An Empirical Study of Remittances and Growth in the Developing World By Elird Haxhiu At the microeconomic level, development is about analyzing the lives of the poor. How do citizens in the developing world live? How do they survive? While these important questions can be answered and studied in many ways, a common theme is surely about how they make a living. How much income do lower income countries have? Where do they get it from? Most importantly, how do they use it? This research paper concerns itself with the issue of how the less privileged parts of society procure the funds necessary to survive on a daily basis. Although wage labor, microfinancing, and government transfers undoubtedly play an important role, this paper will take up the issue of remittances and the role they play in developing countries. Remittances are the financial funds emigrants transfer back to their native countries. In other words, individuals who migrate to other countries often work abroad at higher wages and send a share of their wages back home to family members, friends, or business associates. Development economics may be crudely defined as the study of how poor countries become rich. Although a more nuanced definition would substitute the word poor with the word developing and highlight the key areas of human existence that are associated with development (such as access to adequate healthcare or the choice to pursue education). Regardless, the fundamental issue remains: people in developed countries seem to enjoy richer and fuller lives than their counterparts in developing nations. Whether it is through the development of broad theories that explain capital accumulation or in-depth case studies that seek to understand why some countries are successful at providing dignified lives for their citizens, the end goal is the same. Development economics is about how to help poor countries become rich. If this simple but elusive idea can be understood, developing countries today can take the steps necessary to encourage their development. Why should development economics care about remittances? The first reason is the growing importance of remittances around the world. According to World Bank estimates between 2000 and 2010, the amount of officially recorded remittances received by all countries jumped from about $131.5 billion to $440.1 billion U.S. dollars, representing an increase of over 234 % (World Bank, 2011). More importantly, over 73% of all remittances were sent to citizens in the developing world (World Bank, 2011). Thus, another important reason to study remittances is the role they play in economies of developing nations. Table 1 below presents the 15 countries that receive the most remittances as a share of their gross domestic product (GDP) in the year 2013. It is worth emphasizing that a different ranking would exist in terms of total remittances received; however, remittances as a share of output are the most appropriate measure to assess their impact in the economies of developing nations. For example, although India may receive much more remittances in absolute value than Haiti, those funds are negligible in boosting purchasing power since India s economy (and its population) is far greater than Haiti s.
An Empirical Study of Remittances and Growth in the Developing World Elird Haxhiu Table 1: Top 15 Countries receiving remittances as a share of GDP in 2013 Country Remittances (millions $) GDP (millions $) R/GDP Tajikistan* 3,582 3,945 91% Kyrgyz Republic* 2,278 3,576 64% Moldova* 1,976 4,044 49% Nepal* 5,552 11,370 49% Samoa 165 443 37% Haiti 1,781 4,883 36% Armenia 2,193 6,875 32% Guyana* 328 1,070 31% Liberia* 360 1,286 28% Lesotho* 543 2,029 27% Hondorus* 3,136 12,771 25% Uzbekistan* 6,633 27,198 24% Lebanon 7,551 32,347 23% Gambia* 181 841 21% Tonga 60 280 21% Source: Author s calculations based on World Bank Data *Categorized as medium to low Human Development Index (HDI) score Figure 1: Top 15 Remittance Sending Countries in 2013 (millions of dollars) 10
The Hinckley Journal of Politics 2015 It is clear to see from the table above that remittances play a large role in the countries listed, ranging from one fifth of domestic output in Gambia and Tonga to over 90% of GDP in Tajikistan. More importantly, however, most of the countries in the list above are considered to be underdeveloped. According to the 2013 United Nations Human Development Report, the nations with an asterisk in Table 1 were categorized as having medium or low human development according to their Human Development Index (HDI) score. This means that they all ranked below the top 103 countries in the world in terms of healthcare, education, and per capita income, which are the three main components of HDI (UN Human Development Report, 2013). Subsequently, remittances warrant careful study not only because of their magnitude worldwide, but also due to the critical role they play in the developing world. A natural extension of the data presented in Table 1 is a measure of the top countries from which remittances originate. In other words, what are the top 15 host countries for immigrants that send remittances back home? Figure 1 presents these countries and their contributions to global remittances in millions of dollars in 2013. It is immediately clear that the U.S. retains its reputation as a nation of immigrants. In 2013, immigrants residing in America sent over $52 billion U.S. dollars to friends and family in their native countries. Another striking characteristic of Figure 1 is that all of the top 15 remittance-sending countries are rich and wield considerable political power at the international level. Russia, Germany, France, the Netherlands, Spain, and Italy all join the U.S. as a source of over $1 billion U.S. dollars in 2013 in the form of remittances. This fact seems to confirm the intuitive notion that poor citizens in the developing world migrate to countries where opportunities for employment are brighter so that they, as well as their families back home, can improve their economic situations. Thus far, two reasons have been advanced for the study of remittances: they are large relative to national and global output levels, and there seems to be a link between rich countries sending and poor countries receiving remittances. A third justification is advanced by Michael Todaro and Stephen Smith (2012) in their Economic Development textbook. Although there are clearly benefits to citizens of developing nations (who have added disposable income to spend on food, clothing, shelter and other goods and services), the authors note that there may also exist some hidden costs to remittances. For example, the authors note that there are legitimate concerns that outmigration can hamper development prospects because of the loss of skilled workers via brain drain (Todaro & Smith, 2012, 695). While temporary relief may be provided by the migration of workers from a developing country to a richer one, the long-term costs may be significant. A labor force depleted of educated and driven individuals can doom a nation to perpetual economic stagnation and dependency. Remittances and the Process of Economic Development The final reason that studying remittances is warranted provides the foundation for the empirical enquiry of this paper. There exist valid theoretical and intuitive reasons to believe that remittances contribute to increases in the rate of growth of developing nations. The first mechanism which might explain this causal link is the Keynesian notion that increases in aggregate demand is bound to produce increases in output. With more money to spend on goods and services, citizens receiving remittances not only benefit themselves, but also incentivize suppliers to increase production. This increased production is also likely to be associated with increases in employment levels. Thus, the engine of economic growth is jump-started as the now-employed individuals will have even more income to spend in the economy, further boosting aggregate demand, production, and employment levels. A second mechanism which explains this causal link is the relief that remittances may grant to government discretionary budgets. If citizens find that they can satisfy a portion of their consumption needs from remittances, the necessity for government transfers will be reduced. Wise governments in the developing world could theoretically use this relief to devote a greater share of their revenues toward enhancing their productive structures and moving toward industrialization. As Developmentalist economists such as Paul Rosenstein-Rodan and Ragnar Nurkse argued in the 1940s and 1950s, it is only through a radical transformation toward a manufacturing based economy that a nation can develop (Cypher, 2009). Which key economic agent can (and should) start this transformation? The Developmentalists argued that only the government is large and powerful enough to find and coordinate the big push necessary to jump start industrialization (Cypher, 2009). The relief granted by remittances may be large enough to help governments in the developing world fund such actions. Thus, theory dictates that increases in remittances lead to increases in output. This paper seeks to investigate the theoretical relationship between increased remittance levels and growth, as measured by gross domestic product (GDP) and controlling for all other relevant factors. A multivariate regression model will be estimated for a single cross-section of countries by the Ordinary Least Squares (OLS) method to determine if the relationship holds. In addition, another OLS regression model will be estimated to obtain a better understanding of the determinants of global remittance flows. This second model will seek to explain the level of remittances that each particular country receives. Consequently, a better idea of these determinants will be obtained before the model of interest is specified and estimated. Mathematically, it is evident that in order to estimate these two econometric models, sufficient variation must exist on the dependent variable. Table 2 below provides basic descriptive statistics for GDP growth in the year 2013. Observations exist for 183 countries in the dataset generated from the World Bank Database. The average rate of GDP growth in 2013 for the world was 3.25% with a large standard deviation of 4.84. The ratio of the standard deviation to the mean, which is greater than one for the entire sample, shows that there is enough variation to estimate the models above. Table 2: Descriptive Statistics for GDP Growth in 2013 Quartile Observations (N) Mean (μ) Standard Deviation (σ) [σ / μ] Total 183 3.25 4.84 1.49 1st 46-1.50 5.68 3.79 2nd 46 2.33 0.58 0.25 3rd 46 4.18 0.57 0.14 4th 45 8.11 3.71 0.46 Another interesting aspect of the table above is that the variation in growth rates for poor countries (i.e., those in the first quartile of the sample, which had an average negative growth rate) is far greater than the variation for richer countries. This should further facilitate the estimation of the regression models above, where the focus will be on explaining growth (or the lack thereof) as a function of remittance levels. A more detailed explanation of the specification of these models as well as their estimates will be 11
An Empirical Study of Remittances and Growth in the Developing World Elird Haxhiu presented in the third section of this paper. The following section contains a brief review of the remittance literature in development economics, with a focus on past empirical work that attempted to use remittances as a predictor of growth rates in the developing world. Literature Review Although there is a wide body of literature on the effects of remittances on economic growth, the findings have at best been mixed. Some studies have found positive effects while others report models which imply zero or negative impacts on output. For example, Natalia Catrinescu et al. (2006) summarize the two opposing forces that shape the impact of remittances in their working paper Remittances, Institutions, and Economic Growth. They use a panel dataset consisting of over 120 countries from 1991 to 2003 to estimate numerous cross-sectional OLS models as well as fixed-effects models that control for unobserved heterogeneity. The dependent variable in their models is the annual growth rate of GP, while the regressor of interest is specified as remittances as a share of domestic output. Likewise, Catrinescu et al. (2006) note that in many developing countries remittances drive economic growth through their impact on disposable income. However, there may also be harmful impacts on output as a result of remittances. The authors identify three explanations for the notion that remittances reduce economic growth: they can fuel inflation, hurt the export sector by appreciating the real exchange rate, and reduce the labor force participation of those receiving remittances (Catrinescu et al., 2006). Despite this, the authors do find that models that account for endogeneity concerns indicate that remittances make a positive, albeit modest, contribution to growth (Catrinescu et al., 2006, 11). Another key theoretical insight offered in the literature is that institutions in developing countries play an important role in efficiently investing remittances, which may also lead to increased growth (Catrinescu et al., 2006). A similar conclusion is drawn by Dilip Ratha (2013) who surveys numerous empirical projects to show that the extent to which remittances promote economic growth depends The relief granted by remittances may be large enough to help governments in the developing world fund industrialization. Thus, theory dictates that increases in remittances lead to increases in output. strongly upon the institutions in place in the developing country, as well as the general macroeconomic environment at any particular time (Ratha 2013). The author explains this finding by suggesting that remittance are different from other monetary flows because they are countercyclical and often increase during times of hardship. Therefore, remittances may actually work toward reducing the depth and breadth of peaks and troughs in the business cycle experienced by recipient households, rather than strictly promoting increases in domestic output through their impact on consumption spending (Ratha, 2013, 2). While Catrinescu et al. (2006) and Ratha (2013) have found positive impacts on growth, Adolfo Barajas et al. (2009) present much bleaker findings in their work. They also work with a panel dataset consisting of 90 countries, but specify only fixed effects models using instruments to account for potential endogeneity in worker remittances. Unlike Catrinescu et al. (2006), these authors specify per capita GDP growth while using remittance levels as the main regressor of interest. Here, they find that decades of private income transfers remittances have contributed little to economic growth in remittance-receiving countries and may have even retarded growth in some (Barajas et al., 2009, 16). The authors claim that developing country achieving remittance-driven growth has been completely unsuccessful. Moreover, they find this fact quite peculiar considering the large number of poor countries that receive remittances in excess of 10% of GDP. One explanation for the lack of observed correlation offered by Barajas et al. is the manner in which remittance funds are used: they are not intended to serve as investments but rather as social insurance to help family members finance the purchase of life s necessities. Remittances lift people out of poverty, but they do not typically turn their recipients into entrepreneurs (Barajas et al., 2009, 17). These results are echoed by Michael Clemens and David McKenzie (2014) in their aptly titled paper Why Don t Remittances Appear to Affect Growth? After briefly presenting previous research studies that find no statistically significant or economically important relationship between remittance levels and economic growth, the authors develop three theoretical explanations. The first is that the recent growth in remittance flows (like the 234% increase from 2000 to 2010 referenced in the introduction of this paper) is illusory and the product of changes in measurement, not changes in real financial flows (Clemens & McKenzie, 2014, 3). The second reason is that cross-country regressions much like the fixed effects regression models specified and estimated by Barajas et al. (2009) and Catrinescu et al. (2006), do not have enough power to detect the truly microeconomic impacts of remittances on growth. Remittance funds are assumed to be spent primarily on the basic consumption needs of citizens in the developing world, rather than on large-scale investments that might have an immediate impact on local or national output. Moreover, the spending decisions of those who receive remittances are decentralized rather than coordinated efforts to increase economic activity. As the final explanation for the missing link between remittances and growth, Clemens and McKenzie note that the greatest driver of rising remittances is rising migration, which has an opportunity cost to economic product at the origin. Therefore, it is not surprising that no observable impacts on domestic output have been found (Clemens & McKenzie, 2014, 25). Despite conclusions to the contrary, numerous country-specific statistical studies have found a positive relationship between remittance levels and economic growth. Siddique et al. (2010) investigated this relationship in the countries of Bangladesh, India and Sri Lanka using time series data and employing the Granger causality test. Their results were mixed; while no relationship was discovered in India, increases in remittances did lead to growth in Bangladesh. Interestingly, a two-way causal relationship was discovered in Sri Lanka, where remittances did positively affect economic growth but growth also had a marginal impact on remittance levels themselves (Siddique et al., 2010). Similar results were found by Orrenius et al. (2009) who investigated the statistical relationship between remittances and proxy indicators for economic growth in the federal states of Mexico between 2003 and 2007. However, their dependent variable was not aggregate economic growth, but rather the proxy variables of wages and employment in the formal sector, unemployment rates, wage inequality, and school enrollment rates (Orrenius et al., 2009, 2). The Two Stage Least Squares (2SLS) model utilized the log of wages and employment and the level interpretation of the unemployment rate. Moreover, they differentiated their fixed effects regressions between the full dataset and a subset of Mexican states with high migration rates. The authors discovered in both regressions that remittances lead to improved labor market conditions, with higher employment and lower unemployment rates (Orrenius et al., 2009, 14). Another more aggregate-level study of 37 African countries from 1980 to 2004 by Bichaka Fayissa and Christian Nsiah (2008) affirmed that a causal relationship does indeed exist. However, remittances impacted economic growth mostly in countries where the financial systems are less developed by providing an alternative way to finance investment and helping overcome liquidity constraints (Fayissa & Nsiah, 2008, 12). 12
The Hinckley Journal of Politics 2015 In sum, empirical results on the hypothesized relationship between remittances to developing countries and growth in output are quite mixed. Some authors find a positive relationship in both country-specific case studies as well as datasets aggregating numerous developing nations. Others, however, fail to find such a relationship and conclude that the marginal impact of remittances on GDP growth is truly zero in the population. There even exists research that finds that remittances have a negative impact on economic growth in the developing world due to the high cost of migration that is associated with it. While this paper in unlikely to settle the score, it does take numerous novel approaches to specifying the econometric models used to investigate the elusive relationship between remittances and growth. The following section describes the dataset used, the specifications of the two models be estimated, and the presentation of the results. The final section discusses conclusions and makes suggestions for further research on this important topic. Empirical Analysis Explaining Remittances The data used in this empirical analysis are based on a combination of four World Bank datasets, with observations on 214 nation-states from 2000 to 2013 (these datasets and their online locations can be found in the references section of this paper). Despite this availability of data, both regression models in this section are estimated in 2010. The reason for this is that one of the key independent variables is only observed in this year. The first OLS regression model to be estimated seeks to explain remittance inflows in all countries in the dataset during the year 2010. The total number of independent variables specified is six, although the regression table below presents various specifications for analytical purposes. The following equation specifies the overall regression model, followed by a detailed explanation of each variable as well as the expected signs of coefficients. rem=β 0 +β 1 emig+β 2 pcgdp+β 3 indep+β 4 unemp+β 5 pubhlth+β 6 rural+ε 1. Remittances: This continuous variable is the dependent variable in the regression model above and is officially defined as the amount of migrant remittance inflows into a particular country in millions of U.S. dollars during the year 2010. 2. Emigrant Stock (+): This continuous variable measures the number of citizens in a particular country living abroad during the year 2010. A positive sign is hypothesized for this variable because of the very definition of remittance flows; increases in the number of citizens living abroad should lead to greater remittance levels in the home country because there are simply more people available to send that money. 3. Per Capita GDP ( ): This continuous variable measures per capita gross domestic product in a particular country in 2010. It is included as a control variable, but its negative hypothesized sign is intuitive; the larger per capita income in a certain country, the less is the need for remittances to be sent from abroad. 4. Years Independent ( ): This continuous variable measures the years a country has been independent in 2010 by the formula: (2010 T), where T is the year a country gained independence. The coefficient for this regressor is hypothesized to be negative since countries that have gained independence from colonial rule earlier have had more time to develop, thus reducing the need for remittance flows. Thus, increases in the years that a particular country has been independent should lead to reductions in remittance inflows in 2010. 5. Unemployment Rate (+): This continuous independent variable measures the national unemployment rate in a particular country in 2010. Once again, this regressor is included as a control variable; it is expected to have a positive marginal impact on remittances, as increases in unemployment mean that more people are lacking jobs and wages, thus increasing the need for support from family members abroad. 6. Public Health Spending ( ): As with per capita GDP and the unemployment rate, this continuous variable serves as a control in the regression model. It is defined as the amount of public health expenditures as a share of domestic output in a particular country in 2010. Public health spending is expected to have a negative marginal impact on the dependent variable in this model. The need for support from family members abroad should be reduced when the government increases their spending on healthcare projects that serve the public. 7. Rural Population (+): This continuous explanatory variable measures the rural population level in a certain country in the year 2010. The hypothesized sign for the coefficient is positive as the poor tend to be concentrated in rural areas in the developing world. If there are more people living in rural areas, it is reasonable to predict that there are more poor people in that country needing assistance from friends or family living and working abroad. Thus, increases in the value of this variable should lead to increases in remittances in 2010. Table 3 below presents basic descriptive statistics on these variables, including the number of observations, the mean, and the standard deviation. Based on Table 3, the average level of remittances flowing into all countries in 2010 was more than $2.6 billion with a standard deviation of more than $6.6 billion. These magnitudes are not surprising given the information covered in Table 1 and Figure 1. However, it is clear that there is significant variation around the mean of remittance inflows, making the estimation of a regression model both possible and useful. Similar results hold for the other variables. Table 3: Descriptive Statistics for Relevant Variables Variable Observations (N) Mean (μ) Standard Deviation (σ) Remittances 173 2,645 6,648 Emigrant Stock 213 953,036 1,723,483 Per Capita GDP 187 10,320 15,261 Years Independent 189 139 251 Unemployment 174 9 6 Public Health 188 4 2 Rural Population 212 1.570x10 7 74,900,000 Table 4 below presents the OLS estimates for six different regression specifications, all explaining variations in the dependent variable (Remittances in 2010). The first model is a simple bivariate regression model with the regressor Emigrant Stock explaining Remittances. According to the adjusted-r 2, the model explains over a quarter of the variation in remittances with the single independent variable Emigrant Stock. Moreover, the estimate for the independent variable is a statistically significant explanatory variable 13
An Empirical Study of Remittances and Growth in the Developing World for Remittances at the.01 significance level, meaning that we reject the null hypothesis that β 1 =0 in favor of the alternative hypothesis that β 1 0. The coefficient of 0.0021 means that every additional citizen living abroad leads to an average increase in remittances of $2,100. Table 4: OLS Regression Models - Remittance Inflows in 2010 (millions $) Regressor (1) (2) (3) (4) (5) (6) Emigrant Stock 0.0021 0.0021 Per Capita GDP 0.0153 (0.607) 0.0020-0.0066 (0.850) Years Independent 4.02 Unemployment Rate Public Health Expenditures (0.070) 0.0026 0.0331 (0.485) 3.29 (0.0137) -49.47 (0.595) -389.31 (0.231) -0.00007 (0.790) 0.0140 (0.634) 2.26 (0.099) 7.87 (0.891) 118.07 (0.562) Elird Haxhiu explanatory variables. Even more impressive is the increase in the explanatory power of the model with rural population included; the adjusted-r 2 of 0.7544 suggests that over three quarters of the variation in remittance inflows in 2010 can be explained by the regressors in the full specification of the regression model above. How much of this increase in explanatory power can be attributed to the inclusion of Rural Population in the model? The last specification in Table 4 above drops all of the control variables, keeping just Rural Population. The adjusted-r 2 of 0.7496 means that nearly all of the explanation of variations in remittances can be explained by Rural Population levels. While this discovery is quite astounding, the issue still remains about why Emigrant Stock was highly significant in all specifications, except when Rural Population was included. Some careful consideration of these variables reveals that they are likely highly correlated with one another; that is, the greater a nation s rural population, the greater the number of emigrants that will live and work abroad. The theoretical justification for this expectation is that rural regions tend to be home to the lowest income earners in the developing world. With little hope, many of them are likely to seek work outside of their nation s borders, thus boosting the emigrant stock. Can we provide statistical evidence to support this hypothesis? Table 5 below presents a correlation matrix of all the variables in the model. Table 5: Correlation Matrix of Variables in OLS Regression Explaining Remittances Rural Population 0.00007 0.00007 Remittances Emigrant Stock GDP Years Independent Unemployment Rate Public Health Rural Population Intercept 511.69 (0.0314) 405.11 (0.515) 181.59 (0.780) 1480.91 (0.258) 589.00 (0.468) 1338.56 Remittances 1 Emigrant 0.60 1 N 173 165 159 144 144 172 Stock GDP -0.01-0.06 1 Adjusted-R 2 0.2696 0.2652 0.2772 0.3590 0.7544 0.7496 Years Independent Unemployment Rate 0.17 0.11 0.35 1-0.12-0.08-0.09-0.09 1 Of course, the main problem with a bivariate regression model is that it does not control for additional factors influencing remittances in 2010. Models 2 through 4 in the table above fix this issue, by successively including control variables into the model. It is clear that the results of the bivariate regression are robust across these new specifications. For example, it is consistently estimated that the average marginal impact of the Emigrant Stock on remittance inflows is equal to about 0.0021 after controlling for per capita GDP, the years since independence, the unemployment rate and public health expenditures. Although the control variables are not statistically significant, the regressor Years Independent is statistically different from zero at the 10% significance level in the third model above. However, its estimated coefficient is positive, which contradicts theoretical expectations. In fact, the sign and approximate magnitude of this variable is consistent across all estimations in the table above. Clearly, further research or novel specifications are necessary to settle this issue. Another issue that must be addressed is the lack of economic and statistical significance of the Emigrant Stock regressor in the sixth specification. Not only is the sign for Emigrant Stock reversed (thus defying theoretical expectations), its p-value of 0.790 means that we fail to reject the null hypothesis that in favor of the alternative hypothesis that. However, the new included variable, Rural Population, is statistically significant and has a theoretically consistent sign. Its estimated coefficient of 0.00007 means that every extra person living in a rural region in 2010 will increase remittances in that year by an average of $70, holding constant all other Public Health -0.08-0.07 0.69 0.30 0.17 1 Rural Population 0.87 0.69-0.10 0.08-0.14-0.19 1 A cursory analysis of this table confirms the above conjecture. Pearson s correlation coefficient r (which ranges from 1 to 1) between Rural Population and Emigrant Stock is equal to 0.69. Therefore, these two variables are highly correlated with one another, which explains why the significance of Emigrant Stock was lost when Rural Population was included in the OLS model. Moreover, based on the first column of the matrix above, it is clear that Rural Population is much more strongly correlated with Remittances than Emigrant Stock is (r = 0.87 compared to r = 0.60). This explains why Rural Population remained significant. Although these two variables are measuring a similar economic variable, the average marginal impact of Emigrant Stock on Remittances (0.0021) is far greater than that for Rural Population (0.00007). This statistical discovery makes intuitive sense; if there are more people living in rural regions, then the potential for migration is much higher due to economic hardship. This indirectly increases remittances. However, if the potential increase in the Emigrant Stock are realized, the effect is far greater. 14
The Hinckley Journal of Politics 2015 Empirical Analysis: Explaining Growth These findings conclude the results for the first regression model estimated in this paper. The second model to be estimated seeks to explain growth in national output as a function of remittances. A key question to consider before estimating such a model is which variables must be controlled for to get a valid estimate of the average marginal impact of remittances on economic growth. Research conducted by Robert Barro (1996) of the National Bureau of Economic Research provides some useful guidelines. According to Barro, who used a panel dataset with 100 countries from 1960 to 1990, the key variables that must be included in any regression explaining GDP growth are: initial schooling, life expectancy, fertility, government consumption, rule of law, inflation, trade levels, and political freedom (Barro, 1996). Due to the limitations of the World Bank dataset used in this paper, only some of these regressors are included as control variables in the OLS model explaining output growth as a function of remittances. The following equation specifies the regression model, followed by an explanation of each variable as well as the hypothesized coefficient signs. Y=β 0 +β 1 rem+β 2 un+β 3 infl+β 4 gini+β 5 lit+β 6 life+β 7 educ+β 8 hlth+ε 15 1. GDP Growth: This continuous variable is the dependent variable in the OLS regression model to be estimated. It is simply defined as the rate of growth in gross domestic product (GDP) in 2010. (Note that all explanatory variables are in 2009.) 2. Remittances (+): This continuous explanatory variable is officially defined as the amount of migrant remittance inflows into a particular country in millions of U.S. dollars during the year 2009. For reasons outlined throughout this paper, the expected coefficient of this variable is expected to be positive. That is, increases in remittances in 2009 should lead to increase in growth in the subsequent year, after controlling for all other regressors of GDP growth in this model. 3. Unemployment Rate ( ): This continuous independent variable measures the national unemployment rate in a particular country in 2009. This regressor is included as a control variable; it is hypothesized to have a negative marginal impact on the rate of economic growth. Standard economic theory dictates that increases in the national unemployment mean that fewer national resources are being used toward production, so increases in unemployment should lead to reductions in growth. 4. Inflation Rate ( ): This continuous explanatory variable measures the national inflation rate in terms of the Consumer Price Index (CPI) in a particular country in 2009. Increases in the inflation rate result in reductions in economic growth, since consumer spending is discouraged by the higher prices, resulting in reduced aggregate demand and lower output. 5. Gini Coefficient ( ): This continuous independent variable operationalizes a nation s level of inequality in the year 2009 by the Gini coefficient. This statistic is calculated from the Lorenz curve; it is simply the ratio of the area between the line perfect equality and the Lorenz curve to the total area beneath the line of perfect equality (see Todaro [2012] for a detailed explanation of this statistic). The hypothesized sign is negative since greater inequality has been shown to reduce future long-run growth. 6. Adult Literacy Rate (+): This continuous explanatory variable measures the adult literacy rate in a particular country in 2009. A country with a skilled labor force tends to perform better than one with a less skilled labor force (all else equal), thus an increase in literacy should lead to increases in growth. 7. Life Expectancy (+): This continuous independent variable measures the life expectancy in years in a particular country in 2009. Its positive sign indicates that a healthier population, which lives longer, is expected to contribute to a nation s economy. Therefore, increases in life expectancy should be associated with increases in the growth of GDP, ceteris paribus. 8. Education (+): This continuous explanatory variable tracks the proportion of school-aged children enrolled in primary school in 2009. The reason for its hypothesized positive sign is similar to the theoretical argument offered for Adult Literacy Rate. 9. Per Capita Health Expenditures (+): The last continuous independent variable is the per capita health expenditures in a particular country in 2009. The reason for its hypothesized positive sign is similar to the theoretical argument given above for the explanatory variable Life Expectancy. Table 6 below presents the number of observations, mean value and standard deviation for each of the nine variables described above. For example, it is clear that the average GDP growth rate in the world in 2010 was roughly 4% with a standard deviation of 4%. A similar analysis can be conducted with the other variables. Of particular interest are the Gini Coefficient and Literacy Rate variables, which only have 44 and 30 observations respectively. In a model with eight explanatory variables (and a constant term) to be estimated, this small number of observations could be problematic in terms of the small number of degrees of freedom. As shall be explained in the following pages, this will be a key consideration in specifying the control variables. Table 6: Descriptive Statistics for Relevant Variables Variable Observations (N) Mean (μ) Standard Deviation (σ) GDP Growth 190 4 4 Remittances 173 2,428 5,849 Unemployment 174 9 6 Inflation Rate 180 4 5 Gini Coefficient 44 41 10 Adult Literacy 30 85 16 Life Expectancy 202 70 10 Education 132 91 12 Health Expenditure 187 1,065 1,410 The next step is to estimate the specified model using the Ordinary Least Squares method. Three different specifications of the general model are presented in Table 7 below. As with the model produced earlier in this paper, the first specification is a simple bivariate regression with Remittances explaining GDP Growth Rate. It is evident from the table below that the estimated coefficient is positive (confirming the initial hypothesis) and statistically different from zero at the.1 significance level. The calculated coefficient of 0.00009 suggests that every additional million dollars of 15
An Empirical Study of Remittances and Growth in the Developing World Elird Haxhiu remittances sent to a particular country will increase that country s growth rate by average of 0.00009 percentage points. Similarly, a $100 million increase is expected to increase growth by 0.009 percentage points. Although the marginal impact is clearly small, it is positive and statistically significant. Table 7: OLS Regression Models Explaining GDP Growth Rate Regressor (1) (2) (3) Remittances 0.00009 (0.079) -0.0018 (0.075) Unemployment 0.0042 (0.983) Inflation Rate -0.0284 (0.929) Gini Coefficient 0.0579 (0.575) Adult Literacy -0.1659 (0.131) 0.00009 (0.045) -0.0276 (0.602) -0.0663 (0.295) Before addressing these two problems in the third regression, it is important to verify whether multicollinearity is present. After all, the two key characteristics of high multicollinearity are insignificant estimates and unexpected signs (which could explain the negative coefficient for Remittances). Table 8 presents the Variance Inflation Factor (VIF) scores for the model. Eight separate regressions are run for each explanatory variable as a function of the other seven independent variables. In each case, a VIF score is calculated by taking the inverse of one minus the explanatory power of the model. For example, for the variable the Variance Inflation Factor score is. The reason that VIF scores are utilized to diagnose multicollinearity, rather than a simple correlation matrix as in the previous section, is that the cause of the problem is unknown. In the model explaining remittances, it was theoretically clear that the emigrant stock might be highly correlated to the number of citizens living in rural areas. In this model, it is not as theoretically evident which variables may be causing the multicollinearity, thus VIF scores are a more general diagnostic tool that examines the effect of every regressor on the others. As A.H. Studenmund (2012) recommends in the textbook Using Econometrics, any VIF score greater than five indicates severe problems of multicollinearity. Table 8 suggests that the second specification suffers from this problem, as most of the regressors have VIF scores of over five. Table 8: Variance Inflation Factor Scores for Multivariable Regression Models Life Expectancy 0.3351 (0.187) Education -9213 (0.025) -0.0268 (0.579) -0.0571 (0.113) Regressor Model (2) Model (3) Remittances 3.12 1.04 Unemployment 5.83 1.07 Health Expenditure 0.0096 (0.101) Intercept 4.02 77.81 (0.047) -0.0007 (0.002) 12.64 Inflation Rate 5.07 1.24 Gini Coefficient 4.38 Adult Literacy 5.28 N 167 11 106 Life Expectancy 14.56 2.70 Adjusted-R 2 0.0126 0.08978.2195 Education 2.02 1.87 Health Expenditure 8.18 1.68 Of course, interpreting a simple bivariate regression model is problematic due to the lack of statistical control for other factors influencing the growth rate. This is accomplished in the second specification of the model in Table 7 above, where all seven control variables are included in the regression model. It is evident that the results of the bivariate regression are not robust. The estimated coefficient for Remittances, although statistically different from zero at the.10 significance level, is now negative. However, all of the control variables are not statistically significant (except education, which itself has an unexpected sign) while the explanatory power of the model has increased dramatically from an adjusted-r 2 of just over 1% in the bivariate case to almost 90% in the multivariate specification. These two facts lack of significance but high explanatory power suggest that the second model suffers from high multicollinearity, a problem which must be addressed. Moreover, due to the small number of observations for the Gini Coefficient and Literacy Rate variables, only 11 observations were used to estimate the second model above. Mean VIF Score 6.06 1.60 This problem is remedied in the third specification where the two variables with the small number of observations (Gini Coefficient and Adult Literacy) are dropped from the model. The new model is now estimated with a total of 106 observations and exhibits a much more reasonable adjusted-r 2 of 0.2195. That the problems associated with multicollinearity have been addressed can readily be verified from Table 8 above, where all of the VIF scores for model (3) are below five. As a result, it is clear that the third specification is preferred to the second in terms of econometric theory. While the control variables are still rather insignificant and exhibit unexpected signs (for example, Education has a negative coefficient), the new estimate for the impact of Remittances on Growth is promising. 16
As illustrated in Table 7, the estimate for Remittances has a p-value of 0.045. This means that we successfully reject the null hypothesis that β1 = 0 in favor of the two-tailed alternative hypothesis that β2 0 in the population. Therefore, the estimate for the marginal impact is statistically significant at the standard 0.05 significance level. Moreover, the estimated sign and magnitude of the coefficient is identical to the one found in the bivariate case. Thus, it is determined that, even after the relevant (and available) control variables have been included in the regression model and held constant, the true effect of remittances on growth is positive. In other words, increases in remittances lead to increases in economic growth. As other researchers have discovered, however, this marginal impact is quite small. Conclusion The findings of this paper support the debate summarized in the literature review of this paper about the positive, but average marginal impacts on economic growth by remittances, holding all else the same. However, many of those papers focused on single countries and a limited range of data. The models estimated in this paper relied on a large dataset of 214 different countries. Though such an aggregated setting was unlikely to yield promising statistical results, two novel findings have emerged in this research. The first is that the key determining factor of remittance levels is a country s emigrant stock (or alternatively, the number of citizens living in rural areas). The second is that, after controlling for the effects of other relevant variables, remittances positively affect economic growth, though the magnitude of that effect is small. What do these results mean for developing countries, since most remittances flow to these states? To begin with, the first regression model estimated by this paper suggests that there is little countries can do to change the level of remittances they receive. Aside from encouraging mass migration away from their nations (as the Philippines does, but hardly a feasible policy choice for others), developing countries must accept the level of remittances they receive. However, since the second regression model has demonstrated the positive impact of remittances on growth, developing nations can (and should) take steps to improving the manner in which those remittances are used to enhance their affect. As Catrinescu et al. (2006) have noted, institutions play a key role in the use of remittance funds. Since these remittance flows come at the steep price of out-migration, which likely results in a negative impact on growth, policy must be concentrated to make the best use of these funds possible, directing them toward productive investments. Finally, what do these results mean for future studies on this topic? First, it must be noted that this paper has hardly settled the score on this important issue in development economics. Though the cross-sectional model estimated included a broad range of countries, it was only conducted for the year 2010 due to data limitations. These results must be replicated during other time periods. Moreover, the limitations of the dataset mean that some important determinants of growth were not included as control variables. Future research, which aims to replicate these results for a broad range of countries, must focus on gathering as much data as possible, following the recommendations of Barro (1996). In addition, it is clear that more theoretical work must be done to investigate what types of institutions are conducive to the productive use of remittances. As numerous authors have previously noted, whether or not a nation benefits from remittances depends on the types of institutions in that country. However, some contradictory conclusions have been drawn. While Catrinescu et al. (2006) have claimed that better financial institutions contribute the positive relationship between remittances and growth, Fayissa and Nsiah (2008) claim that the opposite is true. They found that countries with weaker financial institutions benefitted more from remittances than countries with better financial institutions. This issue must be resolved or at the very least studied in further detail to explain the discrepancy. Finally, additional empirical work is needed to study this effect of institutions; potential institutional variables must be operationalized and used to further explain and refine the effects of remittances on growth. References Barajas, Adolfo, Ralph Chami, Connel Fullenkamp, Michael Gapen and Peter Montiel. Do Workers Remittances Promote Economic Growth? Working Paper. International Monetary Fund. July 2009. Barro, Robert. Determinants of Economic Growth: A Cross-Country Empirical Study. Working Paper. National Bureau of Economic Research. August 1996. Catrinescu, Natalia, Miguel Leon-Ledesma, Matloob Piracha and Bryce Quillin. Remittances, Institutions and Economic Growth. Working Paper. Institute for the Study of Labor (IZA). May 2006. Clemens, Michael and David McKenzie. Why Don t Remittances Appear to Affect Growth? Working Paper. World Bank Development Research Group. May 2014. Cypher, James and James Dietz. The Process of Economic Development, 3rd edition. Print. Routledge Publishing. 2009. Fayissa, Bichaka and Christian Nsiah. The Impact of Remittances on Economic Growth and Development in Africa. Journal of Economic Literature. February 2008. Orrenius, Pia, Madeline Zavondy, Jesus Canas and Roberto Coronado. Do Remittances Boost Economic Growth? Evidence from Mexican States. Working Paper. Federal Reserve Bank of Dallas. February 2009. Ratha, Dilip, Sanket Mohapatra and Ani Silwal. The Migration and Remittances Factbook 2011. Migration and Remittances Unit. World Bank. Ratha, Dilip. The Impact of Remittances on Economic Growth and Poverty Reduction. Policy Brief. Migration Policy Institute. September 2013. Remittance Data Inflows, Remittance Data Outflows, Estimates of Migrant Stocks and Raw Macroeconomic Indicators. World Bank Economic Databases. Accessed November 18th 2014. Source: http://data.worldbank.org. Siddique, Abu, E.A. Salvanathan and Seroja Salvanathan. Remittances and Growth: Empirical Evidence from Bangladesh, India and Sri Lanka. Discussion Paper. The University of Western Australia. August 2010. Studenmund, A.H. Using Econometrics, 6th edition. Print. Pearson Publishing. January 2012. Todaro, Michael and Stephen Smith. Economic Development, 11th edition. Print. Addison-Wesley Publishing. 2012. United Nations Development Program. Human Development Report 2013. United Nations. 17
18