Do Developing Economies Respond to Remittance Flows? Some Channels of the Macroeconomic Impact

Supervisor: Dr. Dirk J. Bezemer October 2008

Do Developing Economies Respond to Remittance Flows?

Some Channels of the Macroeconomic Impact

Clara van der Pol

Abstract

This paper investigates the effects of remittances on developing economies receiving a relatively large portion of their national income from remittances. It is hypothesized that in these countries, remittances may have an effect on the aggregate economic indicators of investment, consumption and imports. It is further hypothesized that if this is true, remittances may have more far reaching effects on national economic structure. Overall, the panel OLS and GMM analysis results show that remittances have effects on national investment, consumption, and imports. The systems of equations set up to test the effect of remittances running through these aggregates distill some viable channels by which remittances shape changes in the value added of different economic sectors. Overall, remittances are shown to have both direct and indirect effects on structural change.

JEL Classification Codes: C33, E25, F21, O11, O57

Keywords: remittances, panels, investment, consumption, imports, structural change

1. Introduction

Studying remittances in a macroeconomic context is important for a number of reasons. Remittances represent a large portion of foreign capital entering developing countries, in many cases amounting to double-digit ratios of GDP and much larger than other capital inflows. Considering these astronomical figures, the past decade has shown policy makers giving increased importance to remittances as a tool for development. Ironically, though, remittances are also the least researched capital flow to developing countries in terms of their effects on the receptor economies. Though early research has provided a pool of theoretical foundations regarding the reasons to remit, and a recent stream of household studies provides important insights into the spending patterns of remittances receivers, the focus on remittances as a macroeconomic capital flow is still nascent. Implicit in much economic thinking and policy regarding remittances is that the micro implications of remittances translate automatically into macro indicators. However, the existing research formally analyzing this relationship is meager at best.

macroeconomic spheres. Following Lindahl (1970), microeconomics is viewed as focusing on the actions of individuals and how they respond to incentives. Macroeconomics constitutes the tracing out of the unintended consequences of various actions and sets of individual actions that emerges when each of many individuals responds to the incentives identified and classified by microeconomics. While macro relates to groups of goods and production factors, a micro view focuses on economic behavior of individuals: their active choice guided by incentives. Remitting is inherently an individual or household decision, thus micro in nature. This can yield a couple of macro results, but also some additional micro effects, which in turn and ultimately give way to macro outcomes.

Micro household studies have uncovered the different purposes served by remittances. Ironically, though the effect of remittances on the household is largely deemed positive, the effect of remittances on national aggregates has often been cast in negative prospects and skepticism (i.e. Amuedo-Dorantes and Pozo, 2004). This possible disparity between micro- and macroeconomic consequences of remittances has largely been overlooked. If the micro decision patterns of how to use remittances are translated into macro indicators, they could have lasting effects on an economy’s macro profile. And these needn’t be positive effects, nor lasting, nor directly proportional. They needn’t even be traceable, as the proportion amounting to remittances in many countries may simply not be large enough to show up in aggregate figures.

To investigate this, the paper first looks at whether remittances have an effect on aggregates of household expenditure pointed out in the microeconomic literature. Theoretically, remittances create (1) increased demand for goods and services, and thus consumption, by increasing household disposable income; and (2) a possibly increased supply by facilitating investment through lowered household capital restrictions and risk (de Haas, 2006). At the same time, several authors have expressed concern in (3) reporting imports to be an important absorber of remittances. Remittances that go to imports mean that the capital injection is transferred abroad, likely damaging a country’s balance of payments and leaving domestic production unaltered. On the other hand, domestic sectors that deal with trade may flourish.

Therefore, this study looks at how remittances affect the relevant expenditure posts of consumption, investment, and imports. Once these effects are established, the ‘shock’ created by remittances can then be analyzed in its effect on the structural evolution of an economy. Different (un)productive sectors may react to cater to the increased demand, while these different sectors will allow for investment, and expansion or contraction, in differing degrees. Because of these mechanisms, an economy’s sectoral structure may change, due to remittances. In this study, what is conventionally understood to be ‘structural change’ is analyzed in an unconventional way. Structural change is broadly defined to reflect changes in the productive composition of an economy. It is measured by the evolution in the relative shares of agriculture, industry and services. Nevertheless, instead of explaining structural change from purely supply-side factors as has become standard in the literature (i.e. Temple and Wößmann, 2006, and Timmer and de Vries, 2008), this study encompasses also demand-side determinants of structural change. This is important because, as mentioned previously, remittances create both demand and supply effects. While the effect of remittances on national investment, consumption, and imports has received some attention (i.e. Glytsos 2002), the relationship between remittances and the structural composition of a receptor country is something that, to the knowledge of the author, has not been attempted yet. Formally, the paper investigates three main research questions:

1. What is the effect of remittances on aggregate consumption, investment and import? 2. What is the effect of remittances on the structural composition of economies?

In answering these questions, the paper may contribute to partly answering how the study of remittances differs from studying the impact of any other source of additional income (López-Cordoba and Olmedo, 2006) or foreign capital inflows. The paper is organized as follows. The next section (2) provides a review of the existing theory and evidence regarding remittances in order to create a theoretical platform for analyzing remittances at the macroeconomic level. This is followed by Section 3, which explains the methodology, and Section 4, which outlines the data sources. The results are presented in Section 5, and the paper concludes in Section 6 with a discussion.

2. Literature Review

After exposing the existing theory and evidence on remittances, this section paves the way towards inserting remittances in a macroeconomic framework by theoretically exploring their relation to expenditure aggregates and structural change.

2.1 Remittances: theory and economic relevance

The current remittance literature can be divided into two streams; one dealing with the micro aspects of remittances, and a second one dealing with the macro aspects of remittances. The former is the richer literature field while the latter has been picking up over the past decade. Much of the literature focuses on unveiling the mostly micro, but more recently also macro, determinants of remittances. The research area dealing with the consequences of remittances has seen its greatest contributions from the micro field, where regional household surveys or national data yield valuable insights into the uses of remittance money.

As stated in de Haas (2006), there has been a heated debate over the past decades of the ‘migration optimists’, arguing that migration brings about transfers of capital and accelerates the acquisition of ideas, experience, skills and knowledge in the developing world; versus the ‘migration pessimists’ who highlight growing inequality, the withdrawal of human capital, remittance dependence, and the use of remittances for “unproductive” and conspicuous consumption. The optimist ‘developmentalist’ view is now in vogue and expects that migration and remittances will provide poor countries with the extra push they need for developmental take-off. The pessimist view relies on ‘structuralist’ theory to show how migration is a byproduct of underdevelopment and can even be detrimental for an economy. In this sense, ‘the failure of remittances to provide a positive stimulus to development in a number of labor-exporting countries is largely due to the structural features of underdevelopment’ (Stahl and Habib, 1991) that can’t be overcome merely with remittances.

2.1(i) The micro economy

As voiced by Puri and Ritzema (1999), the expenditure pattern of migrant households is central to any meaningful discussion on the development implications and policy measures to enhance the developmental impact of remittances. Much microeconomic theory and testing has revealed that remittances mostly go into: consumption (Orozco, 2006), human capital investment like education and health care (Cox Edwards and Ureta, 2003; Adams, 2007), capital investment like motor vehicles and agriculture tools (Glytsos, 1993), real estate and home investments (Adams, 1991), and entrepreneurial activities (Adams, 2007 and Woodruff and Zenteno, 2001).

Though it has been found that a large share of remittances is used for consumption purposes, several household studies have found that remittance receiving households have a higher propensity to save and invest and a lower consumption pattern compared to households that do not receive remittances (Adams, 1991, 2007, and Taylor and Mora, 2006). Thus, through the provision of liquidity and income security, remittances can create ‘income multipliers’ within the household. Woodruff and Zenteno (2001) describe the surge in Mexican microenterprises financed with remittances. Rosen (2007) observes that for most rural Pakistani households, investment options financed by saved remittances are limited to low-risk assets with quick financial gains that can be purchased locally: land and housing, agricultural capital, human capital, and startups of non-agricultural businesses. Ballard (2003) has argued that, if invested, remittance tend to go to real estate, but also to services and other ‘unproductive’ activities like hiring workers and ‘building a mosque’ (Lipton, 1980).It is clear that the lack of comparative perspective in much of the existing literature risks missing the broader picture.

2.1(ii) The macro economy

Taylor (1999) exclaims that, for an economist, the issue of how remittances are actually spent is ‘the wrong question to ask’. This, he argues, is because remittances merely increase the household budget for goods. Though this is true, it is also true that remittances have certain microeconomic attributes that make them unique from other sources of income1. Because of this, remittances may also have unique effects on the macro economy, just as other capital flows like foreign development aid have unique effects. The unique macroeconomic attributes of remittances are comparable to the microeconomic attributes, and it is what makes remittances distinctly different from other capital flows like foreign direct investment, official development aid and capital market flows. Remittances have been shown to be more stable over time and countercyclical (i.e. IMF 2005). Also, they largely circumvent problems of bureaucracy, corruption and distribution that plague the other types of capital flows because they go directly to the people for whom most development policy is intended. Just like the NELM argues that remittances ease capital constraints and lower risk for the household, remittances may help overcome balance of payments and (foreign) capital market restrictions in a macro perspective.

Implicitly assuming remittances as a unique capital flow, many researchers have uncovered some of their developmental opportunities or threats. Notable examples of this recent literature include studies on the effect of remittances on poverty and inequality (Acosta, Calderón, Fajnzylber and Lopez, 2008), on the labor supply (Kim, 2007), on growth (Faini, 2002; Chami, Fullenkamp and Jahjah, 2005, Giuliano and Ruiz-Arranz, 2005, and Catrinescu, León-Ledesma, Piracha and

1 _{See for example Stark and Bloom (1985) on altruism and family members as agents in protecting the}

Quillin, 2006), and on financial development (Aggarwal, Demirgüc-Kunt and Martinez Peria, 2006).

Glytsos (2002) argues that the ‘productive use’ of remittances may be served in a variety of ways through the management of remittances (e.g. by banks), extension of investment credit allowed by the increase in the liquidity of banks from remittance deposits, liberalization of other resources from consumption; investment in human capital (through education and healthcare), purchase of more investment goods from abroad, and growth of investment as a result of the multiplier effect of spending on consumption. In addition, market linkages transmit the impact of remittances from the households receiving them to others in the local, regional, or national economy (Taylor, 1999). If one takes the example of Albania, the vast inflow of remittances made it possible for the value of imports to greatly exceed that of exports. At the same time, remittances facilitated a large number of small-scale direct investments in the Albanian clothing and footwear (export) industry (Clunies-Ross and Sudar, 1998).

All this implies that increased remittances may bring about changes in the nongovernmental components of expenditure (GDP) posts: consumption, investment and net imports2. Some important early developments have been made in this field. In his doctoral thesis, Chaney (1984) applied a microeconomic foundation of remittance behavior to understand the macroeconomic impact of remittances via savings in Portugal. Glytsos (1993) finds that for Greece, remittances indeed induced higher investment and multiplier effects through increased consumption. In a study on seven Mediterranean countries using dynamic macroeconomic models, Glytsos (2002) finds that the effect of remittances on consumption, investment, imports and national output varies over time and across countries, pointing to different inter-country priorities of remittances spending. Additionally, he finds an asymmetric impact of remittance changes, with rising remittances positively affecting growth to a lesser extent than the negative impact from falling remittances. In their study on determinants of remittances, El-Sakka and McNabb (1999) found that imports financed through remittance earnings have a very high income elasticity.

Remittances could be susceptible to macroeconomic glitches, to the extent that conclusions drawn from household studies do not hold at a national or cross-national level, boiling down to a fallacy of composition as exposed by Keynes (what is in the interest of an individual may not work in the interest of the [trade as a] whole). The representative agent framework provides macroeconomists with powerful microeconomic tools, but “it has also blurred the distinction between statements that are valid at the individual level from those that apply to the aggregate” (Caballero, 1992). Reasons for why the largely positive micro observations of the effects of remittances do not necessarily hold in macro terms are plenty. The relationship between aggregate supply/demand and domestic output3 may be affected by institutional restrictions on economic activities and market imperfections that obstruct the formation of market linkages; the counter-cyclicality of remittances; interactions between households, even those not directly receiving remittances, so that the group behaves differently than the individual household; or the (disproportional) distribution of remittances in favor of the richer families who do not face capital constraints (Lipton, 1980). Externalities may also play a role: the Dutch disease where a resulting higher exchange rate evaporates the incomes from remittances (i.e. Bourdet and Falck, 2006) or the crowding-out of other sources of (foreign) capital for investment like financial markets.

2

Although several governments do try to proactively manage remittances, such expenditures are not a direct effect of remittances flows and will thus not be analyzed in this paper. Imports are considered instead of net exports so as to disregard the more difficult to conceptualize and possibly less likely effects of exports.

3

2.2. Remittances and Structural Change

Structural change is a term used to define the reallocation of factors across different sectors of the economy over time (i.e. Kuznets, 1971, Chenery et al., 1986, and Blomqvist, 2001). Factors of production are transferred from the sector with the lowest productivity, agriculture, to the industrial sector, characterized by a higher level and growth of productivity. A related concept is Engel’s law which describes how poor countries spend a large portion of their income on agricultural products. As per capita income increases, the demand for agricultural products declines while the demand for industrial products increases. This creates a market for industrial goods that can be domestically catered to, or, if a domestic industrial sector doesn’t develop, these goods will have to be imported. Harrod and Hague (1963) stressed that imports are vital to fill the gap between growing domestic aggregate demand and limited supply during moments of increased growth.

Developing countries are expected to follow a pattern of industrialization as they develop, but this must not necessarily be the case. Savona and Lorentz (2006) found that the largest contribution to structural change favoring services stems from ‘shifts in private domestic consumption, which are in turn … mainly sustained by a positive income effect’. If remittances are large enough to create this shift in domestic demand, then they may stimulate the services sector more than the industrial sector, altering the classical path of structural change. If in addition the increased demand, whether for consumption or investment, is met largely through increased imports then remittances have negative consequences for the Balance of Payments, positive effects on trade industries, and likely negative effects on domestic production. Alternatively, increased demand may also create many inter-industry linkages from higher consumption and remittances may facilitate domestic supply, enhancing the national economy. An economy’s structure is likely to change if remittances are repeatedly channeled into or out of particular industries or sectors. For example, in Albania remittances have mostly gone to small family business, as well as trading and agricultural activities (Nicholson, 2001), favoring the value added of agriculture and services over industry.

3. Methodology

The analysis comprises a system of equations (3.1) relating remittances to measures of aggregate consumption, investment and imports; (3.2) relating remittances to the measures of structural change; and (3.3) combining these two by making remittances run through expenditure on consumption, investment and imports, to structural change. Different measures were used for the same constructs in order to first, assess robustness and second, compare results of popular measures (used in previous studies) that may be the origin of conflicting results and conclusions. Thus: investments, consumption and imports are each measured once as a percentage of output (GDP) and once as a monetary value; remittances are measured once as a percent of GDP and once per capita.

3.1. Which components of GDP expenditure, if any, are affected by remittances? The answer to this question lies in twelve different regressions, four pertaining to each aggregate expenditure post.

Investments

encouraging investments that are not possible otherwise (de Haas, 2006). If one or both of these hypotheses are correct, remittances should show a positive effect on aggregate investment. Some studies attempting to explain aggregate investment are based on the Euler model of a representative firm applied to aggregate data but Gordon (2002) emphasizes that this leads to bad estimates: “if an analyst is interested in estimating the parameters of a representative firm, this is best done using disaggregated data”. Therefore, this study attempts to explain aggregate investment from other economic aggregates. The following Equations portray this:

I(u)i,t = F(Rem1i,t-1, Xi,t) I(u)i,t = F(Rem2i,t-1, Xi,t) (1a-b) I(v)i,t = F(Rem1i,t-1, Xi,t) I(v)i,t = F(Rem2i,t-1, Xi,t) (1c-d)

i=1,2,...,n ; t=1,2,...,T

where I(u)i,t is the ratio of investment (gross capital formation) to GDP in country i in period t, I(v)i,t is the monetary value of investment expenditure (in constant 2000 US$), Rem1i,t-1 is remittances over GDP, Rem2,t-1 is remittances per capita, and Xi,t is a matrix of regressors (control variables) representing factors that influence imports.

The literature revealed the following control variables. A measure of economic activity is crucial: Glytsos (2002) uses GDP while Drinkwater et al. (2006) proxy it by the one-period lag on the growth of real GDP. The lending rate (Drinkwater et al., 2006), overall tightness of the credit markets (i.e. Servén, 2002), capital account openness (by facilitating the efficient allocation of investments across boundaries to reduce the cost of capital (Henry, 2007) but possibly also creating volatility) and institutions (Dawson, 2007) have been found to also play a role in the investment decision.

Thus the matrix of control variables includes: logGDP1 (the one period lag of the growth rate in GDP to proxy for economic activity), RR (real domestic interest rate to proxy for the lending rate),

DOMCR (domestic credit to the private sector as a percent of GDP, to proxy for the overall

tightness of credit markets), CHINITO (the Chinn and Ito (2006) measure for capital account openness), and PR/CL (quality of institutions proxied by the Freedom House score for Public Rights/Civil Liberties).

Consumption

Do remittances increase households’ expenditure on consumption? The microeconomic literature offers a resounding affirmative answer to this question. The following functions reveal the possible relationship in macro terms, where a positive coefficient for remittances needs to be found in order to agree with the micro findings:

C(u)i,t = F(Rem1i,t-1, Yi,t) C(u)i,t = F(Rem2i,t-1, Yi,t) (2a-b) C(v)i,t = F(Rem1i,t-1, Yi,t) C(v)i,t = F(Rem2i,t-1, Yi,t) (2c-d)

i=1,2,...,n ; t=1,2,...,T

where C(u)i,t is the ratio of household expenditure over GDP in country i in period t, C(v)i,t is the monetary value of aggregate household expenditure (in constant 2000 US$), Rem1i,t-1 is remittances over GDP, Rem2i,t-1 is remittances per capita, and Yi,t is a matrix of control variables.

consumption goods goes up, the demand for them will decrease, reducing aggregate household consumption expenditure. A measure for consumer prices, following Holtham and Kato (1986), should thus be included. Lending rates must also be included to account for the saving trade-off in consumption (Hamburger, 1954; Weber, 1970; Holtham and Kato, 1986). Hamburger (1954) also stresses the importance of future income uncertainty. In addition, Borooah and Sharpe (1986) argue for consideration of the fact that the household is not homogenous across units. The composition of aggregate households changes over time with regard to the proportions of households at different stages of the life cycle and with regard to the proportions in different income classes.

The control variables are the following. Data allows the net savings measure to adjust only for capital depreciation (SAV), but a wealth dilution term is included (POPR, the ratio of population growth rate to the total population as done in Ferreira et al., 2008). Also included are: INFL (inflation), RR (real interest rates), UNEMPL (unemployment to proxy for future income uncertainty), POP15 (population between the age of 15 and 65 to proxy for the life-cycle hypothesis), and CPI (Consumer Price Index).

Imports

Do remittances increase the expenditure on imports? A positive relation would reflect a domestic preference for foreign goods that can be afforded with increased income and/or the inability of domestic suppliers to cater to the increased demand. The following Equations are created:

M(u)i,t = F(Rem1i,t-1, Zi,t) M(u)i,t = F(Rem2i,t-1, Zi,t) (3a-b) M(v)i,t = F(Rem1i,t-1, Zi,t) M(v)i,t = F(Rem2i,t-1, Zi,t) (3c-d)

i=1,2,...,n ; t=1,2,...,T

where M(u)i,t is the ratio of imports to GDP in country i in period t, M(v)i,t is the monetary value of aggregate expenditure on imports (in constant 2000 US$), Rem1i,t-1 is remittances over GDP, Rem2i,t-1 is remittances per capita, and Zi,t is a matrix of control variables that influence imports.

The literature was screened for important control variables in the aggregate import function. Dutta and Ahmed (1999) summarize the most important import determinants as: import price, income (GDP), and foreign exchange reserves (which for many developing countries are made up of mostly export earnings, foreign direct investment, foreign aid and remittances). For developing countries, foreign debt is an important means to finance imports. Additionally, in countries with little access to foreign capital, remittances might have a higher effect.

The control variables are described as follows: IVID denotes the relative prices of imports (measured as the Import Value Index deflated by domestic prices, CPI, as in Dutta and Ahmed (1999)), OEXR is the official exchange rate (to proxy for the real effective exchange rate as it has more observations), logGDP1 is the one-year lagged real Gross Domestic Product (to proxy for real domestic activity as in Dutta and Ahmend (1999) and Tang (2003), EXPO is real exports over GDP, FDI is foreign direct investment over GDP, AID is foreign aid received as a percentage of Gross National Income, and CHINITO is the Chinn and Ito (2006) index to control for capital account openness. A measure for foreign debt is not included because of scarce data.

3.2. Do remittances affect structural Change?

2 / 1 1 2 1 , 2 / 1 1 2 , 1 1 , ,

cos

























=

∑

∑

∑

s

s

s

s

θ

Yi,t = S1,i,t + S2,i.t + S3,i,t

where Yi,t is the total output in country i in period t and S1,i,t ,…, S3,i,t refer to the value added of, respectively, the agriculture, industry and service sectors. Each of these value added (S1,i,t ,…, S3,i,t) measures serve as one component of the structural change. They will each be incorporated in their own structural change equation so that the direction of structural change can be observed; in other words: do the explanatory variables move the economy towards agriculture, industry or services?

In addition, a measure of ‘overall’ structural change is used. Vikström’s (2001) θ captures the variation in the value added of the three main economic sectors—agriculture, industry and services—as summarized into a single number. It reflects the net change in the composition of an economy, but says nothing about the direction of the change. The measure is based on the Similarity Index as used in, for example Timmer (2000), but modified to express a geometrical degree:

where si,tare sector i’s value added at time t, and n is the number of sectors.The intuition

behind it is that the structure of output can be described as a vector of which the coordinates are the quantities of output. The angle between two vectors measured at different points in time becomes a measure of structural change. This is illustrated for the case of sectors in Figure 1.

Figure 1. Illustration of Vikström’s θ of Structural Change Source: Vikström (1991)

If vector 0A describes the composition or structure at the starting point and vector 0B describes the structure at the ending point, then angle θis a measure of the extent of structural change between these two points in time. The case of two sectors can easily be extended to an arbitrarily number of sectors, using a vector in n-space. For this study, Vikström’s θ over the three sectors is calculated from one year to the next, so annually. All in all, the study analyses four different aspects, or measures, of structural change.

must bear in mind that given such a construction one sector’s increase is another sector’s decrease. Vikström’s θ is useful to get an idea of the total ‘absolute value’ change across all sectors at once. It depicts how much shifting the economy undergoes from one year to the next, though it says nothing about how much change is attributed to each sector. Vikström’s θ and the values added nicely complement each other to get a balanced view of structural change. This requires eight separate equations:

SC(r)i,t = F(Rem1i,t-1, Ai,t) SC(r)i,t = F(Rem2i,t-1, Ai,t) (4a-h)

i=1,2,...,n ; t=1,2,...,T; r=1,2,…,7

where SC(r)i,t denotes structural change measure r 4

in country i at period t, Rem1i,t-1 is remittances over GDP, Rem2i,t-1 is remittances per capita, and Ai,t is a matrix of regressors that represent factors that are expected to influence structural change.

Much research has focused on the supply-side of structural change by looking at how differences in relative productivity amongst sectors attract more or less labor, in this way contributing more or less to total output (e.g. Timmer, 2000)5. Since remittances may create both a demand and supply shock, it is important to view structural change as having also demand-side determinants. The control variables used in the regressions are taken from the economic growth literature. This is done for pragmatic reasons: (1) there exists scant literature on the demand-side determinants of structural change and (2) data for supply side determinants of structural change as stated in the existing literature (productivity, technological change, etc.) is largely unavailable for the countries in question. Nevertheless, because both structural change and economic growth are meant to reflect economic development and because theoretically they move together6, the control variables for growth can proxy for the control variables of structural change. Indeed, Figure 2 shows that for the dataset used, the ratio of agriculture to GDP has a negative relationship to per capita income.

Figure 2: The Negative Relationship between Income and Share of Agriculture, 1970-2006 Source: dataset used in this study

0 1 2 3 4 5 6.4 6.8 7.2 7.6 8.0 8.4 8.8 9.2 9.6 10.0 Income (Log of GDP per capita, PPP corrected)

A g ri c u lt u re s h a re ( L o g o f % o f G D P ) 4

These structural change measures include (a) Vikström’s θ, (b) % agriculture value added, (c) % industry value added, and (d) % services value added.

5 _{But productivity needn’t be the only reason for shifts in economic structure. One could argue that an}

increase of labor in the service industry is a sign of lack of opportunities in more productive sectors (industry); this normally translates in a lot of informal labor.

The literature has put forward a number of recurring determinants for growth. The initial level of output per capita serves to account for conditional (β-) convergence forces (Barro and Sala-i-Martin, 2003). Secondary enrollment serves as a proxy for the country’s education level, and thus indirectly, for the level of human capital. Another aspect of human capital is health, often included as life expectancy at birth. Other determinants are exports (Krueger and Lindahl 2001) or terms of trade (Barro and Sala-i-Martin, 2003), and Foreign Direct Investment (FDI) (Borensztein, De Gregorio and Lee (1998). The growth literature often also recommends the inclusion of an Africa dummy and other geographical area dummies as additional control variables (i.e. Barro and Lee, 1993; Sala-i-Martin, Doppelhofer and Miller, 2004; Collier and Gunning, 1999). However, such dummies conflict with a possible fixed effects model specification which already accounts for differences between the countries, and can not be incorporated into a possible first-differences model because they are time invariant. Inflation is also deemed important, just like real exchange rate distortions (Sala-i-Martin et al., 2004), as they might exert negative effects on growth. Straub (2008) reviews the importance of infrastructure for growth in developing countries. Democracy and institutional effectiveness is also considered a factor in enhancing growth (Barro and Sala-i-Martin, 2003).

Thus, the control variables for structural change are as follows: income (the log of one-year lagged GDP per capita, PPP corrected) as in Barro and Sala-i-Martin (2003). SECENR is the secondary enrollment rate to proxy for education and health is proxied by the life expectancy (LIFE). EXPO is real exports over GDP and is preferred over the terms of trade because it has more observations. FDI is foreign direct investment as a percentage of GDP, INFL is the percent inflation, and OEXR is the official real exchange rate to proxy for real exchange rate distortions (as it contains more observations). Infrastructure is proxied by the per capita number of telephone mainlines (TELMAIN). The Freedom House Civil Liberties (CL) and Public Rights (PR) are used to proxy for institutional quality.

3.3. Does the effect of remittances on structural change run through investment, consumption or imports?

The analysis proceeds with a system of equations to empirically investigate the relationship from remittances to structural change as flowing through investment, consumption or imports. This is portrayed in the following systems of Equations:

I(u/v)i,t = G(Rem1i,t-1, Xi,t) I(u/v)i,t = G(Rem2i,t-1, Xi,t) (5.1a-b)

SC(r)i,t = F(I(u/v)i,t, Ai,t) (5.2)

C(u/v)i,t = G(Rem1i,t-1, Yi,t) C(u/v)i,t = G(Rem2i,t-1, Yi,t) (6.1a-b)

SC(r)i,t = F(C(u/v)i,t, Ai,t) (6.2)

M(u/v)i,t = G(Rem1i,t-1, Zi,t) M(u/v)i,t = G(Rem2i,t-1, Zi,t) (7.1a-b)

SC(r)i,t = F(M(u/v)i,t, Ai,t) (7.2)

i=1,2,...,n ; t=1,2,...,T; r=1,2,…,7

from such a specification represent the partial derivatives of the predetermined variable (change in investment, consumption, imports) with respect to the change in the measure of structural change.

4. Data

The data is taken from the World Development Indicators (2007), with some complementary data on value added and aggregate expenditure taken from the UNCTAD Handbook of Statistics (2008) online database. The data for institutions—Public Rights and Civil Liberties—are taken from the Freedom in the World (2007) dataset online at Freedom House. Chinn and Ito (2006) have developed an index that reflects the restrictions on cross-border financial transactions reported in the IMF’s Annual Report on Exchange Arrangements and Exchange Restrictions (see Ito and Chinn (2005) for more details).

Because the study attempts to study only those countries that have experienced a relatively large inflow of remittances, the dataset encompasses countries which have received an average remittances inflow that is higher than 2% of their GDP over the time period 1970-2006. Inclusion of countries depended on data availability. The WDI dataset is better suited than the IMF data because it claims to have corrected the raw IMF data to make possible the comparability of the series across countries, and also has the advantage that it has amended for missing data with their own estimates. The resulting unbalanced panel dataset comprises yearly observations for about 60 countries with differing degrees of completeness over the period 1970 – 2006. An overview of the data and variables can be found in Appendix A. A more detailed discussion of the data on remittances follows.

4.1. Data on Remittances

The available data does not capture very well the phenomenon of remittances because of two reasons: unclear national account guidelines for recording remittances, and the fact that vast flows sent through informal channels escape being recorded at all. Over time, the process of data gathering has improved as well as the accessibility of the formal money transaction channels to migrants. The current Balance of Payments (BoP) framework methodology generates ambiguity in the recording of remittances because there are three posts which closely approximate the true measurement of the concept of remittances. These are defined below as in Reinke (2007) who states that, “depending on their specific needs, data users can decide which of these components best represent their notion of remittances”.

Workers’ Remittances covers current transfers by migrants who are employed in new economies

and considered residents there. A migrant is a person who comes to an economy and stays there, or is expected to stay, for a year or more.

Compensation to Employees comprises wages, salaries, and other benefits earned by

individuals—in economies other than those in which they are residents—for work performed for and paid for by residents of those economies. Individuals leaving their country with the intention of living in a new economy for a year or longer will be considered residents of the new economy (with a few exceptions, notably students, medical patients, diplomats and military personnel).

Migrants’ transfers are contra-entries to the flow of goods and changes in financial items that

This study uses the sum of Workers’ Remittances and Compensation to Employees, excluding the third term of the often used-triad for measuring remittances. Workers’ Remittances comprise a good share of the flows which, theoretically, are deemed to be remittances. However, not all people who remit are resident migrants. Compensation to Employees complements Workers’ Remittances by adding the cash earned by seasonal workers, theoretically a component of remittances, but it also includes the incomes of diplomats and the money spent in the host countries by the non-residents. Including or excluding this term becomes arbitrary as neither eliminates measurement error. However, two reasons validate its inclusion. First, it is conceivable that in developing countries the share of income from seasonal workers and non-residents in host countries heavily outweighs the meager salaries of diplomats. Second, because more countries publish data on the combined posts (154) instead of on the separate posts (104/107), it was opted to use the combined measure.

Concerning Migrant Transfers, there are two reasons for its exclusion. The first being its different nature compared to the other two posts. Reinke (2007) asserts that in the BoP framework, Compensation to Employees is a component of income while workers’ remittances are a component of current transfers; but both are part of the current account. Migrants’ transfers are a component of capital transfers, which is part of the capital account. Chami et al. (2008) argue that both types of Migrants’ Transfers (the transfer of accumulated assets by migrant residents and the reclassification of assets as a result of a change in residency status) are fundamentally different from remittances and may not involve actual flows. Second, Durand et al. (1996) argue that “sending monthly remittances to Mexico and returning home with savings are interrelated behaviors and represent different ways of accomplishing the same thing: repatriating earnings from the United States”. A last and no less important reason for its exclusion is the insufficient availability of data; only 49 countries report data on Migrants’ Transfers.

5. Results

5.1. The Econometric Model

First of all, the properties of the variables must be analyzed in order to avoid the possibility of spurious regressions from dealing with nonstationary variables. By testing for unit roots, it is possible to see if variables are, or can become, stationary. For the variables that enter the model in log format, the unit root test is done with these series in logs. Since the dataset is an unbalanced panel, it was decided to use the Fisher unit root test which combines the p-values from N independent unit root tests, as developed by Maddala and Wu (1999). Fisher's test assumes that all country series are non-stationary under the null hypothesis against the alternative that at least one series in the panel is stationary. This alternative hypothesis is quite weak because it is not known how many country series besides the ‘at least one’ are stationary. Indeed, for several variables the nulhypothesis is rejected even though only a couple of countries show stationary series. This is why also the individual unit root tests are considered to form an opinion about the overall stationary of the series7. The variable is deemed stationary when more than 60% of the cross-sections reject the nulhypothesis of a unit root at the 10% significance level. The choice of 60% is somewhat arbitrary but it reflects a bare minimum (most variables have values higher than 80%) and is a more reliable alternative to blindly concluding from the Fisher test.

7

The unit root tests have shown that all variables are nonstationary but contain unit roots, except for Vikström’s θ (sc_proan), which is stationary. This means that the analysis can proceed along two paths. One is to look for cointegrating relationships and apply what is called a general error correction model. This approach enhances the ‘long-run information’ in the data (Maddala, 1992, and Dutta and Ahmed, 1999), but is difficult and tedious to apply in a two-step system of equations. An alternative, simpler, path is to apply a first differences transformation to make the series stationary. Such a model can then be estimated with compatible techniques, such as OLS. If the first difference does not render the data I(0) one can second difference the data (Weshah, 2003), as confirmed in the results of the unit root tests. Thus, the OLS estimators will be unbiased and consistent, though they will be inefficient and the sampling variance will generally not be consistently estimated. Over-differencing is not a serious issue in OLS (Plosser and Schwert, 1977). First differencing eliminates individual country effects in the error term, removing the need to correct for them by means of fixed effects (FE). Additionally, the interpretation of the OLS coefficients from a first-differenced specification is simple: how changes in the dependent variable over a year correspond to the changes of the explanatory variables over that same year.

Knowing the form that the different variables will take when entering the models, the independent variables must be tested for possible collinearity. Tables 1-4 in Appendix C show the correlation matrices for the right-hand side variables of the different models. None of the pair-wise correlations are higher than 0.8, satisfying the rule of thumb indicating no damaging multicollinearity. The highest correlation is 0.47, between the two institutional variables of public rights d(pr) and civil liberties d(cl), which is far from posing a problem. Additionally, auxiliary regressions were made with the explanatory variable(s) of interest on the left-hand side and the control variables on the right-hand side8. Each of these regressions enables the computation of the variance inflation factor (VIF). The results show safe VIF scores for all regressions9, so that multicollinearity is not a problem.

The OLS specification has one drawback: it can not account for possible endogeneity in right-hand side variables. Remittances and the variables of aggregate expenditure are liable to endogeneity because it is conceivable that other variables in the equations contribute towards explaining them. For instance, people who remit may have migrated because of a lack of opportunities to generate (sufficient) income in their home countries. In order to control for possible reverse causality, remittances are always entered with a one-year lag. Still, this does not assure their exogeneity, a necessary assumption of OLS. The relevant explanatory variables had to be tested for endogeneity as embedded in their respective equations. The Hausmann test proposes regressing the presumably endogeneous variable on its instruments and the exogenous variables. The residuals of this auxiliary equations are then included as an additional explanatory variable in the initial equation. If the residuals are statistically unequal to zero, instrumental variables are needed because least squares is not consistent.

In the end, only three regressions showed endogeneity of the explanatory variable (see Tables 1 and 2 in Appendix D). To account for this endogeneity, a Generalized Method of Moments (GMM) estimation procedure was applied. It builds on first-differenced models and can account for endogenous right-hand side variables by working with instrumental variables (IV). As a matter of fact, OLS can be considered a type of GMM estimation when it fulfills the condition that the right-hand side variables are uncorrelated with the residual. The challenge of working

with IVs is to find a set of variables, called instruments, which are correlated with the explanatory variable while being uncorrelated with the disturbances. To achieve this, Anderson and Hsiao (1981) proposed using a first difference transformation to eliminate fixed effects, resulting in an error term amenable to consistent estimation by using past values of the variables as instruments. GMM can be made robust to autocorrelation and heteroskedasticity.

5.2. Do Remittances show Macroeconomic Effects?

The results of Equations 1-3 are shown in Table 1, depicting the effects of remittances on investment, consumption and imports. A White-diagonal correction is applied to all estimates, making them robust to observation specific heteroskedasticity in the disturbances. Autocorrelation is not a problem as seen from the satisfying Durbin-Watson (D-W) statistics accompanying each regression output. Since a panel suffers from the inconvenience that it can have both serial and cross-sectional autocorrelation, lower D-W statistics of around 1.60 are considered good10. All regressions show acceptable D-W scores and all OLS regressions have a significant F-statistic.

The models estimated with GMM require additional considerations. The standard errors are computed using the estimated residuals to correct for any general kind of heteroscedasticity. Two tests, displayed in the tables, are applied to uncover whether the instruments are (1) relevant (instruments must correlate with the endogenous variables) and (2) valid (they must fulfill the overidentifying restrictions). For the first test, the endogenous variable is regressed on its instruments and the exogenous variables. An F-test must reject its nulhypothesis that all coefficients are equal to zero. For the second test, Sargan’s J-statistic (1958) is applied. According to Andrews’ (2000) simulation study, it is the criterion that performs best for moment selection, and it is also the method recommended by Arellano and Bond (1991) for IV selection. If the test rejects the null hypothesis of the instruments being optimal, then the estimates of the model should be interpreted cautiously because this may indicate that either the model is misspecified and/or that some of the instruments are invalid (Jalan and Ravallion, 1999). For each model, the most appropriate batch of instruments was selected according to both the F- and J- statistics. Even so, it is very difficult to find meaningful instruments for remittances. If they are found, the risk is high that the explanatory power of the model will be destroyed because they eliminate most of the variation in the explanatory variables (Bound et al. 1995; Wang and Zivot 1998).

Dependent Variable: Investment (%GDP) Investment (US$) Consumption (%GDP) Consumption (US$) Imports (%GDP) Imports (US$)

constant -0.435** -0.414* 0.000 0.001 constant 0.057 0.046 0.037*** 0.035*** constant 0.040 -0.378 -0.009 -0.010 (0.214) (0.214) (0.011) (0.011) (0.139) (0.151) (0.003) (0.003) (0.409) (0.278) (0.007) (0.007) rem (%GDP) 0.127* 0.005† rem (%GDP) 0.179* 0.002 rem (%GDP) -1.115** 0.005 (0.065) (0.003) (0.105) (0.002) (0.481) (0.003)

rem per capita 0.006† 0.000

rem per capita 0.007† 0.0003*** rem per capita 0.033† 0.0005† (0.004) (0.240) (0.005) (0.000) (0.022) (0.0003) GDP 7.926* 7.409† 0.921*** 0.907*** savings -0.502*** -0.502*** -0.004*** -0.004*** Import Value 0.007 0.000 0.000 0.000 (4.474) (4.499) (0.002) (0.245) (0.000) (0.065) (0.002) (0.002) (0.008) (0.000) (0.000) (0.000) interest rate 0.030 0.029 0.002† 0.002 Wealth dilution 0.000 0.000 0.000 0.000 Exchange Rate 0.005 0.000 0.000 0.000 (0.028) (0.028) (0.002) (0.002) (0.714) (0.000) (0.000) (0.000) (0.013) (0.000) (0.000) (0.000) dom credit 0.038† 0.037† 0.000 0.000 Inflation 0.010*** 0.009*** 0.0001* 0.0001* GDP 4.276 5.587 1.146*** 1.141*** (0.024) (0.024) (0.001) (0.001) (0.002) (0.000) (0.000) (0.000) (8.057) (5.184) (0.157) (0.158) Chin&Ito Index 0.208 0.197 0.004 0.003 Interest rate 0.068** 0.067** 0.000 0.000 Exports 0.689*** 0.719*** 0.005*** 0.005*** (0.274) (0.275) (0.014) (0.014) (0.029) (0.030) (0.001) (0.001) (0.104) (0.067) (0.002) (0.002) public rights -0.005 0.001 -0.009 -0.009 Unemploy ment 0.000 0.005 -0.001 0.000 FDI 0.261* 0.176* 0.004** 0.004* (0.247) (0.246) (0.010) (0.010) (0.072) (0.071) (0.002) (0.002) (0.145) (0.093) (0.002) (0.002) civil liberties -0.192 -0.179 -0.015 -0.015 Pop>15 2.540 2.528 0.037 0.047 Aid 0.001 0.035 -0.002** -0.001* (0.359) (0.359) (0.019) (0.019) (2.840) (2.857) (0.048) (0.047) (0.058) (0.055) (0.001) (0.001) Aid 0.102** 0.100** -0.001 -0.001 CPI 0.032 0.036 -0.001 -0.001 Chin&Ito Index 0.041 0.274 -0.002 -0.002 (0.043) (0.043) (0.002) (0.002) (0.047) (0.044) (0.001) (0.001) (0.419) (0.362) (0.009) (0.009) obs. 767 767 643 643 250 250 204 204 551 552 526 527 r-sq 0.032 0.031 0.054 0.052 0.397 0.393 0.135 0.162 0.318 0.325 0.187 0.187 adj. r-sq 0.022 0.020 0.042 0.040 0.377 0.373 0.099 0.128 0.308 0.315 0.175 0.174 D-W 2.046 2.045 2.093 2.092 2.074 2.071 1.793 1.873 2.210 2.206 2.150 2.153 F-test prob. 0.001 0.003 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 IV F-test - - - 2.033 - - - IV F-test prob - - - 0.018 - - - J-test - - - 72.671 - - - J-test prob - - - 1.110 - - -

Note: standard errors in parentheses. ***, **, *, and † denote statistical significance at the 1, 5, 10 and 15 percent, respectively. Standard errors in parentheses. For equations estimated with GMM, additional information is presented: the IV F-test for instrument relevance, and the J-test for instrument validity.

Table 2 shows the results for model 4 where different measures of structural change are regressed on remittances and the set of control variables.

Table 2. OLS results for Model 4: The Effect of Remittances on Structural Change

Dep. Variable: Viström's θ _{Agriculture VA Industry VA Services VA}

constant 0.027*** 0.027*** -0.308** -0.331** -0.181† -0.169 0.490*** 0.500***

(0.003) (0.003) (0.136 (0.137) (0.123) (0.123) (0.137) (0.139)

rem (%GDP) 0.000 0.051 0.006 -0.057

(0.001) (0.036 (0.038) (0.042)

rem per capita -0.0001* 0.007*** -0.002 -0.005*

(0.00007) (0.002) (0.002) (0.003) GDP 0.121* 0.138** -8.614** -9.158*** 7.000** 7.390** 1.633 1.780 (0.063) (0.063) (3.447) (3.348) (3.201) (3.219) (3.342) (3.343) Education 0.000 0.001 -0.136*** -0.036 -0.023 -0.018 0.055 0.055 (0.001) (0.001) (0.039) (0.042) (0.025) (0.023) (0.042) (0.041) exports 0.002** 0.002** -0.136*** -0.139*** 0.174*** 0.174*** -0.040 -0.037 (0.001) (0.001) (0.039) (0.039) (0.040) (0.040) (0.048) (0.048) FDI 0.001 0.001 -0.102* -0.103* -0.007 -0.008 0.108** 0.111** (0.001) (0.001) (0.053) (0.053) (0.042) (0.042) (0.048) (0.049) Inflation -0.001*** -0.001*** 0.057** 0.056** -0.003 -0.001 -0.053** -0.054** (0.000) (0.000) (0.023) (0.021) (0.026) (0.025) (0.022) (0.021) Exchange Rate 0.0000008*** 0.0000009*** -0.0003 0.00003*** 0.000 0.000 0.00003** 0.00003*** (0.0000002) (0.0000002) (0.000009) (0.000009) (0.000) (0.000) (0.00001) 0.000009) Infrastructure 0.000 0.000 0.001 0.002 -0.017*** -0.017*** 0.016* 0.015* (0.000) (0.000) (0.008) (0.008) (0.006) (0.006) (0.009) (0.009) Civil Liberties -0.017** -0.017** 1.034** 1.027** -0.223 -0.223 -0.811* -0.803* (0.008) (0.008) (0.440) (0.435) (0.288) (0.287) (0.451) (0.445) Public Rights -0.008 -0.008 0.417 0.410 -0.059 -0.042 -0.356 -0.367 (0.007) (0.007) (0.329 (0.324) (0.351) (0.343) (0.372) (0.370) observations 177 177 177 177 177 177 177 177 r-squared 0.184 0.199 0.249 0.260 0.212 0.214 0.139 0.139 adj. r-squared 0.134 0.150 0.204 0.215 0.165 0.166 0.087 0.088 D-W 1.446 1.435 1.732 1.749 1.537 1.539 1.499 1.511 Prob (F-test) 0.000 0.000 0.000 0.000 0.000 0.000 0.005 0.004

Note: standard errors in parentheses. ***, **, *, and † denote statistical significance at the 1, 5, 10 and 15 percent, respectively. Standard errors in parentheses.

this effect is only marginally statistically significant. Interestingly, the import regression estimated in GMM shows a large and negative effect of an increase in remittances (%GDP), while the other remittances coefficients estimated in OLS are slightly positive but only marginally significant.

The measure of remittances per capita seems especially pertinent for the structural change models—it is remittances per capita which arrives at all the statistically significant effects of these regressions while remittances over GDP is constantly insignificant. While an increase in the change of remittances per capita increases the change in value added of agriculture, it decreases the change in the percent value added of services by 0.005, and has no effect on the industrial sector. However, the negative coefficient for Vikström’s θ means that, all in all, an increase in the change in remittances actually reduces the cumulative change in economic structure. This suggests that, by reducing the sum of the each sector individual pace of change, remittances may reduce the rate of overall structural change.

5.3. Do Remittances Affect Structural Change through investment, consumption, and imports?

The regressions in each system of equations (5-7) were estimated simultaneously with alternating OLS or GMM specifications, depending on the need to account for endogeneity. The effect of remittances on structural change happens once directly and once indirectly, through its effect on either investment, consumption or imports. Two conditions must be met for remittances to have this indirect effect: (1) remittances must be statistically significant in the aggregate expenditure regressions (Table 1) and (2) these aggregate expenditure posts must in turn be significant in the structural change regressions11. Under such a setting, the models account for both possible effects of remittances. Only when the respective regressions showed statistically significant outcomes for both these conditions, was the system of equations regressed simultaneously and its results depicted in the following tables12. Table 3 shows the resulting outcome for the Systems of Equations 5 where remittances enter the regressions for structural change (industry and services value added) directly and indirectly by channeling through investment. Table 4 shows the same for the potential channel of consumption, and Tables 5a and 5b illustrate the regressions for the potential channel of imports.

11 _{The regressions for investment, consumption and import on structural change were carried out, but for}

the sake of concision, do not appear here.

12

Table 3. OLS results for System of Equations 5: The Effect of Remittances on Investment and on Structural Change, through Investment

Dependent Variable Investment (US$) Industry VA Investment (US$) Services VA

constant 0.000 -0.300** 0.000 0.581***

(0.010) (0.153) (0.010) (0.165)

rem (%GDP) 0.005† 0.025 0.005† -0.057

(0.004) (0.058) (0.004) (0.062)

rem per capita - -

- - GDP 0.921*** 0.921*** (0.182) (0.182) Interest Rate 0.002*** 0.002*** (0.001) (0.001) Dom Credit 0.000 0.000 (0.001) (0.001)

Chin & Ito Index 0.004 0.004

(0.015) (0.015) Aid -0.001 -0.001 (0.002) (0.002) Civil Liberties -0.015 -0.207 -0.015) -0.880** (0.016) (0.326) (0.016) (0.353) Public Rights -0.009 -0.157 -0.009 -0.310 (0.012) (0.295) (0.012) (0.319) Investment 2.413** -3.241*** (0.975) (1.056) GDP 3.599 6.937† (4.145) (4.489) Education 0.012 0.023 (0.042) (0.045) Exports 0.174*** -0.012) (0.034) (0.036) FDI -0.006 0.100** (0.042) (0.046) Inflation 0.000 -0.062*** (0.019) (0.021) Exchange Rate 0.000 0.000 (0.000) (0.000) Infrastructure -0.018 0.018 (0.013) (0.014) obs. 643 158 643 158 r-sq 0.054 0.243 0.054 0.186 adj. r-sq 0.042 0.186 0.042 0.124 D-W 2.095 1.689 2.095 1.573

Table 3. OLS results for System of Equations 6: the Effect of Remittances on Consumption and on Structural Change, through Consumption Dependent Variable Consumpt (%GDP) Viström’s θ Consumpt (%GDP) Viström's θ Consumpt (%GDP) Industry VA Consumpt (%GDP) Industry VA Consumpt (%GDP) Services VA Consumpt (%GDP) Services VA constant 0.057 0.027*** 0.046 0.028*** 0.057 -0.195 0.046 -0.179 0.057 0.478*** 0.046 0.500*** (0.143) (0.003) (0.151) (0.003) (0.143) (0.147) (0.151) (0.149) (0.143) (0.162) (0.151) (0.163) rem (%GDP) 0.179* 0.000 0.179* 0.034 0.179* -0.074 (0.101) (0.001) (0.101) (0.051) (0.101) (0.056)

rem per capita 0.007 -0.0002** 0.007 0.000 0.007 -0.008†

(0.006) (0.000009) (0.006) (0.004) (0.006) (0.005) Savings -0.502*** -0.502*** -0.502*** -0.502*** -0.502*** -0.502*** (0.044) (0.044) (0.044) (0.044) (0.044) (0.044) Wealth Dilution 0.000 0.000 0.000 0.000 0.000 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Inflation 0.010*** 0.009*** 0.010*** 0.009*** 0.010*** 0.009*** (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) Interest Rate 0.068** 0.067** 0.068** 0.067** 0.068** 0.067** (0.028) (0.028) (0.028) (0.028) (0.028) (0.028) Unemployment 0.000 0.005 0.000 0.005 0.000 0.005 (0.068) (0.069) (0.068) (0.0069) (0.068) (0.069) Pop>15 2.540 2.528 2.540 2.528 2.540 2.528 (2.251) (2.462) (2.451) (2.462) (2.451) (2.462) CPI 0.032 0.036 0.032 0.036 0.032 0.036 (0.037) (0.037) (0.037) (0.037) (0.037) (0.037) Consumption 0.001 0.001 -0.055* -0.057* 0.097*** 0.100*** (0.001) (0.001) (0.032) (0.032) (0.035) (0.035) GDP 0.141* 0.177** 5.360 5.825† 3.766 4.344 (0.077) (0.076) (3.759) (3.806) (4.130) (4.165) Education 0.000 0.001 -0.021 -0.015 0.051 0.053 (0.001) (0.001) (0.035) (0.035) (0.039) (0.035) Exports 0.002*** 0.002*** 0.170*** 0.166*** -0.024 -0.022 (0.001) (0.001) (0.032) (0.032) (0.035) (0.035) FDI 0.001 0.001 -0.010 -0.012 -0.117*** 0.121*** (0.001) (0.001) (0.041) (0.041) (0.045) (0.045) Inflation -0.001*** -0.001*** -0.008 -0.006 -0.044** -0.045** (0.000) (0.000) (0.019) (0.019) (0.021) (0.021) Exchange Rate 0.000** 0.000001** 0.000 0.000 0.000 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Infrastructure 0.000 0.000 -0.017* -0.018* 0.017† 0.016 (0.000) (0.000) (0.010) (0.010) (0.011) (0.011) Civil Liberties -0.018*** -0.018*** -0.316 -0.325 -0.697* -0.682* (0.007) (0.006) (0.324) (0.325) (0.356) (0.355) Public Rights -0.008 -0.006 -0.098 -0.069 -0.332 -0.337 (0.006) (0.006) (0.291) (0.290) (0.319) (0.317) obs. 250 170 250 170 250 170 250 170 250 170 250 170 r-sq 0.397 0.193 0.393 0.224 0.397 0.241 0.393 0.239 0.397 0.177 0.393 0.181 adj. r-sq 0.377 0.137 0.373 0.170 0.377 0.188 0.373 0.186 0.377 0.119 0.373 0.124 D-W 2.132 1.461 2.127 1.485 2.132 1.676 2.127 1.674 2.132 1.586 2.127 1.609

Table 4a: OLS/GMM System of Regressions Results for Equations 7: The Effect of Remittances on Imports and on Structural Change, through Imports

Estimation Method

GMM with IVs for

Remittances OLS OLS

GMM with IVs for

Remittances OLS Dependent Variable Import (%GDP) Viström's θ Import (%GDP) Viström's θ Import (US$) Viström's θ Import (%GDP) Industry VA Import (%GDP) Industry VA cons -0.043 0.027*** -0.116 0.027*** 0.021*** 0.027*** -0.043 -0.173† -0.116 -0.168 (0.364) (0.003) (0.241) (0.003) (0.007) (0.003) (0.364) (0.120) (0.241) (0.143) rem (%GDP) -0.714** -0.001 -0.714** 0.024 (0.348) (0.001) (0.348) (0.035) rem per capita 0.034*** -0.0002** 0.000 -0.0002** 0.034*** 0.000 (0.009) (0.00007) (0.000) (0.00009) (0.009) (0.004) Import Value 0.007 0.000 0.000 0.007 0.000 (0.008) (0.000) (0.000) (0.008) (0.000) Exchange Rate 0.005 0.000 0.000 0.005 0.000 (0.013) (0.000) (0.000) (0.013) (0.000) GDP 2.844 -1.720 0.342*** 2.844 -1.720 (7.485) (4.384) (0.129) (7.485) (4.384) Exports 0.725*** 0.721*** 0.005*** 0.725*** 0.721*** (0.091) (0.049) (0.001) (0.091) (0.049) FDI 0.196* 0.182** 0.005** 0.196** 0.182** (0.119) (0.092) (0.003) (0.119) (0.092) Aid 0.028 0.032 -0.002** 0.028 0.032 (0.054) (0.037) (0.001) (0.054) (0.037) Chin&Ito Index 0.089 0.293 -0.001 0.089 0.293 (0.397) (0.434) (0.012) (0.397 (0.434) Imports 0.001* 0.001** 0.043† -0.059* -0.055** (0.0006) (0.0005) (0.027) (0.034) (0.028) GDP 0.129** 0.149** 0.110 6.596 6.934* (0.060) (0.072) (0.080) (3.249) (3.628) Education 0.000 0.001 0.002* -0.026 -0.021 (0.001) (0.001) (0.001) (0.025) (0.034) Exports 0.001 0.001 0.001* 0.228*** 0.221*** (0.001) (0.001) (0.001) (0.043) (0.038) FDI 0.001 0.001 0.001† -0.002 -0.005 (0.001) (0.001) (0.001) (0.039) (0.040) Inflation -0.001*** -0.001*** -0.001*** -0.006 -0.004 (0.0004) (0.000) (0.000) (0.024) (0.019) Exchange Rate 0.0000008*** 0.0000009** 0.0000007* 0.000 0.000 (0.0000002) (0.0000004) (0.0000004) (0.000) (0.000) Infrastructure 0.000 0.000 0.000 -0.016*** -0.017† (0.000) (0.000) (0.000) (0.006) (0.010) Civil Liberties -0.015** -0.014** -0.015** -0.355 -0.340 (0.007) (0.006) (0.007) (0.277) (0.315) Public Rights -0.008 -0.007 -0.007 -0.032 -0.048 (0.006) (0.006) (0.006) (0.318) (0.279) obs. 339 177 552 177 527 .226 339 177 552 177 r-sq 0.149 0.202 0.324 0.228 0.065 0.168 0.149 0.234 0.324 0.232 adj. r-sq 0.128 0.149 0.314 0.177 0.051 0.027 0.128 0.182 0.314 0.181 D-W 1.991 1.416 2.223 1.412 2.093 1.516 1.991 1.562 2.223 1.553 IV F-test 2.232 - - 2.232 - Prob. 0.010 - - 0.010 - J-stat 5.472 - - 5.495 - Prob. 0.361 - - 0.358 -

Table 4b: System of Regressions Results for Equations 7 (cont.): The Effect of Remittances on Imports and on Structural Change, through Imports

Estimation Method

GMM with IVs for

Remittances OLS OLS GMM with IVs for Imports

Dependent Variable Import (%GDP) d(ser_3) Import (%GDP) d(ser_3) Import (US$) d(ser_3) Import (US$) d(ser_3) cons -0.076 0.503*** -0.116 0.517*** 0.021** 0.289* 0.021** 0.415* (0.408) (0.142) (0.285) (0.143) (0.008) (0.162) (0.008) (0.217) rem (%GDP) -1.074** -0.065 0.004 -0.046 (0.524) (0.053) (0.004) (0.050) rem per capita 0.034† -0.006** 0.000 -0.006* (0.022) (0.003) (0.000) (0.003) Import Value 0.010 0.000 0.000 0.000 (0.009) (0.000) (0.000) (0.000) Exchange Rate 0.010 0.000 0.000 0.000 (0.010) (0.000) (0.000) (0.000) GDP 6.436 -1.698 0.370** 0.341** (7.621) (5.464) (0.170) (0.173) Exports 0.649*** 0.718*** 0.005*** 0.005** (0.101) (0.067) (0.002) (0.002) FDI 0.267* 0.183** 0.005*** 0.005 (0.140) (0.093) (0.002) (0.002) Aid 0.032 0.032 -0.002** -0.002* (0.070) (0.055) (0.001) (0.001) Chin&Ito Index 0.317 0.276 -0.001 -0.002 (0.456) (0.365) (0.010) (0.010) Imports 0.195* 0.188† 5.483** 4.005 (0.117) (0.116) (2.688) (4.479) GDP 1.051 1.376 0.738 0.016 (3.563) (3.532) (4.278) (4.584) Education 0.055 0.055 0.040 0.033 (0.048) (0.047) (0.048) (0.049) Exports -0.200** -0.195** -0.082* -0.066 (0.098) (0.097) (0.050) (0.052) FDI -0.096** 0.099** 0.075* 0.086** (0.044) (0.045) (0.043) (0.043) Inflation -0.035** -0.036** -0.033* -0.034* (0.017) (0.017) (0.019 (0.019) Exchange Rate 0.0002** 0.00002** 0.00003** 0.00003*** (0.00001) (0.00001) (0.00001) (0.00001) Infrastructure 0.017† 0.016 0.010 0.013 (0.011) (0.011) (0.010) (0.012) Civil Liberties -0.277 -0.289 -0.262 -0.503 (0.536) (0.526) (0.490) (0.0607) Public Rights -0.242 -0.215 -0.142 -0.007 (0.326) (0.325) (0.342) (0.329) obs. 346 164 552 164 526 160 527 160 r-sq 0.025 0.191 0.324 0.198 0.067 0.048 0.065 0.087 adj. r-sq 0.002 0.133 0.314 0.140 0.052 -0.023 0.051 0.019 D-W 1.970 1.471 2.223 1.481 2.089 1.479 2.092 1.453 IV F-test 2.318 9.997 - 9.9264 - 13.404 - 4.238 Prob. 0.000 0.000 - 0.000 - 0.000 - 0.000 J-stat 2.521 1.992 7.274 7.054 Prob. 0.925 0.737 0.122 0.133