, Greenland, Grenada, Guam, Guatemala, Guinea, Guinea- Bissau, Guyana, Haiti, Honduras, Hungary, Iceland, India, Indonesia, Iran, Islamic Rep., Iraq, Ireland, Isle of Man, Israel, Italy, Jamaica, Japan, Jordan, Kazakhstan, Kenya, Korea, Rep., Kosovo, Kuwait, Kyrgyz Republic, Lao PDR, Latvia, Lebanon, Lesotho, Liberia, Liechtenstein, Lithuania, Luxembourg, Macedonia, Malawi, Malaysia, Maldives, Mali, Malta, Mauritius, Mexico, Moldova, Monaco, Mongolia, Montenegro, Morocco, Mozambique, Myanmar, Namibia, Nepal, Netherlands, New Caledonia, New Zealand, Nicaragua, Niger, Nigeria, Norway, Oman, Pakistan, Palau, Panama, Papua New Guinea, Paraguay, Peru, Philippines, Poland, Portugal, Puerto Rico, Romania, Russian Federation, Rwanda, Samoa, San Marino, Sao Tome and Principe, Saudi Arabia, Senegal, Serbia, Seychelles, Sierra Leone, Slovak Republic, Slovenia, Solomon Islands, South Africa, South Sudan, Spain, Sri Lanka, St. Lucia, St. Vincent and the Grenadines, Sudan, Suriname, Swaziland, Sweden, Switzerland, Syrian Arab Republic, Tajikistan, Tanzania, Thailand, Togo, Tonga, Trinidad and Tobago, Tunisia, Turkey, Tuvalu, Uganda, Ukraine, United Kingdom, United States, Uruguay, Vanuatu, Venezuela, Vietnam, Yemen, Rep., and Zambia. 44

53 Chapter 2 International Worker Remittances and Economic Growth in a Real Business Cycle Framework 2.1 Introduction Next to foreign direct investments and foreign aid, unilateral transfers such as remittances constitute a significant source of foreign financing in developing countries (see Figure 2.1). The common wisdom is that these transfers are generally spent on consumption goods thus helping recipients improve their welfare. However, beyond enhancing welfare, the effects of remittances on economic performance is not well understood. Some of the initial empirical work found that the impact of remittances on economic growth is positive, although these were later qualified or overturned in subsequent papers. 1 Current research have become skeptical about the developmental effects of remittances. In fact, a majority 1 The differences in findings with respect to effectiveness of worker remittances are highlighted in a literature review by Barajas et al. (2008). 45

54 of recent papers have found that there is virtually no aggregate relationship between economic growth and migrant transfers across developing countries: The results show that, at best, workers remittances have no impact on economic growth. (Barajas et al., 2009) How come worker remittances do not appear to be effective at the aggregate level? To answer this question, this article steers away from the conventional approach of using reduced forms of the neoclassical growth models to study the relationship between remittances and economic growth, and instead uses the Real Business Cycle (RBC) model. 2 The RBC model is appropriate because it predicts how agents will respond to exogenous aggregate unilateral transfer shocks which is what remittances generally are. RBC theory offers a good framework to analyze how consumption and savings allocations are made, and this is particularly useful because in order for remittances to be effective it must be invested. This article contributes to the literature in three respects: First, the theoretical component of this paper makes use of the RBC model to evaluate the effectiveness of remittances to growth in GDP per capita. Through RBC models, it is possible to isolate the economic impact of worker remittances to GDP per capita from other types of shocks (like technology shocks), something that can be difficult to do empirically. Second, this study contributes to the empirical literature on remittance effectiveness by adding permanent and temporary components of migrant remittances to variables that explain economic growth. Within the literature on remittances this paper is, to the best of my knowledge, the first to use such a decomposition as an explanatory variable to growth in GDP. Lastly, to calibrate the model, as well as for the regressions, data from the most number of recipient countries possible (49 for the calibration and 81 for the regressions) and a long time horizon ( ) was 2 This study also has an empirical component but the econometric analysis is motivated by RBC theory. 46

55 Figure 2.1: Net Official Development Assistance, Remittances and Foreign Direct Investments, averages in current US dollars (from World Development Indicators). used. This article has the following main results motivated by RBC theory: First, a permanent change in worker remittances does not affect output in the long run. The idea is that a permanent change in worker remittances produces a positive income effect, and since leisure is a normal good, hours worked falls. A reduction in hours worked is followed by a reduction in output. Second, a temporary change in worker remittances produces a positive effect to output in the short run. Agents want to maintain a smooth consumption profile in the face of a temporary change in worker remittances. This consumption smoothing behavior force agents to optimally save some of their remittances received leading to an increase in investment and output. To test these results empirically, a reduced-form regression model commonly used in the remittance effectiveness literature with the addition of permanent and temporary component of remittances was estimated. The findings suggest that the temporary component of worker remittances positively affect growth rates in GDP per capita, 47

56 although the extent of this effect can be considered mild. On the other hand, the permanent component of remittances does not affect growth in GDP per capita. The rest of this paper is organized as follows. Section 2 provides a brief review of the background literature. Section 3 presents the model with stochastic remittance shocks. A simplified version of the model in the non-stochastic steady state is presented in section 4. Section 5 presents and discusses the calibration and simulation results. Section 6 presents the empirical model and regression results. Section 7 concludes. 2.2 Brief Background Literature Regression-based methods dominate the current literature on the effectiveness of remittance to economic growth. Several studies stand out from this strand of literature. One of the first was by Chami et al. (2003) where they used cross-section and panel data of 113 remittance recipients for 29 years ( ). They found that the share of remittance to GDP is, at best, not a significant variable in explaining GDP per capita growth. According to them remittances are negatively correlated economic growth because of a moral hazard problem. Due to remittances, recipients can decrease their labor force participation, limit their job searches, reduce labor effort or engage in risky ventures. These results were later supported by Rao and Hassan (2012) which found that direct growth effects of remittances as measured by share of remittance to GDP are insignificant. Their results were robust with alternative specifications and estimation methods. Faini (2006) used a cross section of 68 countries in which the dependent variable is the average annual per capita GDP growth rate from 1980 to He found a positive and significant relationship between total remittances and economic growth. However, using an instrumental variable approach with distance between countries as an instrument, the coefficient for total remittance lost its significance but remained positive. 48

57 Another paper that used total remittances as an explanatory variable is Giuliano and Ruiz-Arranz (2005). They used a panel of 73 countries from 1975 to Their basic specification regressed per capita GDP growth on the ratio of total remittances to GDP with the initial level of GDP per capita, the investment rate, population growth, the fiscal balance as a percentage of GDP, years of education, openness, and inflation as explanatory variables. Their specification did not find evidence that total remittances is significantly related to growth. Catrinescu et al. (2009) covered 114 countries during the period and included institutional variables into their analysis. Overall, their study found a positive relationship between growth in GDP per capita and gross capital formation, as well as between growth and some of the institutional variables. Their study also found some evidence of a mild positive relationship between growth and total remittances. In a cross-country study with 84 recipient countries and a longer time period ( ), Barajas et al. (2009) found that remittances do not seem to make a positive contribution to economic growth. Their study was different in two respects: First, they constructed a new instrument which is the ratio of remittances to GDP of all other recipient countries. Second, a trade-weighted average growth rate of real per capita GDP of the remittancereceiving country s top 20 trading partners was added as an explanatory variable. Their findings were consistent with criticisms of foreign aid presented by Rajan and Subramanian (2008) who point out that there is very little evidence that decades of official transfers have contributed much to the growth of developing economies. There are only few studies that used RBC (or, in general, DSGE) models to study the effects of remittance fluctuations on the cyclical behavior of macroeconomic variables. Often, these studies focus only on a single country or a small group of countries with similar characteristics (i.e. developing or major remittance recipient countries). Chami et al. 49

58 (2006) used a DSGE model to investigate the influence of countercyclical remittances on the conduct of fiscal and monetary policy and trace their effects on real and nominal variables. They found that remittances raise disposable income and consumption and insure against income shocks, thereby raising household welfare. Remittances increase the correlation between labor and output, thereby producing a more volatile business cycle and increasing output and labor market risk. Acosta et al. (2009) developed and estimated a two-sector DSGE model to analyze the effects of remittances on the economy of El Salvador. They found that an increase in remittance flows leads to a decline in labor supply and an increase in consumption demand that is biased toward non-tradables. Durdu and Sayan (2010) used a two-sector model of a small open economy with financial frictions calibrated to Mexico and Turkey. They found that remittances dampen the business cycles in Mexico, whereas they amplify the cycles in Turkey. They also found that effects of worker remittances in the long run are mild. In the short run, however, remittances have quantitatively large impacts on the economy, when there are borrowing constraints. Mandelman (2013) constructed a heterogeneous agent model to analyze the role of monetary policy in a small open economy subject to sizable remittance fluctuations. He found that recipient households are better off with the exchange rate peg when facing an uptrend in remittances. Moreover, he was able to show that countercyclical remittances and a flexible policy regime pursuing stabilization goals are useful tools to smooth the consumption path of credit constrained households and achieve macroeconomic stability. 50

59 2.3 Canonical Small Open Economy RBC Model with Remittances The model economy discussed below has the basic structure of the standard small open economy RBC model with the addition of remittance shocks. 3 Time is discrete and indexed by t = 0,...,. Variables with a bar denote a steady state value. The economy is populated by a large number of identical infinitely-lived agents. The expected lifetime utility, U, of the representative agent is defined as: U = E 0 t=0 β t u(c t, l t ) (2.1) where u is the period utility function, β > 0 is the degree of time preference, c t represents consumption and l t represents leisure. Agents are endowed with one unit of time which can be allocated between leisure and work, n. The economy produces an internationally tradable composite commodity using inputs of capital, k, and labor, subject to the following technology: G(k t, n t, k t+1 ) = z t f(k t, n t ) Φ(k t+1 k t ) (2.2) where z t is a random productivity shock and Φ(k t+1 k t ) is the cost of adjusting the capital stock as a function of net investment. 4 In this expression, the term zf( ) is a constant returns to scale production function representing GDP. The function Φ( ) is assumed to satisfy Φ(0) = Φ (0) = 0 thereby ensuring that in the non-stochastic steady state adjustment costs are zero and the domestic interest rate is equal to the marginal product of capital. 5 The 3 With the exception of remittances, the discussion in this section follows the small open economy model described in Schmitt-Grohe and Uribe (2003) and Mendoza (1991). 4 As in Mendoza (1991) this model ignores the existence of non-traded goods and the substitution effects induced by changes in relative prices of the traded non-traded commodities. 5 This configuration of capital adjustment costs avoids excessive volatility in investment as a result of 51

60 random productivity shock is assumed to follow an AR(1) process: ln z t = (1 ξ z ) ln z + ξ z ln z t 1 + ε z,t (2.3) where ξ z < 1 and ε z N(0, σz). 2 Similar to Mendoza (1991), agents have access to a perfectly competitive international capital market in which foreign assets b, that pay or charge the real interest rate r, are exchanged with the rest of the world. The law of motion of foreign asset holdings is expressed as: b t+1 = tb t + (1 + r t )b t (2.4) where the trade balance, tb, reflects the net flow of the good between the small open economy and the rest of the world. Following Nason and Rogers (2006) I assume that the real interest rate is given by: r t = r + p(b t ) (2.5) where r denotes the world interest rate and p(b t ) is a country-specific interest rate premium. For simplicity I assume that the risk premium is strictly increasing and the world interest rate is constant. A bond-elastic interest rate generates some transition dynamics in the model the farther the foreign asset is in its steady state. 6 Furthermore, the risk premium is nil in the non-stochastic steady state, p( b) = 0. Two reasons motivate the assumption of a bond-elastic interest rate: First, it can capture the presence of financial frictions; And second, it gives rise to a steady state of the model that independent of the initial net foreign variations in domestic and world interest rates. 6 This can be demonstrated if one will look at the first order conditions (or Euler equations) with respect to physical capital and foreign asset. 52

61 asset position of the economy. 7 Disposable income in the economy may be allocated to consumption, foreign asset holdings, or investment i. Since residents in the economy being modeled are assumed to receive remittances from migrants abroad, d, the appropriate resource constraint is given by: G(k t, n t, k t+1 ) + d t = c t + i t + tb t. (2.6) From the resource constraint one can see that remittances enter as a form of a unilateral transfer. 8 Remittance are tied to the country s stock of migrants abroad which for simplicity is assumed to be exogenous and embodied in d. Similar to the productivity variable remittances is assumed to follow an AR(1) process: ln d t = (1 ξ d ) ln d + ξ d ln d t 1 + θε z,t + ε d,t (2.7) where ξ d < 1, θ R and ε d N(0, σ 2 d). The parameter θ is included to capture the relationship that may exist between remittance levels and the state of economic conditions in the recipient country. If θ is positive, then the recipient country will receive an increase in remittances when economic conditions are good. However, θ may also be negative, and in this case the economy will receive relatively more remittances during periods where productivity and output are low. 9 7 Schmitt-Grohe and Uribe (2014) notes that equilibrium in a small open economy is non-stationary in consumption, trade balance, and foreign asset holdings, thereby complicating the task of approximating equilibrium dynamics. Stationarity can be induced by, among other devices, introducing a bond-elastic interest rate (e.g. Nason and Rogers, 2006) or an endogenous discount factor in the utility function (e.g. Mendoza, 1991). 8 The left hand side of equation (6) represents the Gross National Product in this economy. 9 There is empirical evidence which suggest that countercyclicality is not commonly observed among remittance-receiving countries (Sayan, 2006). 53

62 Finally, the law of motion of capital is given by: k t+1 = i t + (1 δ)k t (2.8) where δ denotes the depreciation rate of capital. Since there are no externalities and other market imperfections, the competitive equilibrium in this economy can be calculated as the solution to the appropriate social planning problem. The Social Planner seeks to maximize the expected lifetime utility of the representative agent by choosing the optimal sequences {c t, n t, y t, i t, k t+1, b t+1 } t=0, subject to the resource constraint, the law of motion for capital, the law of motion for foreign assets, the production technology, the stochastic processes, and a no-ponzi constraint: lim j E t b t+j Π j s=0(1 + r s ) 0. (2.9) 2.4 Remittances in the Long Run: A Special Case with Inelastic Labor Supply The steady state effects of remittances can be best studied in a simple, deterministic setting. A simple model allows one to study the effects of remittance flows on agents savings and consumption decisions. The simplification was done in three ways: First, the period utility function is assumed to depend only on consumption and labor is supplied inelastically (n = 1). Second, the stochastic components are removed, thus z t = z and d t = d. And third, capital mobility is restricted in a way that unilateral transfers (remittances) are the only source of funds outside the model economy and there are no capital adjustment costs. The third restriction makes the model, in effect, a closed economy. With the exception of remittances, the simplified version of the benchmark model and the discussions in 54

63 this section follow Annen and Kosempel (2012). 10 As a result of these simplifications one can easily find the Euler equation, the resource constraint and the usual transversality condition: βu (c t+1 )[ zf (k t+1 ) + 1 δ] = u (c t ) (2.10) k t+1 = zf(k t ) + (1 δ)k t c t + d (2.11) lim t u (c t )k t = 0 (2.12) Since the general properties of the model are well-known, the discussion will focus only on the dynamics of the model after a change in the remittance parameter d. In the steady state, equations (2.10) and (2.11) become: zf ( k) + 1 δ = 1 β (2.13) c = zf( k) δ k + d (2.14) The phase diagram of the dynamic system in the c and k space is presented in Figure 2. Suppose initially that the economy is at a long run equilibrium at point A and does not receive remittance, d = 0. Now suppose there is an unanticipated inflow of remittances that is expected to be permanent, d > 0. The Euler equation is independent of remittance so it will not shift. However, the unanticipated permanent remittance shock will shift the 10 Annen and Kosempel (2012) studied the impact of foreign aid in a closed economy. Remittances and certain types of foreign aid, being unilateral transfers, have similar properties specially in their impacts to recipient households. Hence, their results for foreign aid can be carried over for the case of remittances. 55

64 c c t = c t+1 B A d > 0 d = 0 0 k Figure 2.2: Effects of a permanent increase in international worker remittance flows. k resource constraint upwards. The extent of the shift in the resource constraint will depend on the size of the remittance. Once the inflow of remittances arrive the saddle path instantaneously shifts and the new long run equilibrium moves to point B. A permanent increase in remittances produces two effects: First, there will be an increase in steady state consumption; and second, the level of steady state capital is unchanged. Agents will want to smooth their consumption profile faced with a remittance shock that is permanent. Thus, they consume all of the remittance per period after the change and leave nothing for saving. Since the saving rate associated with a permanent change in remittance is zero, it will not be effective in raising output per capita. 11 Suppose now that the increase in remittances is temporary and its terminal date is known with certainty. The transitional dynamics of a temporary increase in remittances and its macroeconomic effects is shown in Figure 2.3. Given the unanticipated remittance inflows the level of consumption jumps to point B. The size of the jump from points A to B and 11 Output may decrease if leisure is included in the model. This result stems from standard macroeconomic intuition that since leisure is a normal good, leisure rises in the face of an unanticipated inflow of remittance. Consequently, hours worked drops as well as output per capita. 56

65 c c t = c t+1 B 1 A C 2 d > 0 d = 0 0 k Figure 2.3: Effects of a temporary increase in international worker remittance flows. k k the movement from B to C (segment 1) depends on how long the temporary remittance is expected to last. If remittance is expected to last for a very short period of time, then most of the remittance inflow will be saved and the increase in consumption will be very small. This leads to a temporary increase in capital up to point C. On the other hand, if the remittance inflow is expected to last for a long period of time, marginal propensity to save will be low and much of the remittance will be consumed. These explanations follow the permanent income hypothesis: given a temporary increase in remittance, agents must optimally invest some of the remittances received to maintain a smooth consumption profile. Once the remittances run out the economy reverts back to its initial equilibrium in point A from point C (segment 2). The restricted version of the model provides a good framework to describe the long run effects of remittances. However, it does not allow one to properly quantify the effectiveness of remittances because there is no uncertainty and its impact differ across countries. To better quantify the effects remittance have on macroeconomic performance, it will now be convenient to revert back to the stochastic version of the model and proceed to the calibration exercise. 57

66 Table 2.1: Business cycle statistics from data for remittance recipients for the period All remittance Remittance-GDP ratio Variable recipients > 10% 10% 1% < 1% (x) σ x ρ xt,gdp t σ x ρ xt,gdp t σ x ρ xt,gdp t σ x ρ xt,gdp t GDP GNP Consumption Investment Trade balance/gdp Remittance Remittance/GDP 4.0% 14.4% 3.6% 0.4% Note: Data were expressed in logs and detrended using HP filter with λ = Table 2.2: Model parameter values. d γ α δ n r β ω z θ ξ z σ z ψ ξ d σ d φ b Calibration and Simulations Data and Business Cycles This article follows the World Bank definition of worker remittance as the sum of personal transfers and compensation of employees. 12 Data for worker remittance was obtained from the World Development Indicators database. Data for other macroeconomic aggregates were sourced from the Penn World Tables and UNCTAD database. Following Kydland and Prescott (1990), business cycles are defined as the deviations of macroeconomic aggregates (i.e. output, consumption, investment, remittance) from trend, and business cycle facts are the statistical properties of co-movements of these aggregates with respect to deviations from trend of GDP per capita. When examining business cycle 12 Personal transfers include all current transfers in cash or in kind between resident and nonresident individuals, independent of the source of income of the sender and the relationship between the households. Compensation of employees represents remuneration in return for the labor input to the production process contributed by an individual in an employer-employee relationship with the enterprise. 58

67 Table 2.3: Permanent changes in share of remittance to output per capita. Remittance to GDP ratio Steady state values 0% 1% 4% 10% 20% Consumption/GDP Investment/GDP GDP Consumption Investment Hours worked aspects of the data, each data series was detrended using the Hodrick and Prescott (1981) filter. For any series x t for t = 1, 2,..., T, the HP filter extracts a trend component τ t and a cyclical component s t = x t τ t by minimizing the loss function: T T (x t τ t ) λ [(τ t+1 τ t ) (τ t τ t 1 )] 2 (2.15) t t where λ is a weight that reflects the relative variance of the two components. Since remittances data are available annually the parameter λ was set to 6.25 as suggested by Ravn and Uhlig (2002). Data were expressed in logs to capture percentage deviations from trend. Following Baxter and King (1995) the first and last observations were dropped prior to calculating the moments of the business cycle components. This ensures that the results are not influenced by behavior in the end-points of the filtered series. Finally, contemporary cross-serial correlations of the cyclical components of remittance and output were computed. As is customary in the macroeconomic literature, a positive contemporaneous correlation implies that remittances are procyclical and a negative contemporaneous correlation as countercyclical. Following Pallage and Robe (2001), correlation is judged to be non-different from zero if it lies in the interval (-0.29,0.29). 59

68 Table 2.4: Model economy business cycle statistics Productivity and Productivity Remittance Direction of Cyclicality Variable Remittance Shocks Shocks Only Shocks Only Pro (θ > 0) Counter (θ < 0) (x) σx ρxt,gdp t σx ρxt,gdp t σx ρxt,gdp t σx ρxt,gdp t σx ρxt,gdp t GDP GNP Consumption Investment Trade balance/gdp Remittance n.a. n.a Note: Data were expressed in logs and detrended using HP filter with λ = The re-calibration of the model for different values of θ was conducted purely for sensitivity analysis. 60

69 Table 2.1 reports the standard deviations and cross serial correlations of output and other economic aggregate time series data for 49 remittance recipient countries. Proper computation of business cycle stylized facts require that all observations are non-missing across variables and contiguous in the time dimension. As a result, a recipient is excluded from Table 1 if fewer than 20 consecutive years of data are available for that country. Regression analysis does not have the same requirement. Hence, in the regression analysis later in this article, there were more recipient countries (81) included in the sample. As shown in Table 2.1, investment and remittances are four and almost ten times more volatile than output, respectively. Overall, there is not much difference between GDP and GNP in terms of volatility. Consumption is more volatile than output which suggest difficulty in consumption smoothing, a stylized business cycle fact among developing countries (Rand and Tarp, 2002). Remittance is ayclical for 35 countries (or 71%) in the sample. Trade balance to GDP ratio is almost as volatile as output and is countercyclical on average. Also reported are business cycle statistics according to the share of remittance to output of recipient countries. Remittances are more volatile for recipient countries that receive little (less that 1% of output). There is not much difference in business cycle statistics across other economic aggregates for remittance recipients Calibration To quantify the model one must specify functional forms to be used in the simulations. This study abides by the common practice in the macroeconomic literature and specify the utility function to be: u(c t, l t ) = (cω t l 1 ω t ) 1 γ 1 1 γ (2.16) where γ > 0 is the coefficient of relative risk aversion and ω > 0 indicates the importance 61

70 of consumption relative to leisure in determining instantaneous utility. The production function is specified to be Cobb-Douglas: f(k t, n t ) = k α t n 1 α t (2.17) where α (0, 1) is capital s share parameter. The capital adjustment cost function is assumed to be quadratic: Φ(x t ) = φ 2 (x t) 2 (2.18) where φ > 0 and x t = k t+1 k t. The specification of Φ( ) implies that net investment, whether positive or negative, generates resource costs. Finally, following Schmitt-Grohe and Uribe (2014), the country interest rate premium takes the form: p(b t ) = ψ(e bt b 1) (2.19) where ψ > 0 and b are parameters. The model will be simulated numerically following the method described in King et al. (1988). As shown in Table 2.2, parameter values are set so that the model s properties match averages from data for remittance recipients. The value for the parameter in the utility function ω was set such that the average time spent working n of 30%. 13 Similarly, the discount factor β was set such that the average annual real interest rate r is 4%. The steady state value in the stochastic process for remittances d is set to which implies an average remittance-gdp ratio of 4% for the sample of countries. The parameters for standard deviations of the innovations in the stochastic processes, σ z and σ d, were set to and 0.615, respectively, to match the average annual standard deviations for produc- 13 In the RBC literature estimates for average hours worked varies from 1/5 to 1/3. 62

71 tivity and remittance shocks in the sample. For baseline calibration ξ d is set to 0.67 which represents the median value estimated from the sample of countries. 14 In the remittance process a sensitivity analysis will be performed on the parameter θ. For baseline calibration θ is set to zero which implies that the reason for giving remittances is independent to the state of economic conditions in the recipient countries. The parameters φ and ψ were set to 0.17 and 0.004, respectively, to produce reasonable amount of standard deviations for investment and trade balance, as well as their correlations to output. The next set of parameter values were selected on the basis that they have been previously used in the macroeconomics literature. The values for b, γ and α were set equal to , 2 and 0.32, respectively, following Schmitt-Grohe and Uribe (2003). The depreciation rate δ was set to 10% per annum. The value of the autocorrelation coefficient in the technology process ξ z is set equal to 0.81, which is the annual equivalent to 0.95, used for quarterly series by Prescott (1986). The steady state value of the productivity shock z only affect the scale of the economy and can therefore be normalized to Permanent changes in remittance flows Table 2.3 reports the effects of permanent changes in the remittance-gdp ratio. The baseline calibration is set at a remittance-gdp ratio of 4%. When remittances increase, the investment rate remains the same, and the propensity to consume rises by one percentage point for each percentage point increase in the remittance-gdp ratio; this implies that permanent increases in remittance are consumed rather than invested. An increase in remittances produce a positive income effect, and since leisure is a normal good it rises and hours worked falls. The fall in hours worked causes a reduction in the levels of output, investment, and capital. In the model, each one percentage point increase in the long run 14 Regression results for the autocorrelation coefficient in the remittance process has a sample mean value of 0.64, a minimum value of 0.18, a maximum value of 0.93 and a standard deviation of

72 remittance-gdp ratio reduces GDP per capita by approximately Temporary changes in remittance flows Table 2.4 reports the simulated business cycle statistics. The first column reports the baseline simulation. Statistics generated by the model are consistent with business cycle stylized facts: investment and consumption are procyclical, investment is much more volatile than output and consumption, and trade balance is countercyclical. Remittance is acylical in the model which is consistent with the data for remittance recipients. However, the model does not do a good job in matching consumption and trade balance volatility in the data. The second column reports business cycle statistics for the case where there are no remittance shocks. The business cycle statistics for variables other than remittances are similar with or without remittance shocks because remittance shocks are being drowned out by productivity shocks. This result is highlighted by switching off the productivity shock while keeping the remittance shock. In the third column one can see that volatilities of output, consumption and investment drop significantly. Without productivity shocks, remittance shocks account for only around 24% of volatility in output at most. The relative strength of the productivity shock explains why remittance is ayclical in the model. Virtually all of the volatility in aggregate variables (other than remittance) is explained by the productivity shock. In the benchmark calibration remittance was assumed not to be related to the productivity shock and θ was set to zero. In columns 4 and 5 of Table 2.4 a sensitivity analysis on θ was conducted. In column 4 the value of θ was set to 8.9 to produce a remittance-output correlation coefficient of Column 5 reports the simulation with θ set to -1.8 to produce a correlation of Comparing columns (4) and (5) with the benchmark calibration in (1), except for the correlations, one can find very little difference in output volatility. It 64

73 turns out any cyclical relationship between remittance and output created in the model by changing the sign of θ does not reveal anything about a causal effect of remittance on output. It simply indicates that remittance levels are responding to the same variable causing output fluctuations. 15 The impulse responses of consumption, investment, output and hours to a positive one percent deviation from trend in the remittance level are shown in Figure 2.4. The income effect brought about by a one percentage deviation from trend in remittances leads to a decrease in hours worked, thereby leading to a decrease in output. Demand for the foreign asset rises on impact (not shown). However, since the remittance shock is temporary, agents optimally smooth their consumption profile inducing them to save. The rise in saving leads to an increase in investment, which in turn leads to a positive effect on output in the future. It is important to note from these simulations that the impact of remittances to economic aggregates is very small in magnitude Sensitivity Analysis The impact of remittances is expected to be not uniform across countries and this is driven by three causes: First, some countries simply receive more remittances than others. For instance, the impact of remittances may be felt more in countries like Lesotho where remittances account for about 25% of their GDP. Raising the share of remittances to output by four times the benchmark of 4%, the model predicts that output volatility will increase by roughly 1.08%. This effect is rather small and is another confirmation that remittance shocks are being drowned out by productivity shocks. Second, countries experience remittance shocks with varying degrees of persistence. 15 This result also applies to other forms of unilateral transfers such as foreign aid, see Annen and Kosempel (2012). 65

74 % deviation x 10 3 OUTPUT Years after the shock % deviation 9 x 10 3 CONSUMPTION Years after the shock 0.08 INVESTMENT 2 x 10 3 LABOR SUPPLY % deviation % deviation Years after the shock Years after the shock Figure 2.4: Impulse response to a one-percent remittance shock. According to RBC theory, the lower the persistence is the more immediate the impact of remittances on growth in output and the faster output is back again at trend. Lowering the persistence parameter to 0.33 (from the benchmark of 0.67), output volatility will peak at four periods after the remittance shock and quickly goes back again to trend after 20 periods. On the other hand, raising the persistence parameter to 0.94 not only lowers the effect of remittances to GDP but it also increases the number of periods for GDP to be back again to trend (at 60 years). And third, remittances are more volatile in some countries than others. The model predicts that doubling the remittance volatility raises output volatility by about 33%. Therefore, the impact of remittances may be felt more in countries that have high remittance volatility. 66

75 2.6 Cross-Country Empirical Evidence This section makes use of RBC theory to motivate a regression analysis as it is typically done in the empirical literature on effectiveness of worker remittances Econometric Specification This paper follows the econometric specification commonly used in the remittance effectiveness literature (see Chami et al and Catrinescu et al. 2006). Formally, the following regression model was estimated for each country i: g it = β 0 + β 1 y 0i + β 2 gcf it + β 3 wr it + u it (2.20) where g is growth in real GDP per capita, y 0 is log of initial real GDP per capita, gcf is the log of gross capital formation to GDP ratio, and wr is the log of worker remittance to GDP ratio. The previous sections have shown that the distinction between temporary and permanent component of remittance is important. Hence, in some specifications wr will be replaced with a component of remittance that is either temporary or permanent. The next step is to come up with a measure of temporary and permanent remittance. For each country i the following regression model was estimated: ln D it = ϕ i + η i t + v it (2.21) where ln D is the log of remittance per capita and t is a time trend. In this regression model, the permanent component of remittance is ˆϕ i + ˆη i t and the temporary component of remittance is the residual given by: ˆv it = ln D ˆϕ i ˆη i t (2.22) 67

76 From the theoretical model and simulations the hypothesis is that only temporary remittance was found to influence growth. If the hypothesis is true then by replacing wr with ˆv the estimated coefficient for β 3 should be positive and significant. 16 Panel data was constructed using 5-year averages from Countries included in the sample as well as some descriptive statistics are presented in the Appendix. Data for gross capital formation was collected from WDI Empirical Estimates The results of the empirical analysis are presented in Table 2.5. Columns (1)-(3) report estimates using OLS. The negative coefficients for log initial output suggest evidence for growth convergence. The estimates of the ratio of gross capital formation to GDP per capita are large and significant. Depending on the specification, around 21%-27% of the variation of growth is explained by the model using OLS. Column (1) reports the model estimate in accordance with Chami et al. (2003). The estimated coefficient for the log share of remittances to GDP per capita is positive but not significant. This result essentially reveals that there is no significant correlation between the share of remittance to output and growth in GDP per capita, consistent with those found in Chami et al. (2003). Column (2) reports the estimated coefficients with the permanent component of remittance. Similar to the result in column (1) there is no significant correlation between the permanent component of remittance and growth. Even the lagged permanent component is not significantly correlated with contemporaneous growth. These results are likewise consistent with RBC theory and confirm that growth effects of a permanent increase in 16 One can also de-trend the remittance series using filters. I have tried using HP filter to extract the temporary and permanent components and the results are almost identical with a linear de-trending method. 68

77 remittance are difficult to detect from the data. The results are different, however, if the permanent component of remittance is replaced with a temporary component. Shown in column (3) is the regression estimate using the temporary component. The evidence of the effectiveness of temporary component of remittance to growth is mild (at 10% level of significance). Lagged effect of temporary remittance is not siginificant. These results confirm the predictions of RBC theory that temporary fluctuations in remittances can influence economic growth. 17 It is possible that these results are not robust due to endogeneity of remittance. The model may not be able to properly identify the causal arrow from remittances to economic growth. For instance, a country performing poorly in terms of growth in GDP per capita may experience increased outward migration leading to more remittances sent back home. In this case, the OLS estimates presented earlier may be downward biased. To address the endogeneity problem a two-stage least squares (2SLS) regressions with lagged remittance as instrument were estimated. The results of 2SLS are shown in columns (4) and (5). The estimated coefficient for the permanent component remain positive and insignificant. On the other hand, the estimated coefficient for the temporary component remains positive and but is no longer significant. The estimated coefficient from 2SLS increased substantially suggesting a downward bias in the OLS estimate. From RBC theory, one can infer that the lower the persistence of remittance is the larger its immediate impact to GDP per capita growth. Table 2.6 presents the regression results for countries whose remittance persistence is less than the median value of The results are nearly identical as in Table 2.5 with the exception of the temporary component of remittance which is now significant as shown in column (5). These findings suggest that, even accounting for endogeneity, the permanent component of remittance is uncorrelated 17 The weak relationship between temporary remittances and economic growth may be due to a generated regressor problem. 69

78 with growth. Moreover, the regression estimates suggest a weak but positive (and sometimes significant) relationship between the temporary component of remittance and growth consistent with RBC theory. 2.7 Conclusion This paper develops a simple small open economy RBC model augmented with stochastic remittance shocks. According to RBC theory, only worker remittances that are temporary in nature have an impact on GDP per capita. The intuition is that agents facing a temporary remittance shock will optimally save to smooth their consumption profiles. This smoothing behavior leads to a rise in investment, acquisition of foreign assets, and, consequently, output. On the other hand, permanent increase in worker remittances does not have an impact on GDP per capita. From the calibration exercise it was found that the contribution of remittance to output volatility is very small. An increase in the share of remittances to output brings very little impact to output volatility. The predictions of the model were tested using panel data from 81 remittance recipient countries. A decomposition of remittance data in terms of permanent and temporary components was conducted. The regression results found that the temporary component of worker remittances have a positive but mild impact to economic growth. Permanent component of worker remittances, on the other hand, does not have such an impact. These results are robust even if we account for possible endogeneity bias in the estimations. The aforementioned results confirm the findings from the empirical literature that unilateral transfers, such as remittances, are not going to be very effective at raising output. The principal conclusion of this paper is that the only form of unilateral transfer that could be effective are transfers designed to improve productivity levels in recipient countries; 70

79 such as technical assistance, or possibly foreign aid for public infrastructure projects. 71

80 Table 2.5: Remittances and Economic Growth: Panel Estimation ( ) a Dependent Variable: OLS OLS OLS 2SLS 2SLS GDP Per Capita Growth (1) (2) (3) (4) (5) Log Initial Real GDP Per Capita ** ** ** * * (0.002) (0.002) (0.002) (0.002) (0.002) Log Gross Capital Formation/GDP 0.034*** 0.030*** 0.031*** 0.029*** 0.029*** (0.005) (0.004) (0.005) (0.005) (0.006) Log Remittance/GDP (0.014) Remittance Permanent (0.005) (0.001) Remittance Permanent (t-1) (0.004) Remittance Temporary 0.006* (0.003) (0.013) Remittance Temporary (t-1) (0.003) R-squared F statistic N a Robust standard errors are reported in parentheses. *** significant at 1%, ** significant at 5% and * significant at 10%. Constant, time and continent dummies were estimated but not presented. Columns 4 and 5 report two stage least squares estimation results with lagged remittance as instrumental variable. 72

81 Table 2.6: Remittances and Economic Growth for Recipients with ρ d 0.67: Panel Estimation ( ) a Dependent Variable: OLS OLS OLS 2SLS 2SLS GDP Per Capita Growth (1) (2) (3) (4) (5) Log Initial Real GDP Per Capita *** *** ** ** (0.003) (0.003) (0.003) (0.004) (0.005) Log Gross Capital Formation/GDP 0.036*** 0.031*** 0.028*** 0.027*** (0.006) (0.006) (0.006) (0.008) (0.018) Log Remittance/GDP (0.015) Remittance Permanent (0.009) (0.001) Remittance Permanent (t-1) (0.008) Remittance Temporary * (0.005) (0.034) Remittance Temporary (t-1) (0.005) R-squared F statistic N a Robust standard errors are reported in parentheses. *** significant at 1%, ** significant at 5% and * significant at 10%. Constant, time and continent dummies were estimated but not presented. Columns 4 and 5 report two stage least squares estimation results with lagged remittance as instrumental variable. 73

82 Table 2.7: Appendix: List of recipient countries and descriptive statistics. Country g d/gdp log(gcf) σ y σ c σ i σ d ρ yc ρ yi ρ yd ξ d Algeria Antigua and Barbuda Argentina Bangladesh Barbados Belize Benin Bolivia Botswana Brazil Burkina Faso Cabo Verde Cameroon China Colombia Congo, Rep Costa Rica Cote d Ivoire Cyprus Dominica Dominican Republic Ecuador Egypt, Arab Rep El Salvador Ethiopia Fiji Gabon Ghana Grenada Guatemala Guinea Guinea-Bissau Honduras India Indonesia Israel Jordan Kenya Korea, Rep Lao PDR Lesotho Madagascar Malaysia Mali Malta Mauritania Mexico Morocco Mozambique Namibia Niger Nigeria Oman

83 Table 2.8: Appendix: List of recipient countries and descriptive statistics, cont. Country g d/gdp log(gcf) σ y σ c σ i σ d ρ yc ρ yi ρ yd ξ d Pakistan Panama Papua New Guinea Paraguay Peru Philippines Rwanda Samoa Senegal Seychelles Sierra Leone South Africa Sri Lanka St. Kitts and Nevis St. Lucia St. Vincent Sudan Suriname Swaziland Syrian Arab Republic Thailand Togo Trinidad and Tobago Tunisia Turkey Vanuatu Venezuela, RB Yemen, Rep Notes: The following are the description of the column headings: g average growth rate in real GDP per capita, d/gdp average share of remittances to GDP per capita, log(gcf) average log capital formation, σ y volatility of output, σ c volatility of consumption, σ i volatility of investment, σ d volatility of remittance, ρ yc contemporaneous correlation of output to consumption, ρ yi contemporaneous correlation of output to investment, ρ yd contemporaneous correlation of output to remittance, ξ d is the autocorrelation coefficient of remittances. 75

84 Chapter 3 International Worker Remittances and Human Capital Formation Across Countries 3.1 Introduction Remittances refer to resources, such as money and goods, that are transmitted to households by migrant workers working in host countries. For many developing countries shown in Figure 3.1, remittance inflows represent a significant source of foreign exchange earnings, even exceeding private capital flows, official development assistance or foreign direct investment (The World Bank, 2014). Given its scale, there is now a growing interest on how remittances are spent and whether their use impacts the economic development of recipients. How are these remittances spent or used? Are remittances spent on consumer goods back home, or are they channeled into human and physical investments? Adams and Cue- 76

85 cuecha (2010) summarized the current literature and identified at least three interrelated views on how remittances are spent or used and their effect on economic development. The first view is that remittances are fungible. In other words, remittance income is spent just like any other source of income. The second view argues that the receipt of remittances can cause behavioral changes such that remittances tend to get spent on consumption rather than investment goods (Chami et al., 2003). The third view is that since remittances are a transitory type of income households tend to spend them more on investment goods (human and physical capital investments) than on consumption goods, and that this can contribute positively to economic development (Annen et al., 2014; Yang, 2008). Taking a cue from the third view, the purpose of this article is to refine and extend the debate concerning how remittance inflows are allocated and their impact on economic development. Remittances are likely to affect employment and consumption decisions of their recipients as well as the composition of their investment expenditures (Ngoma and Ismail, 2013). From an investment perspective, one feasible indirect effect of migrant remittances have on economic development is its effect on human capital formation through education. Precisely, to study the impact remittance inflows have on economic development, this article builds a real business cycle model with two productive sectors: a sector that produces physical goods and another sector that produces human capital. From a methodological point of view, the real business cycle model is attractive in the sense that: First, it allows one to quantify the general equilibrium effects of altering how remittances are allocated across sectors in the economy; Second, the model can be calibrated to mimic the economy of a typical remittance recipient; And third, counter factual studies can also be conducted to measure the economy-wide impacts of raising the levels of remittance inflows. This article is closely related to four studies which examined the impact of remittances using calibration 77

86 (a) India (b) Mexico (c) Philippines (d) Bangladesh (e) El Salvador (f) Nigeria Figure 3.1: Composition of various inflows to selected countries,