Remittances, financial development and economic growth : is the effect of remittances on economic growth conditional on financial development and is this effect different for countries in the Sub-Saharan African region?

(1)

Remittances, Financial Development and Economic Growth

Is the effect of remittances on economic growth conditional on financial development and is this effect different for countries in the Sub-Saharan African region?

Master Thesis MSc Economics

Specialization International Economics and Globalization Faculty of Economics and Business

August 15, 2017

Author: Annick Besançon Student number: 10283366

Email: annickbesancon@hotmail.com Supervisor: Ms. N.J. Leefmans

Second reader: Prof. dr. F.J.G.M. Klaassen Word count: 13101

(2)

STATEMENT OF ORIGINALITY

This document is written by Student Annick Besançon who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Abstract

The past two decades, the amount of remittances has increased significantly in Sub-Saharan Africa and in developing countries in general. As a result of this, increased attention has been given to these financial flows as it is believed that remittances could play an important part in poverty reduction and economic development. Despite the increasing importance of

remittances, the effect of remittances on economic growth is still not well-understood.

This study investigates the effect of remittances on economic growth in developing countries, and in particular, whether this effect is conditional on the presence of a developed financial sector and whether this effect is different for Sub-Saharan African countries compared to other developing countries.

A panel dataset of 120 developing countries, of which 40 Sub-Saharan African countries, over the period 1990 to 2015 is used to perform a pooled ordinary least squares estimation, a fixed-effects ordinary least squares estimation and a system generalized method of moments

estimation. I find a weak positive effect of remittances on economic growth and a negative coefficient for the interaction term between remittances and financial development, indicating substitutability between remittances and financial development. These findings are controlled for the endogeneity of remittances and do not depend on the particular measure of financial sector development used. Whether the effect of remittances on economic growth is different for Sub-Saharan African countries specifically, depends on the dataset used. When annual data is used, no evidence is found for a separate effect for Sub-Saharan African countries. When 4-year averages are used, evidence is found that the positive effect of remittances on economic growth is reduced when countries are situated in the Sub-Saharan African region.

(4)

1. Introduction

For several decades, foreign direct investment and official foreign aid used to make up the biggest share of total international capital flows to Sub-Saharan Africa (SSA). However, the last decade the amount of remittances - the cross-border money transfers from migrants to their country of origin - has increased significantly (Figure 1). In some SSA countries, like Lesotho and Kenya, the amount of remittances received has even surpassed that of foreign direct investment. As a result of this, attention to this source of external finance has also increased. This is especially so, since it is believed that remittances could play an important part in poverty reduction and economic development.

Figure 1: Financial inflows in Sub-Saharan Africa

Source: Graph constructed by the author using data from the World Development Indicators

Extensive research has been performed on the micro and macro determinants of remittances and on the effects of remittances on poverty and economic growth. This research agrees that the most important micro determinants of remittances are both altruism and self-interest. There is also a general consensus that there is a positive effect of remittances on poverty reduction (e.g. Adams, 2005). However, there is no consensus on the effect of remittances on economic growth. Some papers find a positive effect (Giuliano and Arranz, 2009), others a negative effect (Chami et al, 2003) and some even find no effect at all (Barajas et al., 2009).

(6)

One of the reasons for the mixed results might be omitted variable bias. According to some papers (Bjuggren et al (2010); Aggarwal et al., 2010) the marginal importance of remittances as a financial source for investment decreases with a more developed financial sector. This suggests that an interaction term between remittances and financial development should be included. There are already some papers that have studied the effects of remittances on economic growth conditional on financial development (Guiliano and Arranz, 2009; Singh et al., 2010), however again with mixed results. Therefore, more research is needed in this area. Besides that, most studies focus on the developing countries as a group or on Sub-Saharan Africa specifically. However, there is no study yet that answers the question whether the effect of remittances on economic growth (conditional on financial sector development) is significantly different for countries in the Sub-Saharan African region from the group of developing countries as a whole.

This thesis will contribute to the existing literature in several ways. First of all, as more data have become available this study will look at the effect of remittances on economic growth in developing countries by using data on six more years compared to previous studies. Second, besides looking at the effect of remittances on economic growth this thesis will also investigate whether this effect is conditional on the presence of a developed financial sector and whether this effect is different for Sub-Saharan African countries compared to other developing countries.

The research questions that this study will answer are therefore: 1) What is the effect of remittances on economic growth in developing countries and, specifically, in Sub-Saharan Africa and 2) is the effect of remittances on economic growth in developing countries and Sub-Saharan Africa specifically, conditional on the financial sector development of the remittance-receiving country.

This thesis tries to answer these questions by performing an empirical analysis using a panel dataset of 120 developing countries, of which 40 Sub-Saharan African-countries, over the period 1990 to 2015. The dataset that will be used contains annual data and is derived from the World Bank’s Development Indicators. Besides performing the empirical analysis using annual data, I will also use 4-year averages to control for business cycle fluctuations.

The methodology is partly based on the paper of Giuliano and Arranz (2009), who perform a pooled OLS regression and a system generalized method of moments (GMM) estimation in order to investigate whether the level of financial depth in the recipient country affects the impact of remittances on economic growth. Besides using these two estimation methods proposed by Giuliano and Arranz, I will also perform a fixed-effects OLS estimation to

(7)

account for unobserved country-specific effects. I use the same growth equation as Giuliano and Arranz to estimate the effect of remittances on economic growth conditional on financial development, and augment this equation with a dummy for Sub-Saharan Africa to answer the question whether remittances have a different effect on economic growth for countries in this region.

The remainder of the thesis is structured as follows. Chapter 2 provides an overview of what is already known in the literature on remittances, economic growth and financial development. Chapter 3 describes the methodology and estimation method that will be used to answer the research questions. Chapter 4 gives more information on the data used in this thesis. Chapter 5 presents the results of the empirical model and chapter 6 concludes.

(8)

2. Literature review

This chapter gives an overview of the existing literature on remittances, financial sector development and economic growth, both theoretical and empirical. Section 2.1 will discuss the theoretical channels through which remittances can affect economic growth and the significance of these channels according to the empirical literature. Section 2.2 will focus on the role of financial sector development in the remittance-growth nexus and elaborates on the research that has been done in this area so far.

2.1 Remittances and economic growth

According to Barajas et al. (2009), there are three main channels through which remittances can affect economic growth; capital accumulation, labour force reduction and total factor productivity growth.

2.1.1 The capital accumulation channel

The capital accumulation channel leads to a positive effect of remittances on economic growth in several ways. First of all, an increase in remittances leads to higher disposable income and savings, which could lead to higher economic growth when used for investment. Evidence for this channel is for example found by Woodruff and Zenteno (2007) and by Yang (2005). Woodruff and Zenteno (2007) show, by performing OLS- and IV-estimation based on survey data of more than 600 self-employed workers and small firm owners located in Mexico, that remittances lead to higher investment. Their result is consistent with the results of Yang (2005). Based on four surveys conducted by the Philippine government he performs an OLS estimation and finds that positive migrant shocks lead to enhanced human capital accumulation and entrepreneurship in the country of origin. He also finds that households work more hours in self-employment and become more likely to start up enterprises.

It is sometimes argued that the capital accumulation channel is not very significant as most remittances are used for consumption and not for investment. However, even if remittances are mostly used for consumption, they can still have a positive effect on economic growth through the multiplier effect. This is for example the case in Mexico. Adelman and Taylor (1990) find that for every dollar received in Mexico through remittance inflows, Mexico’s gross national product increases on average with 2.90 dollar.

The second way through which remittances affect capital accumulation and economic growth is through their role as an automatic stabilizer. Several studies indicate that

(9)

remittances are a relatively stable source of income that can act as collateral for loans, improving households’ creditworthiness and alleviating credit constraints (e.g. Ratha, 2007 and Chami et al., 2009). Both of these make it easier for households to invest in high return projects, adding to economic growth. In order to answer the question whether remittance inflows reduce output growth volatility in remittance-dependent economies, Chami et al. (2009) use a sample of 70 countries over the 1970-2004 period and estimate an OLS-, IV- and GMM estimation. Their results provide evidence that on average remittances have contributed to reduce output volatility in these countries.

Third, remittances can also affect human capital accumulation. Research shows that most households invest remittances in health, education and other basic spending, adding to human capital and resulting in economic growth (e.g. De Haas, 2009). The findings of Calero et al. (2009) support this. Based on a household survey including 13581 households in Ecuador, they find that 89.4% of remittance income is used for education, health and rent.

2.1.2 The reduced labour force channel

The reduced labour force channel leads to a negative effect of remittances on economic growth. When remittances are seen as a substitute for labour income, the labour force participation or labour supply might be reduced. This view is enforced by the moral hazard effect first described by Chami, Fullenkamp and Jahjah (2005). Based on an OLS estimation with up to 113 countries over 29 years, they conclude that remittances have a negative effect on economic growth, because remittance transfers occur under asymmetric information and in a context where monitoring is difficult, leading to a decrease in labour market effort and work force participation by the recipients.

However, according to Catrinescu et al. (2006) the results of Chami et al. (2005) suffer from endogeneity between remittances and economic growth. They show, by estimating a dynamic panel data model for the same time period as Chami et al. (from 1970 until 1998), that there is actually a slight positive effect of remittances on economic growth when endogeneity is properly controlled for. To control for the endogeneity between remittances and economic growth, they use the first- and second lag of remittances in the first-difference equation as an instrument for remittances. Moreover, they show that the positive effect of remittances on economic growth slightly increases when institutional quality is included as a control variable. However, this variable is only significant in a few of the regressions.

(10)

2.1.3 The total factor productivity channel

Remittances can also have an effect on economic growth through a third channel, total factor productivity (TFP) growth. This effect occurs in two ways; through the effect of remittances on the efficiency of domestic investment and through the effect of remittances on the real exchange rate (Barajas et al., 2009).

Remittances can affect the efficiency of domestic investment when the remittance inflows change the quality of domestic financial intermediation. An example of how this can happen, is when the recipient or the remitter himself invests the remittances while having informational disadvantages compared to financial intermediaries or banks. As a result, the recipient might invest in a less efficient investment than a financial intermediary would have done when the remittances would have entered the country as formal capital inflows. This way the efficiency of domestic investment is reduced and therefore economic growth. On the other hand, this effect could also work the opposite direction when the remitter or the recipient has informational advantages compared to financial intermediaries and banks. In that case, the remittances would have a positive effect on economic growth. A second way for remittances to affect the efficiency of domestic investment, is through the ability of an economy’s formal financial system to allocate capital. An increase in remittances leads to an increase in the amount of funds flowing through the banking system. Aggarwal et al. (2010) find that an increase in remittances can lead to enhanced financial development and economic growth, because the increase in the amount of funds leads to economies of scale.1

The second mechanism through which remittances affect total factor productivity growth, is through the Dutch disease effect. When large and sustained remittance inflows lead to an appreciation of the real exchange rate, the so-called ‘Dutch disease’ effect occurs, making the tradable sector less competitive and the production of tradable goods less profitable. Proof for this negative effect of remittances on TFP growth and economic growth has been found by Amuedo-Dorantes and Pozo (2004). Based on 13 Latin-American nations, they find that remittances lead to a real exchange rate appreciation of 22%, negatively affecting economic growth. However, it is not clear to what extent the Dutch disease effect really occurs. Rajan and Subramanian (2005) find that there is no appreciation effect at all and Acosta et al. (2009) find that the Dutch disease effect depends on the quality of the institutions present in the country.

(11)

2.2 Financial sector development, remittances and economic growth

The previous paragraph showed that both in theory and in empirical research, the effect of remittances on economic growth is ambiguous. Besides not properly controlling for endogeneity between the dependent and explanatory variables, another reason for the mixed results might be the omitted variable bias caused by leaving out financial sector development in the growth equation. Section 2.2.1 explains how financial sector development is defined, section 2.2.2 elaborates on the motivation for including financial sector development in the growth equation and section 2.2.3 discusses a few studies that have already investigated the relationship between remittances, financial sector development and economic growth.

2.2.1 Defining financial sector development

The financial sector consists of the set of institutions, instruments, markets, and the legal and regulatory framework that permits transactions to be made through the extension of credit (World Bank, 2017a).

With financial sector development one means the extent to which financial institutions, instruments and markets ameliorate the problems created by information and transaction frictions (Levine, 1997). This can be done by producing information about possible investments and allocating capital; monitoring firms and exerting corporate governance; trading, diversification, and management of risk; mobilization and pooling of savings; and easing the exchange of goods and services.

2.2.2 The role of financial sector development in the remittance-growth nexus

Some recent papers suggest that an interaction term between remittances and financial development should be included in the growth equation (Bjuggren et al., 2010; Aggarwal et al., 2011). The inclusion of the interaction term in the growth equation is based on the discussion in the literature whether financial development and remittances are complements or substitutes.

Remittances can act as a substitute for financial development, as remittances provide an alternative way of financing investment and overcoming liquidity constraints. The positive effect of remittances on investment and economic growth is therefore expected to be larger for countries where financial sector development is relatively low. Bjuggren et al. (2010) find evidence for this view. In their paper, they use a dynamic panel data approach and data for 79 developing countries during 1990-2005 and conclude that the marginal importance of

(12)

remittances as a financial source for investment decreases with a more developed financial sector.

On the other hand, remittances can also act as a complement to financial development as recipients might have a need for financial products or bank deposits to storage received remittances. Aggarwal et al. (2011) find evidence for this complementarity between remittances and financial development, based on a dataset of 109 developing countries for the period 1975-2007. In their paper, several empirical approaches are used. They conduct an OLS estimation including country and time fixed effects, a dynamic system Generalized Method of Moments (GMM) estimation following Arellano and Bover (1995) and an instrumental variables estimation, using two sets of instruments to control for the endogeneity between remittances and economic growth. The first instrument set contains measures of economic conditions in remittance-sending countries and the second set contains variables that capture the views held and the policies pursued by policy-makers in remittance-sending countries with respect to international immigration. Regardless of the estimation method, they find evidence of a positive and significant link between financial development and remittances, indicating complementarity between the two.

2.2.3 Research on remittances, financial sector development and economic growth so far There are a few papers that have already studied the effect of remittances on economic growth conditional on financial development. In this section I will discuss five of them.

The most cited study in this area is the paper of Giuliano and Arranz (2009), who find evidence for the substitutability of remittances and financial development. They use a dataset of 100 developing countries for the period 1975-2002 to perform an ordinary least squares (OLS) regression and, to correct for endogeneity, a system generalized method of moments (SGMM) regression. They divide their dataset into six non-overlapping 5-year periods to control for business cycle fluctuations. Four measures are used to proxy financial development: 1) liquid liabilities of the financial system (M2/GDP), 2) the sum of demand deposits, time deposits, saving deposits and foreign currency deposits to GDP, 3) financial sector claims on the private sector divided by GDP and 4) the ratio of credit provided by the banking sector to GDP. They find that remittances have a positive effect on economic growth and that this effect is even larger in countries with less developed financial systems, providing evidence that remittances and financial development are substitutes.

Bettin and Zazzaro (2011) base their study on the paper of Giuliano and Arranz, but use an additional financial development indicator and also include institutional quality as an

(13)

interaction term with remittances in their model. Their dataset comprises annual data for 66 developing countries for the period 1970-2005, that they divide into seven non-overlapping 5-year periods. Using OLS and SGMM, they estimate the effect of remittances on economic growth conditional on financial development both for the annual data and for the seven non-overlapping 5-year data periods. Their results are consistent with Giuliano and Arranz; they find a positive effect of remittances on economic growth and a negative interaction term between remittances and financial development.

The previous two studies look at the effect of remittances on economic growth in a global perspective. The papers of Nyamongo et al. (2012), Lartey (2013) and Singh et al. (2010) look at the same effect in Sub-Saharan African countries only. Nyamongo et al. (2012) use a dataset of 36 SSA countries from 1980-2009. In contrast with the two aforementioned studies, they split their dataset into ten non-overlapping 3-year periods instead of 5-year periods to avoid losing a substantial number of countries. To estimate the effect of remittances on economic growth conditional on financial development, they conduct a fixed effects OLS regression and a Two Stage Least Squares (TSLS) regression, where lagged variables of the explanatory variables are used as instruments to control for endogeneity. They use two financial development indicators: 1) the ratio of broad money to GDP and 2) the ratio of credit to the private sector to GDP. Nyamongo et al. find a positive effect of remittances on economic growth and in contrast with the two aforementioned studies, a positive interaction effect between remittances and financial development on economic growth, indicating that remittances and financial sector development are complements.

These findings are consistent with the findings of Lartey (2013). He uses annual data for 36 SSA countries from 1990-2008 to conduct a system GMM regression. The financial development indicators used in his study are: 1) private credit as a percentage of GDP, 2) deposit money bank assets as a percentage of GDP, and 3) liquid liabilities of the financial system as a percentage of GDP. Besides looking at the effect of remittances on economic growth conditional on financial depth, he also tries to answer the question whether the impact of remittances on growth is through capital accumulation or through other mechanisms. The results of his regression of remittances on investment provide evidence for the existence of an investment channel through which remittances affect growth.

Singh et al. (2010) use the same dataset as Lartey (2013) and perform a Fixed Effects OLS estimation and a TSLS estimation, using lags of the variables in their system as instruments. Their dependent variable is the log difference of per capita real GDP and the explanatory variables, which are all in logs, are the ratio of remittances to GDP, an interaction

(14)

term between financial development and remittances, and a number of control variables. They also include an interaction term between remittances and institutional quality, to look at the effect of remittances on economic growth conditional on institutional quality. They find a significant negative effect of remittances on economic growth and a positive coefficient for both interaction terms. Therefore, they conclude that a more developed financial sector and higher institutional quality could offset the negative effect of remittances on economic growth.

Although the above studies use more or less the same methods, their results are different. A reason might be the lack of proper data. As remittances have become more important the last decade, more attention has been paid to collecting data on remittances. Above studies all used data no later than 2009. Therefore, I want to contribute to the existing literature by performing a panel-data analysis with data for 6 additional years.

Besides that, the above studies only looked at the world as a whole or a specific region. I want to look whether the effect for the specific region Sub-Saharan Africa is different compared to the effect for other regions. Sub-Saharan Africa is still the region with the least amount of bank accounts and one of the regions that has the least developed financial sector. Therefore, I hypothesize that the effect of remittances on economic growth conditional on financial sector development will be bigger for Sub-Saharan Africa, when remittances act as a substitute for financial sector development.

(15)

3. Methodology

This thesis tries to answer the question whether the effect of remittances on economic growth is conditional on financial sector development and whether this effect is different for Sub-Saharan African countries compared to other developing countries. In order to answer this question, I will perform an empirical analysis partially based on the methodology of Giuliano and Arranz (2009). Section 3.1 specifies the model that will be used in this study. Section 3.2 gives more information on the control variables used in this thesis. Section 3.3 elaborates on the estimation technique, the problem of endogeneity and on possible solutions to correct for endogeneity.

3.1 Model specification

To empirically analyse the effect of remittances on economic growth conditional on financial sector development, I use an unbalanced panel dataset of 120 developing countries over the period 1990-2015. The dataset contains annual data and includes six additional years compared to previous research in this area. Chapter 4 gives more information on the data used in this thesis.

Just like Giuliano and Arranz (2009), the starting point of my analysis is an equation that estimates the effect of remittances on economic growth, where real GDP per capita growth is the dependent variable and the ratio of remittances to GDP the explanatory variable. The equation is specified as follows:

Δ𝑌𝑌𝑖𝑖𝑖𝑖 = 𝛽𝛽0+ (𝛽𝛽1− 1)𝑌𝑌𝑖𝑖,𝑖𝑖−1+ 𝛽𝛽2∙ 𝑋𝑋𝑖𝑖𝑖𝑖+ 𝛽𝛽3 𝑅𝑅𝑅𝑅𝑅𝑅𝑖𝑖𝑖𝑖 + 𝜇𝜇𝑖𝑖+ 𝜂𝜂𝑖𝑖 + 𝜀𝜀𝑖𝑖𝑖𝑖 (1) This equation is derived from 𝑌𝑌_{𝑖𝑖𝑖𝑖} = 𝛽𝛽0+ 𝛽𝛽1𝑌𝑌_{𝑖𝑖,𝑖𝑖−1}+ 𝛽𝛽2∙ 𝑋𝑋𝑖𝑖𝑖𝑖+ 𝛽𝛽3 𝑅𝑅𝑅𝑅𝑅𝑅𝑖𝑖𝑖𝑖+ 𝜇𝜇𝑖𝑖+ 𝜂𝜂𝑖𝑖 + 𝜀𝜀𝑖𝑖𝑖𝑖, by subtracting Yi,t-1 on both sides. In equation (1) i and t index country and time respectively; Δ𝑌𝑌𝑖𝑖𝑖𝑖 is the growth rate of real GDP per capita in annual percentages; 𝑌𝑌𝑖𝑖,𝑖𝑖−1 is the logarithm of real GDP per capita lagged one year; 𝑋𝑋_{𝑖𝑖𝑖𝑖} represents a matrix of control variables found in standard growth models; 𝑅𝑅𝑅𝑅𝑅𝑅_{𝑖𝑖𝑖𝑖} denotes the logarithm of the ratio of personal remittances received to GDP; 𝜇𝜇𝑖𝑖 denotes the unobserved time-specific effects; 𝜂𝜂𝑖𝑖 the unobserved country-specific effects and 𝜀𝜀_{𝑖𝑖𝑖𝑖} is the idiosyncratic error term. More information on the control variables used in this thesis will be given in section 3.2.

In the above equation (𝛽𝛽₁− 1) is the convergence coefficient, which is included to control for the general view found in neoclassical theory that countries with a lower initial level of GDP have a higher GDP growth rate compared to countries with a higher initial level of GDP

(16)

(Barro and Sala-i-Martin, 1992). Following previous studies, all explanatory variables are in logarithms (Chami et al., 2005; Giuliano and Arranz, 2009; Singh et al., 2010).

To estimate the effect of remittances on economic growth conditional on financial sector development, I augment equation (1) by adding financial development as a separate variable and by adding an interaction term between financial development and remittances.2 Adding financial development, I obtain the following equation (2):

Δ𝑌𝑌𝑖𝑖𝑖𝑖 = 𝛽𝛽0+ (𝛽𝛽1− 1)𝑌𝑌𝑖𝑖−1+ 𝛽𝛽2∙ 𝑋𝑋𝑖𝑖𝑖𝑖+ 𝛽𝛽3∙ 𝑅𝑅𝑅𝑅𝑅𝑅 + 𝛽𝛽4∙ 𝐹𝐹𝐹𝐹 + 𝛽𝛽5∙ (𝑅𝑅𝑅𝑅𝑅𝑅 ∗ 𝐹𝐹𝐹𝐹) +

𝜇𝜇𝑖𝑖+ 𝜂𝜂𝑖𝑖+ 𝜀𝜀𝑖𝑖𝑖𝑖 (2)

Financial development, FD, is either measured by the ratio of broad money to GDP or by the ratio of credit provided to the financial sector to GDP. These two financial depth measures are often used in the literature and are the same ones that were used by Nyamongo et al. (2008). In equation (1), the effect of remittances on economic growth was measured by 𝛽𝛽₃. Now the effect of remittances on economic growth is measured by 𝛽𝛽₃+ 𝛽𝛽₅∙ 𝐹𝐹𝐹𝐹. In order to have a positive effect of remittances on economic growth, the sum of 𝛽𝛽3+ 𝛽𝛽5∙ 𝐹𝐹𝐹𝐹 has to exceed 0. When 𝛽𝛽₅ has a positive sign, it indicates that remittances and financial development are complements. Assuming that the effect of remittances on economic growth, 𝛽𝛽₃, is positive, this would mean that the effect of an increase in remittances on economic growth is enforced by a more developed financial sector. A negative sign for 𝛽𝛽5 would indicate that remittances are a substitute for financial development. In that case, remittances are more effective in boosting economic growth when a country has a less developed financial sector (Giuliano and Arranz, 2009).

As this study also tries to answer the question whether the effect of remittances on economic growth is different for Sub-Saharan African countries compared to other developing countries, I will also estimate an altered version of equation (1) that includes an interaction term between remittances and a dummy for Sub-Saharan Africa. This results in equation (3):

Δ𝑌𝑌𝑖𝑖𝑖𝑖 = (𝛽𝛽1− 1)𝑌𝑌𝑖𝑖−1+ 𝛽𝛽2∙ 𝑋𝑋𝑖𝑖𝑖𝑖+ 𝛽𝛽3∙ 𝑅𝑅𝑅𝑅𝑅𝑅 + 𝛽𝛽4∙ (𝑅𝑅𝑅𝑅𝑅𝑅 ∗ 𝑆𝑆𝑆𝑆𝑆𝑆) + 𝜇𝜇𝑖𝑖+ 𝜂𝜂𝑖𝑖+ 𝜀𝜀𝑖𝑖𝑖𝑖 (3) Furthermore, to test whether the conditionality of financial development in remittances on economic growth is different for Sub-Saharan Africa, a three-way interaction term between remittances, financial development and a dummy for Sub-Saharan Africa is included in an adjusted version of equation (2). This results in equation (4):

2_{Adding financial development as a separate variable has also been done by other studies (Giuliano and Arranz,}

2009; Lartey, 2013). The reason for including financial development as a separate variable as well is to ensure that the interaction term does not proxy for the level of development of financial markets.

(17)

Δ𝑌𝑌𝑖𝑖𝑖𝑖 = (𝛽𝛽1− 1)𝑌𝑌𝑖𝑖−1+ 𝛽𝛽2 ∙ 𝑋𝑋𝑖𝑖,𝑖𝑖+ 𝛽𝛽3∙ 𝑅𝑅𝑅𝑅𝑅𝑅 + 𝛽𝛽4∙ (𝑅𝑅𝑅𝑅𝑅𝑅 ∗ 𝐹𝐹𝐹𝐹) + 𝛽𝛽5∙ (𝑅𝑅𝑅𝑅𝑅𝑅 ∗ 𝐹𝐹𝐹𝐹 ∗ 𝑆𝑆𝑆𝑆𝑆𝑆) + 𝜇𝜇𝑖𝑖+ 𝜂𝜂𝑖𝑖 + 𝜀𝜀𝑖𝑖,𝑖𝑖 (4) The interaction term between remittances and the dummy for Sub-Saharan Africa is not included in this equation to avoid imperfect multicollinearity.3

3.2 Control variables

There are many different control variables that have been used in economic growth regressions in previous studies. According to Sala-I-Martin (1997), 62 different variables can be identified that have been found significantly correlated with economic growth at least once. In his paper, he runs nearly two million regressions to identify which variables are systematically significantly correlated with economic growth. He finds that 25 of the 62 control variables appear to be significant.

Based on the variables that he identified to be significant and based on the control variables used in previous research by Giuliano and Arranz (2009) and Singh et al. (2010), I decided to use the following set of control variables in this thesis:

• Trade openness, defined as the sum of exports and imports of goods and services as a share of GDP. According to theory, trade openness has a positive effect on economic growth. So, I expect a positive sign here (Sala-I-Martin, 1997).

• Government expenditure, measured as the general government final consumption expenditure as a percentage of GDP, which is included to control for the effect of public policy on economic growth. There is no consensus on the sign of the coefficient of government expenditure. Whether this one is positive or negative depends on the kind of projects the government spends its money on and whether it crowds in or crowds out private investment.

• Inflation, measured by the annual % change in GDP deflator, which is included to capture the macroeconomic stability of the economy. Research shows that inflation distorts economic agents’ decision-making as it is a source of uncertainty, and leads to reduced investment and economic growth. A negative coefficient is expected for this variable.

3_{Imperfect multicollinearity arises when one of the regressors is very highly correlated with the other regressor,}

(18)

• Investment, measured by the ratio of gross fixed capital formation to GDP. Based on previous studies, I expect a positive effect of investment on economic growth (Sala-I-Martin, 1997).

• Population growth, as a measure for labour supply. There is no consensus in literature on whether this variable has a positive or negative effect on economic growth.

• Human capital formation measured by the level of people above age 25 that have finished secondary education.4 Based on previous studies, I expect a positive effect of human capital formation on economic growth (Sala-I-Martin, 1997).

3.3 Estimation technique

3.3.1 Ordinary Least Squares (OLS) estimation and the problem of endogeneity

In order to be able to compare my results with the studies discussed in the literature review, I start by performing a pooled Ordinary Least Squares (OLS) estimation. Pooled OLS ignores the heterogeneity in the data and does not correct for unobserved country-specific fixed effects. It is however, very likely that the pooled OLS estimation suffers from endogeneity. Endogeneity occurs when an explanatory variable is correlated with the error term and can be caused by several sources, including model misspecification or omitted variables, measurement error and simultaneity.5 When this is the case, pooled OLS estimation leads to biased and inconsistent results (Stock & Watson, 2012, p. 462).

In my model, there are several sources for endogeneity. The first one being the possible correlation between the included right-hand side variables and the unobserved country-specific fixed effects. It is very likely that the lagged real GDP per capita, Yi,t-1, is correlated with the standard errors, because the latter includes time-invariant country-specific fixed effects. As OLS estimation does not take into account these fixed-effects, this endogeneity caused by omitted variable bias leads to inconsistent results when pooled OLS estimation is used (Lartey, 2010).

The second source is the simultaneity between remittances and economic growth. It is likely that remittances do not only affect economic growth, but that economic growth in the recipient country also affects the remittances inflow. This simultaneity arises in two ways; through increased migration and through an increased remittances level. Countries with low economic growth and a less successful economic performance, will have more outward

4_{See table 2 in appendix B for an elaborate description of the variables used and their sources.}

5_{Simultaneity means that the causality effect is not purely from the right-hand side variables to the left-hand}

(19)

migration and therefore higher remittance inflows. Next to that, countries that experience less successful economic performance might receive more remittances because of altruistic behaviour (Giuliano & Arranz, 2009; Singh et al., 2010; Gupta et al., 2009).

3.3.2 How to solve the problem of endogeneity?

To address the first problem of endogeneity described in the previous section, I use fixed effects (FE) and random effects (RE) OLS. With fixed effects OLS the variables in the model are adjusted by subtracting the mean of each variable over time. By doing this the unobserved country-specific effects are removed. To decide whether the fixed effects model or the random effects model is the best fit for my regression, a Hausman-test is conducted. The null-hypothesis of the Hausman-test states that both the random effects model and the fixed effects model are consistent estimators. Rejecting the null-hypothesis would mean that fixed effects OLS is preferred. However, fixed effects OLS and random effects OLS do not correct for other sources of endogeneity bias like time-specific unobserved effects, time-specific measurement error and, more importantly, simultaneity. To correct for the time-specific unobserved effects, I include time-dummies in all regressions.

There are different ways to correct for the endogeneity caused by simultaneity. As was described in the literature review in chapter 2, other studies that investigated the effect of remittances on economic growth mostly used an Instrumental Variable (IV) estimation, a Two-Stage Least-Squares (TSLS) estimation with lagged explanatory variables or a Generalized Method of Moments (GMM) estimation to solve this problem.

To conduct an IV estimation, one needs a valid instrument. A valid instrument satisfies two conditions: 1) instrument relevance and 2) instrument exogeneity. Instrument relevance means that the instrumental variable is correlated with the potential endogenous explanatory variable, remittances in my case, and uncorrelated with the dependent variable, economic growth. So, the effect of the instrumental variable on economic growth can only go through the explanatory variable (Stock & Watson, p. 476).

Several instruments have been proposed in the literature to control for the simultaneity between remittances and economic growth. One of the most often used instruments for remittances, is the distance between the migrants’ home country and their main destination country (IMF, 2005; Faini, 2006; Rajan and Subramanian, 2005). However, this instrument suffers from the drawback that it does not vary over time and therefore, cannot be used in a panel framework.

(20)

As a response to this problem, Fajnzylber and López (2007) changed this time-invariant instrument into a time-varying instrument by multiplying the inverse of the distance between the migrants’ destination country and the remittance-receiving country by a measure of the respective country’s economic performance (Barajas et al., 2009). However, as is the case for the instruments used by Chami, Fullenkamp and Jahjah (2005), who use the ratio of the country’s income to U.S. income and the ratio of the country’s real interest rate to the U.S. real interest rate, this instrument is criticized for not being valid as it also affects economic growth and therefore violates the condition of instrument relevance.

Barajas et al. (2009) argue that instead of using macroeconomic instruments to correct for endogeneity, one should use microeconomic instruments as these affect the amount of remittances, but do not have a direct effect on economic growth. The transaction costs of transferring remittances would be an ideal candidate, but unfortunately data for this variable is only available from the year 2008 on and can therefore not be used in this study.

Since there is no consensus yet on a suitable instrument, some studies use lagged explanatory variables as instruments to correct for endogeneity (Catrinescu et al. (2006); Singh et al. (2010)). Another solution that is often used in the literature, is the use of a Generalized Method of Moments (GMM) estimation. As the GMM estimator is designed to address potential endogeneity issues, I will use this solution as well. More specifically, I will follow the estimation method used by Giuliano and Arranz (2009), who perform a system Generalized Method of Moments estimation following Arellano and Bover (1995).

3.3.3 System Generalized Method of Moments (SGMM)

The generalized method of moments (GMM) estimator is designed to address potential endogeneity issues and for estimations using a sample with few time periods and many individuals. I will use a two-step system GMM estimation, which combines an equation in first-differences with an equation in levels, and simultaneously estimates this system of two equations. The equation in first-differences uses lagged levels as instruments and the equation in levels uses the lagged difference of the endogenous variables as instruments.

The system GMM estimator is used instead of the difference GMM estimator, because the first one is said to be more efficient. The reason for this is, that the inclusion of the levels equation with the system GMM estimator allows the use of information on cross-country differences, which is not the case with the difference GMM estimator. A disadvantage of the two-step system GMM estimator, is that the standard errors obtained using this estimator have an asymptotically downward bias. However, the Stata command that I use to estimate the

(21)

two-step system GMM, xtabond2, automatically corrects for this by applying a finite-sample correction to the two-step covariance matrix as derived by Windmeijer (2005).

As Roodman (2009) indicates in his paper, classifying the endogenous variables (and therefore the instruments) should be based on economic theory and cannot be chosen by looking at which endogenous variables give the best results. Furthermore, as a rule of thumb one should not use more instruments than the number of countries used in the estimation to avoid an ‘overfit of endogenous variables’. Based on economic theory I classify lagged real GDP per capita, remittances, investment, inflation, government consumption, openness and the financial development indicators as potential endogenous variables, education and population growth are classified as predetermined variables. When conducting GMM estimation, it is customary to instrument endogenous variables with lagged terms of t‐2 periods and back, while for pre‐determined variables, lagged terms of t‐1 periods and back are used.

For the system GMM estimations using annual data, the endogenous variables are instrumented using lags two through six of the levels as instruments for the first-difference equation and the first-differenced lag one value for the equation in levels. This lag depth was chosen by I use the collapse option in Stata to reduce the number of instruments to a minimum. For the system GMM estimations using the 4-year averages, the same applies, but then with lags 2 to 3 instead of 2 to 6. The collapse option is not used for the 4-year averages as it reduced statistical efficiency.

To test the validity of the instruments used, I employ two specification tests; the Arellano-Bond test to check for serial autocorrelation in the first-differences equation and the Hansen test for over-identifying restrictions. The null-hypothesis for the Arellano-Bond test states that the error terms are not serially correlated. So, rejecting the null-hypothesis would indicate that the instruments used are not valid. Autocorrelation of order 1, AR(1), is expected in the first-differences equation, as the difference of the standard errors and the lagged difference of the standard errors both contain the lagged standard error.6 The joint null-hypothesis of the Hansen test states that the instruments are valid instruments, i.e., uncorrelated with the error term, and that the excluded instruments are correctly excluded from the estimated equation. The p-value should be as big as possible in order not to reject the null-hypothesis. However, an implausibly good p-value of 1.00 is not good either as it indicates that there is something wrong with the model (Roodman, 2009).

6

(22)

3.4 Robustness tests

Besides performing the regression analysis using annual data, I will also perform the regression analysis using the 4-year averages of the annual data to control for business cycle fluctuations. To derive this dataset, I split the sample period 1990-2015 into seven non-overlapping 4-year periods.7

As a second robustness test, I will exclude investment as a control variable for some of the regressions. Previous studies pointed out that including investment as a control variable influences the size and significance of the remittances variable (e.g. Faini, 2006). The reason for this is, when investment is included the effect of remittances on economic growth may partially be captured by the investment variable. The only remaining effect of remittances on economic growth not captured by investment, is the effect through the total factor productivity channel rather than through the capital accumulation channel (Chami et al., 2008). Therefore, as a robustness test, I will exclude investment as a control variable for some of the regressions.

7_{The periods run from 1990-1993, 1994-1997, 1998-2001, 2002-2005, 2006-2009, 2010-2013, 2013-2015. The}

(23)

4. Data

This chapter provides more information on the data used to perform the empirical analysis and on their sources. Section 4.1 elaborates on the sample selection that was used for the empirical analysis. Section 4.2 discusses some limitations to the data on remittances.

4.1 Sample selection

To answer the question on the effect of remittances on economic growth, I use a panel dataset of 120 developing countries, of which 40 Sub-Saharan African-countries, for the period 1990-2015.8 The country selection is based on the list of developing countries published by the World Bank. Furthermore, the decision on the study period and on which countries to include from this list is entirely based on the data availability of remittances and real GDP per capita growth. Unfortunately, there are only 55 countries, of which 20 SSA-countries, for which data is available for all variables and all points in time. The other countries are missing one or two variables for some years. As a result of this, the dataset is unbalanced. However, with on average 1930 observations per regression this dataset is still the most comprehensive one so far.9 The list of countries used in this study can be found in appendix A.

The dependent variable in this study is the real GDP per capita growth in annual percentages. The explanatory variables consist of the amount of personal remittances received as a percentage of GDP, the two financial development indicators measured as the ratio of broad money to GDP and the ratio of credit provided to the financial sector to GDP, and the control variables described in section 3.2. Except for the education variable, all data is retrieved from the World Bank’s Development Indicators database (World Bank, 2017a). The available 5-year data for the education variable is retrieved from the Barro-Lee database. Annual data for this variable is obtained by using linear interpolation. An elaborate description of all the variables used in this thesis and their sources, can be found in table 2 in appendix B.

Table 3 in appendix C shows the descriptive statistics and table 4 in appendix C presents the correlation matrix. The sign of the correlation coefficients between the control variables and economic growth are consistent with theory. Openness, education and investment are positively correlated to economic growth and population growth, government expenditure and inflation are significantly negatively correlated with economic growth. Based

8

A list of all the countries used in this study can be found in appendix A.

9

(24)

on the correlation matrix, remittances seem to have a significant positive correlation with economic growth (0.0428), as well as with the two financial development indicators. The two financial development indicators have a high significant correlation coefficient (0.7102), which is good because these will be the indicators used to answer the question whether the effect of remittances on economic growth is conditional on financial sector development.

4.2 Data limitations

It should be noted that there are some limitations to the quality of the data for remittances. First of all, the recorded data for remittances only contain remittance flows send through formal channels and might therefore understate the actual remittance flows. Data for remittances send through informal channels like mail carriers or Hawala is limited or not available at all.10 According to Ratha (2003), the share of unrecorded remittances ranges from 10 to 30%. According to Freund and Spatafora (2008) this share range is even higher, namely from 35 to 75%. Although the numbers that they have estimated are quite different, they do agree on the fact that the recorded remittances understate the actual remittance flows.

Secondly, two decades ago there were no guidelines yet on how to classify remittances. As a result of this, some countries classified remittances as workers’ remittances, while others registered the amount of remittances under migrants’ transfers or as a completely different entry on the balance of payments. Only the last couple of years the World Bank set up guidelines on how to classify remittances. Therefore, an increase in remittances can represent a real increase in the amount of remittances, but could also be a result of improved registration of remittances or a decrease in the unrecorded portion of remittances sent through informal channels. As Catrinescu et al. (2009) indicate, better technology, decreased transfer transaction costs and efforts to reduce money laundering have generated a decrease in the unrecorded portion of remittance.

10_{Hawala is an informal system of transferring money whereby the money is paid to an agent who then instructs}

an associate in the relevant country or area to pay the final recipient. The system is based on trust and the money is not actually moved to the recipient country.

(25)

5. Results

This chapter presents the results of the empirical analysis to answer the question whether the effect of remittances on economic growth is conditional on financial sector development and whether the effect of remittances on economic growth is different for the Sub-Saharan African region. Section 5.1 describes the results for the different estimations using annual data. Section 5.2 describes the results for the empirical analysis using the 4-year averages.

5.1 Results of the annual data-analysis

This section presents the results of the annual-data analysis. Section 5.1.1 discusses the results for the total group of developing countries and section 5.1.2 those for the Sub-Saharan African region specifically.

5.1.1 Results for the total group of developing countries

Table 1 presents the results of equation (1), which is a regression of the dependent variable, real GDP per capita growth, on remittances and the control variables and does not include the financial development indicators yet. The standard errors reported in this table and in all subsequent tables, are robust to heteroscedasticity after a Breusch-Pagan/Cook-Weisberg test showed evidence for heteroscedasticity in the standard errors.11

Column (1) and (2) present the results for the pooled OLS estimation, column (3) and (4) for the fixed-effects OLS estimation and column (5) and (6) for the system Generalized Method of Moments estimation, both including and excluding the investment variable. 12 As explained in section 3.4, the reason for excluding this variable is, that the effect of remittances on economic growth is mainly through the investment channel. Therefore, leaving this variable out would change the coefficient of remittances upwards. The results of this can be found in column (2), (4) and (6).

Looking at the results of the pooled OLS estimation and the fixed-effects OLS estimation, all control variables have the expected sign. Moreover, all are significant except for the variables openness and population growth. Openness, education and investment have a positive effect on economic growth and lagged real GDP per capita, population growth, government expenditure and inflation have a negative effect on economic growth. The negative coefficient

11_{The results of the Breusch-Pagan/Cook-Weiberg test can be found in appendix D.}

12_{The results for the random-effects OLS estimations are not shown here, as the Hausman-test showed evidence}

(26)

of lagged real GDP per capita supports the convergence theory that countries with a lower initial level of real GDP per capita experience a higher growth rate. The system GMM estimation finds the same signs for the coefficients of the control variables, but now only the coefficient for lagged real GDP per capita and for investment are significant.

Table 1: Pooled OLS, Fixed Effects OLS and System GMM estimates of the effect of remittances on economic growth

Pooled OLS FE OLS SGMM

(1) (2) (3) (4) (5) (6)

Lagged real GDP per capita -0.616*** -0.401* -5.436*** -4.895*** -1.675** -1.244 (0.218) (0.214) (1.165) (1.058) (0.852) (1.105) Remittances -0.012 0.006 0.167 0.186 0.003 0.135 (0.081) (0.083) (0.136) (0.135) (0.276) (0.377) Population growth -0.500* -0.448 -0.507 -0.333 -1.001* -0.637 (0.261) (0.276) (0.400) (0.399) (0.587) (0.629) Openness 0.281 1.027*** 1.080 2.228*** 0.922 4.201 (0.354) (0.390) (0.663) (0.681) (2.232) (2.756) Education 0.404*** 0.412*** 0.335** 0.349** 0.429*** 0.401** (0.138) (0.142) (0.141) (0.144) (0.164) (0.180) Government expenditure -1.477*** -1.578*** -1.917*** -2.185*** -2.223 -3.564 (0.480) (0.500) (0.725) (0.709) (1.659) (2.435) Investment 2.618*** 2.509*** 3.593*** (0.639) (0.756) (1.257) Inflation -0.411*** -0.422*** -0.522*** -0.529*** -0.518 -0.397 (0.138) (0.140) (0.154) (0.154) (0.687) (0.842) _cons 0.329 3.655 40.258*** 39.046*** 8.307 4.165 (3.100) (3.097) (10.903) (10.240) (14.517) (17.215) N 1981 1998 1981 1998 1976 1993 R2 F Instruments AR(1) AR(2) Hansen-test 0.156 420.73 0.129 294.48 0.187 14.10 0.158 10.39 287.89 69 0.001 0.683 0.621 231.36 63 0.000 0.576 0.437 Note: Robust standard errors in parentheses. All variables are in natural logarithms, except for the dependent variable (real GDP per capita growth). All regressions include time dummies. * p < 0.10, ** p < 0.05, *** p < 0.01.

Our main variable of interest, remittances, in all cases has a positive coefficient except for the pooled OLS estimation. For all estimation methods, this coefficient is small and insignificant. For the fixed-effects OLS estimation I find a remittances coefficient of 0.167 and for the system GMM estimation a coefficient of 0.003. These results are largely consistent with the results of Giuliano and Arranz (2009). They also find a positive and insignificant coefficient for remittances, 0.043 and 0.010 using OLS- and SGMM estimation respectively.

(27)

Leaving investment out of the equation does not change the results. As expected, the coefficient for remittances becomes a bit bigger, but is still insignificant and small. A reason for the somewhat disappointing insignificant results might be the omitted variable financial development.

Table 2: Pooled OLS, Fixed Effects OLS and System GMM estimates of the effect of remittances on economic growth conditional on financial development measured as M2/GDP.

(1) (2) (3) (4) (5) (6)

Lagged real GDP per capita -0.532* -0.550** -5.238*** -5.173*** -3.132** -4.495*** (0.276) (0.262) (1.215) (1.224) (1.368) (1.729) Remittances -0.058 0.985** 0.120 0.691 0.148 3.673** (0.095) (0.420) (0.147) (0.686) (0.388) (1.861) M2/GDP -0.122 -0.055 -0.849 -0.866 3.864* 3.546 (0.421) (0.390) (0.583) (0.580) (2.331) (2.223) Population growth -0.691*** -0.686*** -0.753** -0.767** -1.188* -1.779** (0.239) (0.228) (0.334) (0.329) (0.683) (0.748) Openness 0.288 0.272 1.246* 1.163* 0.651 -0.642 (0.371) (0.351) (0.683) (0.668) (2.326) (2.183) Education 0.441*** 0.438*** 0.365*** 0.367*** 0.393** 0.393** (0.129) (0.129) (0.133) (0.133) (0.165) (0.169) Government expenditure -1.357*** -1.486*** -1.729** -1.808** -2.872 -3.550* (0.475) (0.473) (0.713) (0.712) (1.851) (2.002) Investment 2.264*** 2.239*** 1.989*** 1.976*** 3.060*** 3.049*** (0.487) (0.471) (0.504) (0.500) (1.057) (1.071) Inflation -0.351*** -0.332** -0.480*** -0.465*** -0.690 -0.695 (0.135) (0.131) (0.157) (0.154) (0.543) (0.518) REM x M2/GDP -0.295** -0.161 -1.137** (0.115) (0.168) (0.561) _cons 1.057 1.545 42.657*** 42.810*** 11.499 32.553* (2.889) (2.727) (10.616) (10.408) (14.794) (17.405) N 1930 1930 1930 1930 1926 1926 R2 F Instruments AR(1) AR(2) Hansen-test 0.146 371.60 0.150 377.67 0.180 14.96 0.181 14.77 218.42 75 0.001 0.659 0.412 254.64 81 0.000 0.631 0.368 Note: Robust standard errors in parentheses. All variables are in natural logarithms, except for the dependent variable (real GDP per capita growth). All regressions include time dummies. * p < 0.10, ** p < 0.05, *** p < 0.01.

(28)

Table 3: Pooled OLS, Fixed Effects OLS and System GMM estimates of the effect of remittances on economic growth conditional on financial development measured as domestic credit provided by the financial sector to GDP.

(1) (2) (3) (4) (5) (6)

Lagged real GDP per capita -0.358 -0.379 -5.057*** -5.041*** -1.461 -1.778* (0.239) (0.234) (1.176) (1.187) (1.067) (1.059) Remittances -0.079 0.419 0.056 0.196 0.179 0.931 (0.088) (0.274) (0.151) (0.435) (0.338) (0.645) Credit/GDP -0.228 -0.211 -0.688*** -0.687*** 0.670 0.530 (0.231) (0.224) (0.259) (0.260) (0.685) (0.682) Population growth -0.678*** -0.693*** -0.737** -0.739** -0.625 -0.832* (0.228) (0.222) (0.332) (0.330) (0.540) (0.440) Openness 0.230 0.254 1.128* 1.099* 2.133 1.820 (0.345) (0.336) (0.673) (0.659) (2.175) (1.893) Education 0.420*** 0.407*** 0.320** 0.318** 0.315** 0.327*** (0.128) (0.129) (0.133) (0.133) (0.131) (0.125) Government expenditure -1.747*** -1.851*** -1.701** -1.724** -2.246 -1.935 (0.433) (0.425) (0.737) (0.733) (1.568) (1.327) Investment 2.281*** 2.255*** 1.967*** 1.957*** 3.476*** 3.215*** (0.497) (0.483) (0.524) (0.523) (1.150) (1.103) Inflation -0.377*** -0.367*** -0.533*** -0.527*** -0.581 -0.738 (0.123) (0.122) (0.144) (0.142) (0.616) (0.555) REM x Credit/GDP -0.148** -0.042 -0.218 (0.075) (0.104) (0.208) _cons 1.260 1.721 41.103*** 41.189*** -0.683 4.205 (2.817) (2.703) (10.400) (10.257) (11.534) (10.254) N 1911 1911 1911 1911 1907 1907 R2 F Instruments AR(1) AR(2) Hansen-test 0.150 361.77 0.150 370.80 0.186 16.84 0.186 16.11 311.44 75 0.000 0.750 0.938 301.70 81 0.000 0.752 0.923 Note: Dependent variable is real GDP per capita growth. Robust standard errors in parentheses. All variables are in natural logarithms, except for the dependent variable. All regressions include time dummies. * p < 0.10, ** p < 0.05, *** p < 0.01.

(29)

Table 2 and 3 present the results for equation (2), the effect of remittances on economic growth conditional on financial sector development. In both tables, column (1) and (2) present the results for the pooled OLS estimation, column (3) and (4) for the fixed-effects OLS estimation and column (5) and (6) for the system GMM estimation, both including and excluding the interaction term between remittances and financial development.

Table 2 presents the results using broad money to GDP as the financial development indicator and table 3 presents the results using domestic credit provided to the financial sector to GDP as the measure for financial development. The results for equation (2) are more or less similar with the ones from table 1. The control variables again mostly have the expected sign and are significant.

Striking is that the coefficient for remittances in table 2 is positive and significant. Although the coefficient of remittances is only significant in 2 out of 6 regressions, there seems to be a slight positive effect of remittances on economic growth when financial development is included in the regression. The pooled OLS and SGMM estimates indicate that an increase of 1% in the ratio of remittances to GDP leads to an increase in economic growth of 0.985 and 3.673 percentage point respectively.

The coefficient for remittances in table 3, is again positive for most regressions but insignificant. Here I find a remittances coefficient of -0.079, 0.056, 0.179 for pooled OLS, fixed effects OLS and system GMM respectively. It should be noted that the coefficient of remittances becomes bigger when the interaction term between remittances and financial sector development is included in the regression.

Our main variable of interest in this equation is the interaction term between remittances and financial development. When financial development is measured as broad money to GDP, two out of three regression show a significant negative effect of the interaction term between financial development and remittances on economic growth. This could indicate that remittances act as a substitute for credit obtained through the financial sector. However, as Barajas et al. (2009) indicate there could also be an alternative interpretation of this result. It could mean that the increase in financial depth achieved through remittances is of a lesser quality. Although the financial sector is larger, it is not intermediating resources more efficiently (Barajas et al., 2009).

The negative coefficient is also found when we measure financial sector development as domestic credit provided by the financial sector to GDP. Although the results are not significant for all estimations, it gives proof to conclude that if there is an interaction effect between remittances and financial development, it is a negative one. The result of this

(30)

negative interaction effect between remittances and financial development, is consistent with the findings of Giuliano and Arranz (2009).

Financial development itself, does not seem to have a clear effect on economic growth. When financial development is measured as the ratio of broad money to GDP, an insignificant negative effect on economic growth is found based on the pooled OLS and fixed-effects OLS estimations. The system GMM estimates without the interaction term between remittances and financial development, do show a positive and significant effect of financial development on economic growth. However, as four out of six regressions show a negative coefficient for this variable and the positive effect is only significant at the 0.10 significance-level, it is not a very convincing effect.

This effect becomes even less convincing when we look at the results of financial sector development measured as the ratio of domestic credit provided by the financial sector to GDP in table 3. Here financial development has a significant negative effect on economic growth when fixed effect OLS estimation is used.

As a robustness test, the system GMM estimation for equation (2), that includes an interaction term between remittances and financial development, is also performed without including investment as a control variable. Table 4 shows the results for the system GMM estimates of equation (2) excluding investment. As expected, the coefficient for remittances becomes larger and has increased significantly for the estimation including broad money to GDP and the interaction term between broad money to GDP and remittances. The coefficient for broad money to GDP becomes much larger and significant as well. The coefficient of credit to GDP does not change significantly compared with the results in table 3 and is positive. Again, a negative coefficient is found for the interaction term between financial development and remittances. Although the one where financial development is measured as credit to GDP, is not significant.

(31)

Table 4: System GMM estimates of the effect of remittances on economic growth conditional on financial development excluding investment.

SGMM

(1) (2) (3) (4)

Lagged real GDP per capita -3.730*** -1.035 -4.810*** -1.447

(1.379) (1.410) (1.312) (1.188) Remittances 0.165 0.270 6.596*** 1.309 (0.349) (0.405) (2.107) (0.877) M2/GDP 5.842*** 4.381** (2.240) (1.760) Openness 0.813 5.020* -0.843 4.296* (2.526) (2.692) (2.273) (2.423) Inflation -0.407 -0.537 -0.390 -0.688 (0.577) (0.825) (0.564) (0.709) Education 0.424** 0.286* 0.383** 0.311** (0.175) (0.164) (0.154) (0.139) Government expenditure -4.119* -4.337 -6.180*** -4.026* (2.429) (2.916) (2.044) (2.308) Population -1.138* -0.447 -2.229*** -0.655 (0.670) (0.671) (0.692) (0.568) Credit/GDP 0.336 0.418 (0.935) (0.867) REM x M2/GDP -2.001*** (0.608) REM x Credit/GDP -0.304 (0.270) _cons 19.858 -1.136 49.232*** 4.720 (15.522) (18.085) (17.354) (15.485) N 1943 1924 1943 1924 R2 F Instruments AR(1) AR(2) Hansen-test 241.35 69 0.002 0.606 0.470 236.89 69 0.000 0.669 0.529 239.32 75 0.001 0.771 0.921 225.45 75 0.000 0.758 0.708 Note:Dependent variable is real GDP per capita growth. Robust standard errors in parentheses. All variables are in natural logarithms, except for the dependent variable. All regressions include time dummies. * p < 0.10, ** p < 0.05, *** p < 0.01.

(32)

5.1.2 Results for Sub-Saharan African countries specifically

In table 5 an interaction term between remittances and a region dummy for Sub-Saharan Africa is included. Except for openness, all control variables have the expected sign and are significant. Although the size and significance change a bit by estimation method.

When the interaction term between remittances and the dummy for Sub-Saharan Africa is added, the coefficient for remittances becomes larger and changes from -0.012, 0.167 and 0.003 to -0.105, 0.310 and 0.181 for the pooled OLS-, fixed-effect OLS- and system GMM estimation respectively. However, the coefficient for remittances is still insignificant. There is no significant effect either for the interaction term between remittances and the region dummy for Sub-Saharan Africa on economic growth. These results indicate that there is no significant effect of remittances on economic growth and that the effect of remittances on economic growth is not different for Sub-Saharan Africa compared with the developing countries in the other regions.

Table 6 shows the results for equation (4), the effect of remittances on economic growth conditional on financial sector development and including a three-way interaction term between remittances, financial development and the region dummy for Sub-Saharan Africa. Including financial development to the equation by either measuring it as broad money to GDP (column 1, 3, 5) or credit to GDP (column 2, 4, 6), changes the significance and size of the remittances coefficient significantly. Whereas remittances had no effect in the previous estimation, they have a significant positive effect when financial development is included. The pooled OLS and SGMM estimates indicate that an increase of 1% in the ratio of remittances to GDP leads to an increase in economic growth of 0.927 and 3.094 percentage point respectively, when financial development is measured as broad money to GDP.

Just like in equation (2), a significant negative effect is found for the interaction term between financial development and remittances. Although this effect is not significant for all estimations, it does indicate that if there is any interaction effect between financial sector development and remittances it is a negative one. This result is robust to the measure of financial sector development.

Neither of the estimation methods show a significant three-way interaction term. In combination with the previous estimation results, we can conclude that based on annual data there is no separate effect of remittances on economic growth in Sub-Saharan Africa.

(33)

Table 5: Pooled OLS, Fixed Effects OLS and System GMM estimates of the effect of remittances on economic growth including an interaction term between remittances and a region dummy for Sub-Saharan Africa.

Pooled OLS (1)

Fixed Effects OLS (2)

SGMM (3)

Lagged real GDP per capita -0.608*** -5.654*** -1.327*

(0.213) (1.121) (0.757) Remittances -0.105 0.310 0.181 (0.123) (0.211) (0.479) REM x SSA 0.223 (0.165) -0.293 (0.241) -0.024 (0.632) Population growth -0.536** -0.458 -0.690* (0.272) (0.396) (0.405) Openness 0.288 1.116 1.704 (0.349) (0.679) (1.999) Education 0.405*** 0.333** 0.401** (0.138) (0.142) (0.171) Government expenditure -1.494*** -1.924** -1.442 (0.477) (0.734) (1.545) Investment 2.618*** 2.514*** 3.636*** (0.633) (0.757) (1.255) Inflation -0.412*** -0.524*** -0.537 (0.139) (0.153) (0.704) _cons 0.312 41.927*** -0.324 (3.037) (10.573) (10.966) N 1981 1981 1976 R2 F Instruments AR(1) AR(2) Hansen-test 0.156 434.60 0.188 13.45 290.92 75 0.001 0.744 0.676 Note: Dependent variable is real GDP per capita growth. Robust standard errors in parentheses. All variables are in natural logarithms, except for the dependent variable. All regressions include time dummies. * p < 0.10, ** p < 0.05, *** p < 0.01.

Remittances, financial development and economic growth : is the effect of remittances on economic growth conditional on financial development and is this effect different for countries in the Sub-Saharan African region?

Abstract

Table of contents

1. Introduction

2. Literature review

2.1 Remittances and economic growth

2.2 Financial sector development, remittances and economic growth

3. Methodology

3.1 Model specification

3.2 Control variables

3.3 Estimation technique

3.4 Robustness tests

4. Data

4.1 Sample selection

4.2 Data limitations

5. Results

5.1 Results of the annual data-analysis